You are on page 1of 153

Distributed Thinking Symposium VI

Cognition Enacted

31 March 2014 Kingston University London

DTS6

2

Monday 31 March Rooms JG2012 and JG2003 (John Galsworthy Building)

09:00 Coffee JG2012 09:15 Opening remarks JG2003 09:30 Adam Toon 10:00 Gaëlle Vallée-Tourangeau 10:30 Stephen Cowley 11:00 Coffee 11:15 Nick Shipp 11:45 Chris Baber 12:30 Lunch and poster presentations Johanne Stege Bjørndahl Charlotte Harris Matthew Harvey Angeliki Makri 13:30 Lisa Guthrie 14:00 Jens Koed Madsen 14:30 Lucas Bietti 15:00 Frédéric Vallée-Tourangeau 15:30 Coffee 15:45 Orestis Palermos 16:15 Dan Hutto 17:00 Discussion (Pub) 19:00 Dinner

DTS6 Talks No Anticipation without Representation? Chris Baber University of Birmingham This talk will consider the challenge that ‘anticipation’ raises for Distributed Cognition. I will use examples drawn from my work on how people use hand-tools. If one takes a Radical Embodied Cognition stance (à la Chemero) and assumes that human activity can be explained through a combination of affordances and dynamical systems (as I do) then this raises a question of how one deals with anticipatory control. A traditional (representationalist) view might argue that people form a model of the world, then use this model to plan a course of action and then perform this action. The modelling and planning, therefore, form an essential part of anticipation. In the motor control literature, notions of feed-forward control are popularly invoked to explain how neural activation arises prior to movement. My talk will review some neuroimagining studies (relating to tool use) and anticipatory control studies. I want to ask whether there is too much reliance on the assumption that anticipation rests on a ‘model’ that is used to define a ‘program’ to be performed, and too little on the moment-bymoment correction of action-in-the-world.

3

co-speech gesture, postural sway, and eye gaze has different interactional dynamics while interactionally fostering collaborative remembering among small groups. Moreover, these analyses show whether and how the differences in these behaviors’ roles in joint remembering are reflected in the temporal dynamics of the alignment patterns observed in our data. Afterwards, in order to examine whether the instances of simultaneous and sequential alignment during joint remembering found in our data set can be associated to either priming effects (automatic) or conscious aspects of the interactions (task-mediated), I report preliminary results of an agent-based computer simulation. Finally, I discuss the potential of combined qualitative-quantitative approaches for illuminating the interplay of verbal and bodily coordination during contexts such as interactive memory construction.

Replacing Representations Stephen J. Cowley University of Southern Denmark The new consensus in cognitive science is that a system’s embodiment grounds cognition. For some, an agent’s embodiment constitutes cognition, for others this grounds conceptualizing and, for radicals, embodiment can replace representation. Taking a systemic perspective, the paper pursues the replacement view. From a systemic perspective, living human beings animate distributed cognitive systems. This results in a species specific, future-directed sense-making that depends on non-local or second-order constructs. For example, humans alone use music, money, numbers and language. Before arguing that representation can be replaced by a clear view of the second-order, it is important to show what is at stake. First, what we know about cognition depends on modeling systems (e.g. cockpits) that track environmental features. This tracking can be applied to pilots, ants, thermostats and rats. When describing tracking, appeal is often made to representations (Rs): these function as such for human observers. Though the approach has many uses,

Aligning Behaviors in Everyday Conversations about Shared Memories Lucas Bietti Telecom ParisTech In this talk, I investigate the roles that the interactive alignment of manual gesture, postural sway and eye-gaze play in small groups engaged in collaborative remembering. The video data used for these analyses come from an ongoing project on collaborative remembering in small groups. All participants were native Spanish speakers and data was collected in real-world environments (participants’ homes in Buenos Aires). Qualitative and quantitative analyses of a video corpus demonstrate how the alignment of

DTS6 one need not posit that Rs function as representations for the pilot, ant or thermostat. Indeed, they are likely to arise from the coupling of organism-environment systems that link up world, embodiment and brains. That is not my concern. I focus on representations that, by definition, count as such for the living (representing) organism. Tracking mechanisms (Rs) thus contrast with entities that perform a modeling function within an animal-environment system. By definition, these representations or Rmodels permit a person to separate out patterns from ongoing action/perception. An R-model can thus function as if ‘off-line’: it can be associated with, for example, coins, images, music or thoughts. As a result R-models can underpin observer-dependent ways of thinking, managing time and acting. For Vallée-Tourangeau and Vallée-Tourangeau (in press), higher cognition depends on getting –and using- ‘the gist’ of things. R-models clearly describe gist generation and manipulation. Do they serve any explanatory role? The Vallée-Tourangeaus are surely correct that they cannot be traced to on-line coupling: in Shapiro’s terms, there is nothing radical about the view that living bodies that conceptualize and/or constitute cognition. In pursuing a replacement view, I therefore turn to language. I argue that, by using interactivity within distributed systems, living human infants use routines and ways of feeling to self-constitute as speakingobservers. By acting strategically, a baby develops ways of languaging that show extraordinary sensitivity to the verbal patterns of languages: a child comes to create and construe speech-in-action while also hearing utterancetypes. As children come to believe in language (qua ensemble of verbal patterns), they link activity and hearing to a history of interactivity that becomes embrangled with imaginings of speech. While enabled by brains, skills in using patterns and routines thus transform cognitive powers based on the discovery of mental timetravel (e.g. by pretending or telling people what happened). As Dartnall (2005) suggests, this ensures that the world leaks into the head. Far from depending on Rs that sustain R-models, the latter are the child’s own cognitive constructs.

4

Radical Enactivist: Rethinking Basic Minds Dan Hutto University of Hertfordshire and University of Wollongong The cognitive revolution deposed behaviourist thinking (in both philosophy and psychology) and licensed a return to active theorizing about mental states and their place in nature. Promoting representational and computational theories of mind, many researchers in diverse fields have assumed that the contentful properties of such mental states play critical causal roles in computational processes enabling intelligent activity. But serious problems have been identified with the very idea that contentful mental representations (of the kind that might do such work) exist. Moreover, new, nonrepresentationalist approaches in the philosophy of mind and cognitive science – enactive, embodied approaches – have emerged and are growing in popularity. These developments suggest that the time is ripe for a complete rethink of the cognitive revolution. Against this backdrop, I give reasons to preferring radically embodied/enactive accounts of cognition than abandon traditional assumptions of classical cognitive science, proposing a fundamentally shift in how we might conceive of the basic nature of minds.

Drawing in Negatives: Exploring the Theoretical Placement of the SPIMP Cognitive Model of Persuasion Jens Koed Madsen Birkbeck College A recently developed cognitive psychological model of persuasion argues for a subjectiveprobabilistic, immersed, and cultural description of how humans deal with uncertain information from uncertain sources in potentially misleading situations (the so-called SPIMP). The model represents a departure from previous models of persuasion that focus on a dual-process perspective, which divides reasoning into rulebased and heuristic (e.g. the ELM and HSM). Further, the model is in line with a growing body

DTS6 of research applying a Bayesian probabilistic approach to reasoning. Centrally, this moves the description of reasoning towards a subjectively interactive and contextually immersed perspective while retaining empirical testability and support. The model performs well in predicting and describing how humans seem to computationally deal with uncertain information from an uncertain source from a subjective perspective. Conceptually, however, the underlying framework remains under-developed, as the notion of subjectivity and immersion lack definitions and descriptions. Thus, the talk points to a conceptual framework where the model is grounded in the persuadee’s interaction with the environment in order for subjectivity to emerge. Further, in order to move towards a theory of influence, the model needs to be placed in a larger theoretical framework concerning the link between belief and behaviour. Here, the model offers a potential piece of the puzzle in that it describes the process of evaluating uncertain information from an uncertain source, but fails in accounting for influence. The talk explores two limitations of the current model. Firstly, in terms of its theoretical foundation and secondly in terms of its placement in a larger conceptual framework that points towards interaction, distributed cognition, and the limits of persuasion. The talk is an invitation for a holistic perspective on the psychology of persuasion and influence in which the SPIMP offers a central, but inadequate element.

5

resources and better distribution of cognitive load. In attempting to simulate these moves made in the world, different levels of interactivity were examined with a series of mental arithmetic problems. The use of artefacts, such as tokens or a pen promoted more accurate and more efficient mental arithmetic performance. Participants were also profiled in terms of attitude to varying problem presentations as an assessment of their engagement in the task. They felt more positive about and better engaged with the task when they could reconfigure the problem presentation through interactivity. These findings underscore the importance of engineering task environments that support distributed problem representation and adequate levels of interactivity that creates a dynamically shifting topography of action affordances.

Knowledge and Cognitive Integration Orestis Palermos University of Edinburgh Cognitive integration is a defining yet overlooked feature of our intellect that may nevertheless have substantial effects on the process of knowledge-acquisition. To bring those effects to the fore, I explore the topic of cognitive integration both from the perspective of virtue reliabilism within externalist epistemology and the perspective of extended cognition within externalist philosophy of mind and cognitive science. On the basis of this interdisciplinary focus, I argue that cognitive integration can provide a minimalist yet adequate epistemic norm of subjective justification: so long as the agent’s belief-forming process has been integrated into his cognitive character, the agent can be justified in holding the resulting beliefs merely by lacking any doubts there was something wrong in the way he arrived at them. Moreover, since both externalist philosophy of mind and externalist epistemology treat the process of cognitive integration in the same way, we can claim that epistemic cognitive characters may extend beyond our organismic cognitive capacities to the artifacts we employ or even to other agents we interact with. I conclude by

Dynamic Problem Solving: The Effect and Affect of Interactivity on Mental Arithmetic Lisa G. Guthrie Kingston University Individuals often gesture, point or use objects as an aid to solving quotidian arithmetic problems. In attending to a problem, the dynamic loop of information and action flows between the person and the outside world, the very nature of these interactions constrain and guide strategic choices. This interactivity has been linked to better performance in problem solving, possibly due to a more efficient allocation of attentional

DTS6 hinting towards some of the possible theoretical and practical ramifications of this move.

6

Influencing Categorical Choices through Physical Object Interaction Nick Shipp University of Hertfordshire Recent theories on semantic memory have proposed that concepts are grounded in sensorimotor activity and action information plays a strong role in tasks even when such knowledge is not necessary. Research shows that participants are more likely to use action information as a basis for categorisation than perceptual information. The following experiment examined whether participants would be more likely to match items based on shared actions following interacting with the objects on a physical basis. Participants engaged in one of three priming tasks either using a series of objects for their functional capacity (action priming), grouping them into categories (taxonomic priming) or moving them from one table to another (movement priming). Following this participants were shown a forced-choice triad task in which items could be matched based on either taxonomic relations (rifle + sword) or a shared action relation (rifle + water pistol). Items within the triads were presented as an image either on a white background (context-lean condition) or as a functional scene with the object being used by an agent (context-rich condition). The results showed that participants were more likely to select the action related item when they were primed with the functional action of the objects and when the images were presented within context. The results are discussed as a result of the context-dependent nature of action knowledge.

Situating Styles Adam Toon University of Exeter In a series of influential articles, the philosopher Ian Hacking has argued that we find a number of

different styles of reasoning within scientific practice, each with its own history. Furthermore, Hacking argues that, when taken together with positivist theories of meaning, styles of reasoning lead to a form of relativism: if the meaning of a scientific claim depends upon the style of reasoning appropriate to establishing its truth or falsehood, then the birth of a new style brings new propositions into being as candidates for truth or falsehood. As a result, styles cannot be subjected to independent criticism, since the propositions they evaluate have no meaning outside of the style. In his more recent work on styles, Hacking has placed less emphasis on their relativistic implications. Two other developments are also important for the present paper. The first is that Hacking is keen to stress that styles of reasoning are not styles of thinking, since “thinking is too much in the head” and omits “the manipulative hand and the attentive eye” (1992, pp. 3 – 4). Styles involve an “embodied creature [that] uses not just its mind but its body to think and to act in the world” (2012, p. 600). The second important development is that Hacking now links styles of reasoning to a burgeoning form of inquiry that he calls cognitive history. This is “the study of how an organism with certain cognitive capacities, on a planet like this, developed (etc.)” (2012, p. 607), exemplified by works such as Renfrew, Frith, and Malafouris’ The Sapient Mind: Archaeology Meets Neuroscience (2009). Recently, Malafouris (2013) has argued that an appropriate theoretical framework for such studies can be found in recent work in areas such as situated, embodied, extended, and distributed cognition, which reveals the importance of interaction between the brain, body and environment in our cognitive processes. In this paper, I will ask how we might draw on these frameworks to understand styles of reasoning in science, thereby underpinning Hacking’s own emphasis on the role of the body in scientists’ reasoning. Interestingly, I will argue, this approach to understanding scientific reasoning might also be thought to give rise to a form of relativism, since some work in situated cognition suggests that people are unable to engage in certain thought processes in the absence of

DTS6 particular external, material devices. I will examine this ‘situated’ reading of styles of reasoning in detail, and ask what it might mean for our understanding of science.

7

Insight Enacted Frédéric Vallée-Tourangeau Kingston University People solve problems by recruiting and interacting with artefacts. Through this interactivity, a problem representation evolves and is distributed across resources internal and external to the reasoner. Real-world interactivity was domesticated in a laboratory environment by offering different artefacts to participants in a low and a high interactivity group. An insight problem involving the spatial rearrangement of sets, but masquerading as an arithmetic problem, was employed. Participants in the low interactivity group were never able to break the impasse, that is to abandon their representation of the problem as one involving an arithmetic solution. Participants in the high interactivity group were more likely to break the impasse and discover a productive representation of the problem. Video evidence reveals substantial differences in the manner with which participants ‘thought’ about the problem as a function of the nature and degree of interactivity; insight was enacted through the manipulation of certain artefacts.

cognition. We argue that a radical departure from the classical information-processing model is untenable because higher-level cognition is fundamentally representation-based. However, we also argue that classical accounts of thinking put too great an emphasis on the role of internal representations and mental processing. This obscures the symbiotic relationship between thinking and acting and the role of spatiotemporal dynamics and ecological affordances on thinking. To fully understand how people think, solve problems, and make decisions, we need to break from traditional conceptions of thinking activities as sequestered in a static mind, transcend current debates about the localisation of cognition and, instead, focus our efforts towards better understanding how thinking emerges in ecological space and ecological time from the transactional flow of action and representational opportunities outcropping from a dynamic agent-environment interface.

!!!!!!

Posters Thinking Together with Material Representations: The Role of Joint Epistemic Actions in Creative Cooperation Johanne Stege Bjørndahl Aarhus University In many situations in people’s everyday lives, cognition is a highly situated and social activity. This study investigated the role of material representations in collective creative processes. How do material objects shape and aid joint epistemic processes? How do people coordinate and spontaneously distribute cognitive labor when solving an open-ended creative task together? These questions guided a qualitative study of social interactions in 6 groups of 4-5 participants solving a series of creative workshop tasks involving LEGO blocks as part of a psychology experiment. A qualitative microanalysis of the video recordings of the interactions identified three different roles that the

The Spatio-temporal Dynamics of Systemic Thinking Gaëlle Vallée-Tourangeau Kingston University Recent developments in cognitive science reject the classical view of cognition as a cerebral activity involving the rule-based processing of symbols inside the mind and call for a reconceptualization of cognition as emerging in a system encompassing the brain and the body in situ. Current dissenting views comes with two corollaries. First, representations are unnecessary to explain complex behaviours. Second, there is a spatio-temporal dimension to

DTS6 material representations play in the joint epistemic processes involved in the tasks: illustration, elaboration and exploration. Firstly, the LEGO blocks were recruited for illustration, to support top-down structured processes, such as to represent already well-formed ideas in order to support communication and epistemic alignment. Furthermore, the LEGO blocks were examined and questioned in ways that gave rise to discussions, clarifications and highlighted underlying disagreements, this we call elaboration. Lastly, the LEGO blocks were used for exploration, that is, the material representations were experimented on and physical attributes were explored resulting in discoveries of innovative practices and new meaning potentials. The study points to a tendency for the more top-down ways of engaging material representations to be supportive of gaining and maintaining authority as well as to provide powerful tools for achieving epistemic alignment. However, more bottom-up oriented approaches were characterized by more distribution of cognitive labor, more unconventional usage of materials and more possibility for unanticipated solutions to take place that shaped the problem solving processes.

8

It All Adds up: Interactivity and Expertise in Mental Arithmetic Charlotte L. Harris and Lisa G. Guthrie Kingston University People solve problems in a situated and contingent manner. The problem presentation shifts dynamically as people interact with its constituent elements, either through the rearrangement of physical artefacts or the space that configures and contains the problem. Recently, laboratory-based research on problem solving aimed to recreate the dynamic nature of problem solving by designing problem tasks in mental arithmetic that encourage interactivity. The fixed, paper-based problem presentation of conventional mathematics may discourage ingenuity and re-formulation of complex sums. In contrast, allowing participants a more dynamic

problem space offers the prospect of improving efficiency in problem solving. However, any benefits derived by re-designing the problem task have the potential of being eroded by expertise in that given field, equally those low expertise folk may benefit from an altered problem presentation. The experiment presented here investigated problem solving from a distributed cognition perspective, examining the impact of interactivity on mental arithmetic for ‘maths experts’ and ‘non-maths experts’. Results showed that in the high interactivity condition, low expertise participants equipped with tokens were more accurate and efficient in completing the addition problems than in a low interactivity paper-based condition. In addition, average deviation from the correct solution was also reduced, however the use of tokens increased latency. Results showed the effect of interactivity was different for participants with greater expertise: Efficiency and accuracy only marginally improved, and average deviation from the solution increased in the high interactivity condition, compared to the low interactivity problem presentation. Latency in both high and low interactivity conditions remained constant. These results indicate that mental arithmetic performance may be improved for those with low expertise by using a distributed problem-solving environment. The benefits of interactivity, however, may be reduced for those with greater maths expertise. Enaction, Anti-representationalism, and Content in Natural Languaging Matthew Harvey University of Cambridge This paper outlines an enactivist alternative to the idea that natural language is necessarily content-involving, framed as a reply to Daniel Hutto and Erik Myin (2013). The authors simultaneously defend a thoroughgoing antirepresentationalism about sensorimotor engagement with the world and (at least prima facie) accept a classical representationalist account of meaning in natural language. I argue that they are mistaken, however, in identifying both of these two positions with radical

DTS6 enactivism, because the idea of enaction - taken to be a positive argument to the effect that biologically autonomous organisms “enact” or “bring forth” their worlds – (i) entails antirepresentationalism without reducing to it but also (ii) precludes any account of linguistic meaning in terms of informational content. In this light it is important that Hutto and Myin’s antirepresentationalism is logically independent from their position on linguistic meaning; the former is motivated by the “Hard Problem of Content” (HPC), which demonstrates that explanatory naturalism is incompatible with an account of informationally contentful representations. The latter is an assertion, for which Radicalizing Enactivism does not offer much support, that language-involving behavior both instantiates conditions of felicity (i.e., is informationally contentful) and is exempt from the HPC. I depart from Hutto and Myin in adopting what they call “Really Radical Enactive Cognition”, which holds that cognition is never, under any circumstances, contentful or representational. I accept as premises both (a) that informationally contentful accounts of perception are untenable (due to the HPC and related considerations), and (b) that languaging “behaves as if” it were contentful. I outline a conception of this phenomenal aspect of languaging that draws on core enactive accounts of non-contentful meaning-making activity complexified by Steiner and Stewart’s (2009) notion of “heteronomy” as a socialized complement to the “immanent normativity” of biological autonomy. I introduce the idea of “attentional technologies” to describe the coenaction of joint-attentional structures (that is, minimally heteronomous perceptual entities). This theoretical construct allows me to link bottom-up enactive theory to the philosophy of techniques and to distributed linguistics, and so to resolve the tension between the HPC and the seeming-contentfulness of meaning in natural language. The view of languaging that emerges is, I believe, broadly compatible with Hutto and Myin’s position as well as with early enactive accounts of language.

9

Changing the Cognitive Landscape: Material Engagement in Problem Solving Angeliki Makri, Erica Mundahl, and Andra Popa Kingston University Traditional cognitive psychology research on problem solving proceeds on the basis of experimental procedures wherein the degree of interactivity and the possibility of manipulating the problem presentation are limited or eliminated. As a consequence, participants’ problem solving performance is constrained and limited by a sterilized environment which amputates cognition, an environment that lacks ecological validity. The experiment reported here investigated problem solving performance with an analytic and an insight problem as a function of the level of interactivity. In a low interactivity condition, participants worked on the problems using an electronic tablet. In a high interactivity condition, participants were given artefacts that configured the problem presentation and were invited to manipulate them in order to solve the problems. The high interactivity condition fostered a dynamic and fluid problem presentation that anticipated and complemented the participants’ hunches as they worked through a solution. Solution rates in the high interactivity condition were higher for both the analytic and insight problem, and significantly so for the latter. Measures of creativity and need for cognition were significant predictors of performance for the analytic problem in both low and high interactivity conditions, but not for the insight problem. In turn, working memory was a significant predictor of performance for the insight problem, again, across conditions. High interactivity enhanced problem solving performance and led to a higher solution rates especially for the insight problem. This suggests that in the high interactivity condition, the correct solution is the product of a largely contingent trajectory guided and constrained by the shifting action affordances cued by the dynamic problem presentation. These findings support the fruitfulness of adopting a distributed cognition perspective on problem solving and encourage researchers in this area to question the traditional efforts to limit interactivity in thinking experiments.

DTS6

10

!

!"#$%&'# !"#$$%&$'&(!)"#$%$*)+& (#")*+%"&, ",-.-/$%$*)&0&!$"-$%$*) !"#$"#

1234567286&9:&(;<=>9?9@<

! " Systemic Cognition Lab
http://syscoglab.wordpress.com/about/ http://issilc.org/

http://fass.kingston.ac.uk/research/

Philosophical Psychology, 2014 Vol. 27, No. 1, 112–125, http://dx.doi.org/10.1080/09515089.2013.828371

Friends at last? Distributed cognition and the cognitive/social divide
Adam Toon
Downloaded by [University of Exeter] at 03:39 18 February 2014

Distributed cognition (d-cog) claims that many cognitive processes are “distributed” across groups and the surrounding material and cultural environment. Recently, Nancy Nersessian, Ronald Giere, and others have suggested that a d-cog approach might allow us to bring together cognitive and social theories of science. I explore this idea by focusing on the specific interpretation of d-cog found in Edwin Hutchins’ canonical text Cognition in the wild. First, I examine the scope of a d-cog approach to science, showing that there are important disputes between cognitive and social theorists on which d-cog remains silent. Second, I suggest that, where social explanations can be recast in d-cog terms, this reformulation will not be acceptable to all social theorists. Finally, I ask how we should make sense of the claim that, on a d-cog analysis, social factors are cognitive factors. Keywords: Distributed Cognition; Edwin Hutchins; Nancy Nersessian; Ronald Giere; Sociology of Scientific Knowledge

1. Introduction There is sometimes thought to be an opposition between cognitive and social theories of science. Perhaps the clearest instance of this opposition is Latour and Woolgar’s infamous “ten-year moratorium on cognitive explanations of science” (1986, p. 280). On one side of the divide, social accounts emphasize scientists’ social and political interests or institutional structures. On the other, cognitive accounts refer mainly to scientists’ cognitive processes. In philosophical discussions, the term ‘cognitive’ is often associated with terms such as ‘rational’ or ‘truth-conducive’. In the present context, however, ‘cognitive’ is used in the sense found in cognitive science, to refer to psychological processes such as perception, reasoning, memory, and so on (Giere & Moffatt, 2003, p. 302). Processes that are cognitive in this sense can, of course, fail to be rational or truth-conducive. The divide at issue here is therefore distinct from that
Adam Toon is a Marie Curie Fellow at the University of Exeter. Correspondence to: Adam Toon, Department of Sociology, Philosophy and Anthropology, University of Exeter, Amory Building, Rennes Drive, Exeter EX4 4RJ, United Kingdom. Email: a.toon@exeter.ac.uk
q 2013 Taylor & Francis

Philosophical Psychology

113

concerning the relationship between social factors and the rationality of science (e.g., Longino, 1990; Solomon, 2001). Recently, Nersessian (2005) has argued that the perceived divide between cognitive and social theories is based on a mistaken “Cartesian” view of cognition, summed up in the tenets of GOFAI (“Good Old Fashioned Artificial Intelligence”) (Haugeland, 1985). According to GOFAI, cognition involves computational processes on symbolic representations internal to the individual mind. Within cognitive science, GOFAI is increasingly challenged by a range of different approaches. Particularly important for Nersessian is distributed cognition (d-cog). In contrast to GOFAI, d-cog claims that many cognitive processes are “distributed” across social groups and the wider material and cultural environment. Together with a team of researchers, Nersessian has offered an analysis of the laboratory as an “evolving distributed cognitive system” (e.g., Nersessian, Kurz-Milcke, Newstetter, & Davies, 2003a; Nersessian, Newstetter, KurzMilcke, & Davies, 2003b). She suggests that this approach allows us to “integrate” cognitive and social accounts of science, as well as yielding other important insights. For example, d-cog might allow us to see that the nature of cognitive processes in science has changed over time (2005, p. 52). Giere (2002a, 2002b, 2002c, 2004, 2006, 2007, 2012; Giere & Moffatt, 2003) has expressed similar hopes for d-cog. As well as applying the approach to a number of aspects of scientific practice, such as the use of diagrams and models, Giere has suggested that we reinterpret well-known studies by sociologists such as Latour and Knorr-Cetina in d-cog terms (Giere, 2002b; Giere & Moffatt, 2003). Like Nersessian, Giere believes that d-cog “bridges the often perceived gap between cognitive and social histories of science” (2002a, p. 285) and that the approach offers important historical insights. For example, it allows us to see that “what powered the scientific revolution was an explosion of new forms of distributed cognitive systems” (2002a, p. 298). At times, Giere also goes further than Nersessian, arguing that, on a d-cog approach, the cognitive and social “overlap” (2002a, p. 296) or “merge” (Giere & Moffatt, 2003, p. 304). For example, referring to Knorr-Cetina’s analysis of experiments in highenergy physics, Giere argues that d-cog provides “a complementary, cognitive, account of these experiments” (2002b, p. 639). In this account, “we can now say that aspects of the situation that seemed only social are also cognitive” (2006, p. 114). In this paper, I try to assess d-cog’s potential to bring together cognitive and social approaches to science. Exactly what is meant by ‘distributed cognition’ is sometimes unclear. In section 2, I distinguish between a general and a more specific interpretation of the approach. Without further development, the general interpretation of d-cog is difficult to assess. I therefore focus on the more specific interpretation, found in Hutchins’ Cognition in the wild (1995). Hutchins’ approach is introduced in section 3. In section 4, I consider the scope of d-cog: which aspects of science might be understood using this approach, and which, if any, will remain beyond its reach? In doing so, I argue that there are important disputes between cognitive and social theorists on which d-cog remains silent. In section 5, I suggest that, where social explanations can be recast in d-cog terms, this reformulation will not be acceptable to all social theorists. Finally, in section 6, I ask how we should make sense of Giere’s

Downloaded by [University of Exeter] at 03:39 18 February 2014

114

A. Toon

claim that, on a d-cog analysis, social factors are cognitive factors. Can this aspect of dcog help us to reconcile cognitive and social approaches? 2. What is Distributed Cognition? The canonical text for distributed cognition is Hutchins (1995). This is an ethnographic study of navigation on a U.S. Navy ship, which Hutchins calls the Palau. Hutchins offers a detailed analysis of how the Palau’s crew accomplishes certain tasks, such as determining the ship’s position and planning its course. Typically, he finds, these tasks are not performed by any one individual, but by a group of crew members working together. Each of the crew members also makes use of a variety of different tools. The processes that accomplish the tasks are therefore distributed across the members of the team and the tools they use. Moreover, the social structure plays a crucial role in the crew’s activity: the way in which tasks are performed depends upon the command structures that govern the crew’s interaction. On the ship there is a “social distribution of cognitive labor” (Hutchins, 1995, p. 228). For example, consider the “fix cycle.” This is the procedure in which the navigation team determines the ship’s position by taking visual bearings to landmarks on either side of the ship. The fix cycle involves a number of different crew members. Inside the pilothouse are the navigation plotter and the bearings recorder. Together, they decide on suitable landmarks for obtaining a position fix. They then pass on the names of these landmarks to the pelorus operators, who stand on either side of the ship. The pelorus operators must identify these landmarks on the horizon and take their bearings. To do so, they use a device called an alidade, which has a hairline sight aligned with a gyrocompass scale. The pelorus operators relay the bearings to the pilothouse, where the recorder notes them in the log and the plotter plots them on the chart to determine the ship’s position and project its future course. The plotter uses various tools to carry out this task, such as a hoey, which is a special protractor with a long arm that can be set to the recorded bearing. Hutchins’ analysis thus encompasses both the cognitive acts of individuals (such as reading the bearing of a landmark in the alidade sight) and the social structure guiding their interactions (such as the ship’s command hierarchy). In this respect, his analysis is perhaps not so remarkable. After all, as both Giere and Nersessian acknowledge, not all sociological accounts are as hostile to cognitive or psychological explanations as Latour and Woolgar (e.g., Bloor, 1976). Conversely, cognitive theorists sometimes recognize the importance of social context (e.g., Dunbar, 1995). What exactly is distinctive about a d-cog approach? In particular, how does d-cog offer a reinterpretation of sociological accounts, such as Knorr-Cetina’s? The distinctive feature of d-cog, it seems, is that it treats social groups, along with tools and parts of the material and cultural environment, as cognitive systems. Hutchins’ key claim is that the navigation team may be analyzed as a “cognitive and computational system” (1995, p. xiv). This passage already points to two different interpretations of the approach, however. According to the first interpretation, to analyze an activity as d-cog is to understand it as a cognitive process. According to the

Downloaded by [University of Exeter] at 03:39 18 February 2014

Philosophical Psychology

115

Downloaded by [University of Exeter] at 03:39 18 February 2014

second, it is to understand the activity as a computational process. While the first approach is not committed to any particular view of the nature of cognition, the second claims that cognition takes a specific form, namely computation. As we will see in section 3, it is the computational version of d-cog that underpins much of Hutchins’ (1995) analysis. Both Nersessian and Giere sometimes distance themselves from the claim that cognition is computational (Giere, 2006, pp. 107– 108; Osbeck & Nersessian, 2006; see also Brown, 2011, p. 28). Nevertheless, it is worthwhile assessing the computational approach, for a number of reasons. First, Hutchins’ book remains the locus classicus for work on distributed cognition and forms the central inspiration for both Nersessian and Giere. Second, as we shall see in section 3, the computational version of d-cog has much to recommend it. Third, the computational view has the virtue of being more specific, and therefore easier to assess, than the first interpretation of d-cog. If d-cog claims only that social groups are cognitive systems, without saying more about what cognitive systems are, then it is difficult to see how to evaluate the approach. In what follows, I shall therefore focus on the computational interpretation of d-cog, although some of what I say (especially in section 6) will apply to the more general interpretation as well.

3. Ships as Computers At the outset, Hutchins tells us that his study is an attempt “to apply the principal metaphor of cognitive science—cognition as computation—to the operation of [the navigation team]” (1995, p. 49). Although he is skeptical of GOFAI’s claim that what goes on inside the head is computational, Hutchins believes that this analysis may be applied to the navigation team as a whole: “the system formed by the navigation team can be thought of as a computational machine” (1995, p. 228). Moreover, “the computation observed in the activity of [the navigation team] can be described in the way cognition has been traditionally described—that is, as computation realized through the creation, transformation, and propagation of representational states” (Hutchins, 1995, p. 49). It is important to note that Hutchins does not intend his analysis to be merely metaphorical. In his view, applying the language of computation to the navigation team is “not a metaphorical extension at all” (1995, p. 364). Hutchins develops his analysis by drawing on Marr’s (1982) distinction between three different levels on which a cognitive system may be understood. At the computational level we have an abstract description of the computation that a system carries out. Thus, fixing a position may be understood as combining two one-dimensional constraints to give a unique position in two-dimensional space. The representational level concerns “the choice of representation for the input and output and the algorithm to be used to transform one into the other” (Marr, 1982, pp. 24– 25). For example, the crew of the Palau uses the standard Western coordinate system and one-dimensional constraints are given by lines of position. Finally, the implementation level specifies how the representational system and algorithm are physically implemented. On the Palau, for

116

A. Toon

instance, the pelorus operators determine the bearings that give the line of position, while the plotter determines their intersection on the chart. It will be helpful to see some examples of how Hutchins applies this analysis to social interactions. Within computer science, a daemon is an agent that monitors for certain trigger conditions, and takes a specified action when those conditions are met. During the Palau’s voyage, Chief Richards instructs another member of the crew, Smith, to watch the fathometer and report when the depth of water falls below 20 fathoms. Hutchins claims that here we have “the social construction of an information-processing mechanism” (1995, p. 192). By giving the order to Smith, Chief Richards reconfigures the system formed by the navigation team to create a daemon which will respond to certain conditions and output a symbolic signal when those conditions are met. Hutchins offers a similar analysis of “phone talkers.” These are members of the crew posted at each end of the ship’s telephone lines. The job of a phone talker is to receive messages and relay them to the relevant crew members when there is a suitable break in their activity. In Hutchins’ analysis, phone talkers are information buffers, which “[permit] communication to take place when the sender and the receiver are not overloaded” (1995, p. 195). The bearing recorder and log are also information buffers, since they enable the pelorus operators and the plotter to work asynchronously: the plotter need not plot each bearing as it is reported, but can instead refer back to the bearing recorder or log. The log is also a memory, as well as a filter, which “passes the bearings without passing the temporal characteristics of their production” (1995, p. 195). In Hutchins’ original formulation, then, d-cog is a claim about what happens at the implementation level of a computation. Rather than taking place within the mind of any individual, the algorithm determining the ship’s position is implemented by a distributed system consisting of the entire navigation crew. This computational version of d-cog has a number of attractions. First, and perhaps most important, it offers a clear rationale for the claim that d-cog integrates the cognitive and the social: according to the computational view, social organization becomes part of the way that the relevant computation is implemented. As Hutchins puts it, we may “treat the social organization as a computational architecture” (1995, p. 185). Second, the computational approach provides a powerful tool for understanding different social arrangements and assessing their epistemic merits. In this respect, it shares the attraction of Thagard’s earlier proposal (1992) that we understand science using distributed artificial intelligence (D.A.I.). Third, d-cog also avoids one of the main objections against Thagard’s proposal because, unlike D.A.I., d-cog is not committed to claiming that scientists’ internal cognitive processes are computational (Thagard, 1992, p. 58). 4. Uncharted Waters Neither Nersessian nor Giere claim that d-cog offers a complete theory of science, nor that it allows us to reconcile all disputes between cognitive and social theorists. In light of this, it is important to ask how far d-cog might be able to take us. Which aspects of science might be analyzed using a d-cog approach and which, if any, will remain out of

Downloaded by [University of Exeter] at 03:39 18 February 2014

Philosophical Psychology

117

its reach? These questions are rarely explicitly addressed in the literature. And, at first glance, it might seem difficult to answer them at this stage, since even the computational version of d-cog remains rather general. In itself, the approach claims only that laboratories carry out some form of computation; it does not say what that computation is or exactly how it is implemented. In fact, however, I want to suggest that any computational d-cog analysis will remain silent on some parts of scientific practice. To see this, recall that d-cog is a claim about the implementation level of a computation: Hutchins shows how the computation determining the Palau’s position is implemented by the entire navigation team and their social interactions. It is in this sense that d-cog offers a means of integrating cognitive and social factors. Notice, however, that even if we accept such an analysis, a number of issues will remain to be addressed. First, a d-cog analysis will not tell us why a particular computation is being performed or why a specific representational system is being used to carry it out. Second, the analysis will not tell us how a representational system gains its representational status. When applied to science, I suggest, both of these questions receive competing answers from social and cognitive accounts. As a result, they will remain areas where a d-cog analysis of the laboratory cannot help to reconcile debate between the two. First, consider the choice of representational system. Alongside his study of the Palau, Hutchins offers an analysis of traditional navigation techniques in Micronesia. Hutchins argues that the Western and Micronesian systems are, in fact, the same at the computational level: both combine one-dimensional constraints to give a unique position in two-dimensional space. But the two differ radically at the representational level. For example, rather than taking the boat to be moving across a fixed twodimensional space, Micronesians think of the canoe as stationary while the water moves past it. Progress during a journey is represented not by a unit of length but by the changing bearing of a reference island. One-dimensional constraints are sometimes provided by sightings of birds, which indicate the canoe’s distance from land. There are many questions that we might ask about these representational systems. Why do Micronesian navigators think of the canoe, rather than the water, as fixed? Why do we employ standard units of distance and time and embody these in charts and other devices, while the Micronesians do not? Are there still other ways in which humans might represent the world to find their way around it? Each of these questions concern the choice of representational scheme used to carry out a computation, not the way that the computation is implemented. As a result, they cannot be answered through a d-cog analysis of a particular navigation practice. Hutchins himself offers many fascinating insights into the origins of the representational scheme used in Western navigation (1995, chapter 2). He does so, however, not by relying on his analysis of the Palau, but instead by looking at the history of charts and other navigational tools in the West. In Hutchins’ view, the features of the Western representational system are historically contingent, rather than “natural and inevitable or simply the consequences of the interaction of human nature with the demands of a given task” (1995, p. 114). Many of the questions that we ask about the sciences also concern representational systems. For example, consider classification schemes. These have been the focus of

Downloaded by [University of Exeter] at 03:39 18 February 2014

118

A. Toon

Downloaded by [University of Exeter] at 03:39 18 February 2014

considerable attention in the sociology of scientific knowledge. In fact, according to Bloor, “one of the central propositions in the sociology of knowledge” is that “the classification of things reproduces the classification of men” (1982, p. 267). By contrast, some cognitive scientists claim that there are important similarities in the way that different cultures categorize the world that are due to universal cognitive constraints (e.g., Berlin, 1992). I do not wish to enter into this debate here. All I want to point out is that, like questions concerning the differences between Western and Micronesian navigation, this debate about classification schemes concerns the choice of representational system, rather than its implementation. As a result, if there is an opposition here between sociological and cognitive accounts, then a d-cog analysis of the laboratory cannot help to overcome this opposition. The second issue that d-cog does not address is that of how representational systems gain their representational status (that is, why they have meaning). The reason for this, of course, has to do with a general feature of the computational view. As Crane puts the point, “a computational process is, by definition, a rule-governed or systematic relation among representations. To say that some process or state is computational does not explain its representational nature, it presupposes it” (2003, p. 169). In line with Hutchins’ analysis, d-cog would treat the laboratory as a computational system creating, manipulating, and destroying representations. This leaves open the question of how those representations come to represent the world. Once again, this is a question that receives very different answers from social and cognitive accounts of science. For example, members of the Edinburgh Strong Programme in the sociology of science propose a theory of meaning known as finitism, drawn from the later Wittgenstein (Barnes, Bloor, & Henry, 1996). According to finitism, meaning is a fundamentally social phenomenon. By contrast, of course, many cognitive scientists and philosophers reject this view, instead seeking to explain meaning in terms of nonsocial factors, such as causal relations. It might be thought that d-cog may legitimately defer these disputes about the nature of meaning. And yet finitism is arguably the central element of the Strong Programme, which gives rise to many of its key claims, such as epistemic relativism (e.g., Kusch, 2002). Here again, then, we find an important dispute between which d-cog cannot help to reconcile. 5. A Storm Brewing? Section 4 pointed to aspects of science that will be omitted from a d-cog analysis. Let us now consider those parts of science that d-cog does seek to analyze, and ask whether this analysis can help to reconcile cognitive and social accounts. Unfortunately, I believe that the prospects for reconciliation here may be limited in an important respect, since the analysis offered by d-cog will not be acceptable to all social theorists. In fact, rather than bridging the gap between cognitive and social theories, d-cog threatens to return us to an old debate between the two, sparked by Slezak’s paper, “Scientific discovery by computer as empirical refutation of the Strong Programme” (1989). Slezak focused on computer discovery programs such as BACON (Langley, Bradshaw, & Simon, 1983). Given the relevant data, BACON is able to “rediscover”

Philosophical Psychology

119

Downloaded by [University of Exeter] at 03:39 18 February 2014

various empirical laws, such as the ideal gas law. According to Slezak, programs like BACON constituted a decisive refutation of the Strong Programme in the sociology of scientific knowledge (SSK), which claims that “social factors are an irreducible component of scientific discovery” (1989, p. 564).1 In fact, for Slezak, the “very possibility of computer programs making scientific discoveries poses a fundamental challenge to SSK’s radical claims” (1989, p. 564). Slezak’s argument proved controversial. Even supporters of cognitive accounts took issue with his claim that BACON was capable of making scientific discoveries. For example, Thagard noted that such programs “are highly simplified and do not constitute full simulations of how discoveries were actually made” (1989, p. 654). Interestingly, however, most commentators agreed that, if computer programs could make discoveries, SSK would be refuted. For example, in response to Slezak’s paper, Giere wrote:
A minimal claim of a sociology of scientific knowledge (SSK) would be that the course of science is necessarily influenced by the human interests of scientists, which are in turn partially shaped by their social relationships. If computers have no human interests and no social relationships, and yet can make scientific discoveries, this minimal claim would be refuted. (1989, p. 639)

Similarly, Collins agreed that “if [Slezak’s premise] is taken to mean that BACON reproduces social collectivities then, if true, it would be fatal for ‘the strong programme’ and its variants.” Why? “Because BACON is not a social collectivity” (1989, p. 614). The trouble, I think, is that d-cog poses a similar challenge to social theories. Recall that, according to Hutchins, the navigation team is a “computational machine” (1995, p. 185). As Hutchins himself reminds us, computation is independent of the physical medium in which it is implemented (1995, p. 51). As a result, if the laboratory is a distributed cognitive system, then the computation it performs could in principle be carried out on some other system entirely (Magnus, 2007, p. 299). And this system need not be a social one. In fact, as far as d-cog is concerned, the social is only the “hardware” on which the computation happens to be run. Consider phone talkers, for example. According to Hutchins, they are information buffers, controlling the flow of information between different parts of the system. On the Palau this is achieved by assigning crew members a particular role within the social organization of the ship. But there is nothing essentially social about an information buffer. If Hutchins’ analysis is correct, the phone talker might as easily be replaced by a silicon circuit, so long as it implements the same computation. Of course, d-cog would not support all of the conclusions that Slezak wanted to establish using BACON. For example, Slezak argued that, since BACON implements general rules of problem solving, its success points to the existence of “context-free or super-cultural norms of rationality” (1989, p. 572). A d-cog approach does nothing to suggest the existence of such norms: the computations carried out in different laboratories might vary widely with different social contexts. Nevertheless, like BACON, d-cog would appear to be at odds with SSK’s claim that science is essentially

120

A. Toon

Downloaded by [University of Exeter] at 03:39 18 February 2014

social. Similar claims are also found in the work of authors outside SSK. For example, Longino writes that science is “necessarily social” (1990, p. 12). There are a number of ways in which we might try to resolve the apparent conflict here. Recalling our discussion in section 4, social theorists might argue that social factors are necessary for those aspects of science omitted from d-cog, such as the semantics of representational schemes. Alternatively, we might construe claims regarding the essentially social nature of science in some weaker sense, which is not threatened by the possibility of non-social implementation raised by d-cog. For example, perhaps some social theorists might rest content with the claim that, while a non-social science is possible, actual scientific practice is social, and likely to remain so. Others, however, are keen to argue for a stronger position (e.g., Kusch, 2002). Taking a slightly different line, Thagard argues that D.A.I. is compatible with Longino’s view, since computer networks can capture the critical interactions that she takes to be required for scientific objectivity (1992, p. 61). Once again, these issues cannot be explored fully here. Much will depend upon the details of particular social theories. Nevertheless, I think it is important to note that there is an apparent conflict here and that this conflict must be resolved if d-cog is to bridge the gap between cognitive and social accounts of science. 6. Merging the Cognitive and Social One way in which d-cog promises to bridge the gap between social and cognitive theorists is by revealing that social factors are cognitive factors. For example, recall Giere’s claim that, on a d-cog analysis, “aspects of the situation that seemed only social are also cognitive” (2006, p. 114). In a similar vein, he writes that:
Thinking of science in terms of systems of distributed cognition enlarges the domain of the cognitive in our understanding of science. It is typically assumed that there is a sharp divide between the cognitive and the social. From the perspective of distributed cognition, what many regard as purely social determinants of scientific belief can be seen as part of a cognitive system, and thus within the purview of a cognitive understanding of science. There is no longer a sharp divide. The cognitive and the social overlap. (Giere, 2002a, p. 296)

With d-cog, we are told, the cognitive and social “merge” (Giere & Moffatt, 2003, p. 304). How exactly should we understand these claims? And how might this aspect of dcog help us to reconcile cognitive and social theories of science? To answer these questions, we first need to understand what is meant by ‘cognition’ in this context. In everyday contexts, we normally use the term ‘cognitive’ to refer to internal, psychological processes such as perception, memory, reasoning, and so on. Moreover, it was broadly this sense of cognition that was at stake in our original dispute between cognitive and social theories of science: social theorists argue for the importance of social factors, such as social status or institutional structures, while cognitive theorists stress internal, psychological processes such as reasoning, perception, memory, and so on. Given our usual notion of cognition, we might wonder how external processes such as social interactions could possibly count as cognitive. However, Giere is keen to stress

Philosophical Psychology

121

that the concept of cognition invoked by d-cog is not our everyday one: “we are developing a science of cognition. In so doing we are free to make cognition a technical scientific concept different from everyday notions” (2002b, p. 642; see also Giere, 2006, p. 112). What is this technical notion? What does it mean to say, for example, that the Palau or the laboratory is a cognitive system in this technical sense? Giere suggests that “the reason for calling these systems cognitive systems rather than, say, transport systems or agricultural systems, is that they produce a distinctly cognitive product, knowledge” (2002b, p. 642). Elsewhere, he expands on this idea:
A distributed cognitive system is a system that produces cognitive outputs, just as an agricultural system yields agricultural products. The operation of a cognitive system is a cognitive process. . . . But what makes the output of the system a cognitive output? Here I think the only basis we have for judging an output to be cognitive is that it is the kind of output we recognize as the result of human cognition, such as a belief, knowledge, or a representation of something. (Giere, 2006, pp. 112– 113)

Downloaded by [University of Exeter] at 03:39 18 February 2014

Thus, a cognitive system is one which produces a cognitive output, such as a belief, knowledge, and so on. In this new, technical sense, even distant galaxies used as gravitational lenses for the Hubble Space Telescope may count as part of a cognitive system (Giere, 2012, p. 201). If we understand ‘cognition’ in this way, then to claim that the laboratory is a cognitive system is simply to claim that its output is a cognitive state, such as a belief, knowledge, and so on. While this claim might be uncontroversial, it is difficult to see how it will help to reconcile debates between cognitive and social theories of science. After all, both sides in this dispute would agree that science produces a cognitive output, such as beliefs (and perhaps also knowledge). What they disagree about, of course, is the nature of the processes that produce that output. Showing that the laboratory is cognitive in the minimal sense that it produces a cognitive output would not appear to resolve any debates over the nature of the processes that lead to that output. So it seems that, if d-cog’s merger of social and cognitive factors is to bridge the gap between cognitive and social theorists, it must employ a more substantial concept of cognition. One obvious strategy would be to stress the computational aspect of d-cog. On this interpretation, to claim that social interactions are cognitive would be to claim that they implement a computation (and perhaps also that they result in a cognitive output, such as a belief). As we saw in section 5, this claim would be likely to receive more resistance from social theorists. But let us suppose that this resistance could be overcome. Would a computational analysis of social factors then help to reconcile cognitive and social theorists? Of course, it is likely that disputes will remain. Even if social theorists were to accept that social interactions are computational, they might still disagree over their relative importance compared to internal psychological processes, such as memory or reasoning. Nevertheless, d-cog’s proponents might argue that there is now an overarching concept of cognition—namely, cognition as computation—that encompasses both social processes and internal, psychological ones. The difficulty with taking this line, however, is that Hutchins himself does not think that internal

122

A. Toon

cognitive processes are computational. And, as we have seen, both Giere and Nersessian are also wary of the computational view. Put simply, then, the challenge facing d-cog would seem to be as follows. In the original dispute between social and cognitive theorists, ‘cognitive’ took something close to its usual meaning, referring to internal, psychological processes such as memory, perception, reasoning, and so on. According to Giere, d-cog invokes a new, and much broader, notion of cognition that applies to systems involving many individuals and their social interactions, as well as models, diagrams, instruments, and sometimes even distant galaxies. At the same time, Giere also argues that d-cog helps to resolve the dispute between social and cognitive theorists since it reveals that social processes are cognitive processes. The trouble is that, the further d-cog’s technical notion of cognition moves away from the everyday notion at stake in the original dispute, the harder it is to see how d-cog’s merger of cognitive and social factors helps to resolve that dispute. Unless d-cog’s notion of cognition displays at least some significant similarities with our usual notion, it is difficult to see how it enables us to close the gap between cognitive and social theories of science. Instead, what d-cog would seem to offer is simply a new way of analyzing the social aspects of science (one which understands social interactions as computational). At this point, one option for d-cog’s proponents is to endorse the extended mind thesis (Clark & Chalmers, 1998). Like Giere, proponents of the extended mind thesis talk of cognitive systems that exist outside our heads, and even outside our bodies. Unlike Giere, however, they argue that such claims do not involve a merely technical sense of ‘cognition’. Consider the well-known case of Otto and Inga. When Inga hears of an exhibition at the Museum of Modern Art (MoMA) she recalls that the museum is on 53rd Street and heads off. Otto is an Alzheimer’s patient who carries a notebook with him wherever he goes to record useful information. When Otto hears of the exhibition, he looks up the information in his notebook and heads off. Clark and Chalmers claim that Otto’s notebook plays a similar functional role to Inga’s biological memory. As a result, they argue, the notebook counts as part of Otto’s cognitive processes, not in any merely technical sense, but in precisely the same sense as Inga’s biological memory. Before consulting his notebook, Otto literally believes that MoMA is on 53rd Street, just as Inga believes this before consulting her memory. If external objects like notebooks can implement recognizably cognitive processes, like memory, then perhaps social interactions can too (Clark & Chalmers, 1998, pp. 17–18). Giere rejects the extended mind thesis, since he regards it as leading to unnecessary and unanswerable metaphysical puzzles about the mind (e.g., 2006, pp. 110– 113). It is for this reason that he insists that d-cog involves a technical concept of cognition. If dcog’s proponents were sympathetic to the extended mind thesis, however, then this would seem to offer a way to bridge the gap between cognitive and social theorists. Just like Otto’s notebook, scientists’ social interactions might be said to implement processes that are cognitive, not in any merely technical sense, but in exactly the same sense as internal, psychological processes, such as memory. Of course, one problem with this approach is that the extended mind thesis remains highly controversial (e.g., Menary, 2010). But another is that supporters of the extended mind use ‘cognitive’ in a

Downloaded by [University of Exeter] at 03:39 18 February 2014

Philosophical Psychology

123

Downloaded by [University of Exeter] at 03:39 18 February 2014

far more restricted manner than Giere. For example, in their original article, Clark and Chalmers argue that, in order to count as part of the cognitive process, an external device must meet conditions of “glue and trust” (Clark, 2010). These conditions require that, like Otto’s notebook, an external process must be a constant in the person’s life, directly available and automatically endorsed. The glue and trust conditions might perhaps be met in some cases on the Palau or in the laboratory. For example, perhaps the plotter always has his slide rule with him and trusts what it says without question. But many processes, and in particular many social processes, will not meet these conditions. The reports of the pelorus operators are not always directly available to the plotter, for example, and he sometimes questions what they tell him. So even if d-cog’s supporters were sympathetic to the extended mind thesis, further work would be needed to show that social processes are also cognitive.

7. Conclusion Distributed cognition offers a promising approach for understanding scientific practice. My aim in this paper has been to clarify the nature of that approach and explore its potential for bridging the gap between cognitive and social theories. In particular, I have discussed three apparent limitations on the computational form of d-cog as a means of reconciling cognitive and social theories. First, while d-cog might show how social interactions can implement a computation, it will not tell us why a particular representational scheme is used to carry out that computation or why those representations have meaning. Both of these aspects of science receive competing social and cognitive explanations. Second, the computational form of dcog will not be acceptable to all social theorists, since it implies that science is not essentially social. Finally, I’ve suggested that, if d-cog’s claim to merge social and cognitive factors involves only a technical concept of cognition, then it is difficult to see how it will help to reconcile debate between cognitive and social theorists. None of these issues should lead us to reject a d-cog approach to science, but they might perhaps help us to understand what such an approach can hope to achieve.

Acknowledgements I would like to thank Giovanna Colombetti, Ronald Giere, Jeff Kochan, Sabina Leonelli, Nancy Nersessian, Lisa Osbeck, Thomas Sturm, and Paul Thagard for extremely helpful discussion and correspondence about distributed cognition. Thanks also to audiences in Aberdeen, Athens, Bielefeld, Exeter, and Nancy, and to two anonymous referees, for very insightful comments and criticism. Note
[1] Following Slezak, I will use ‘SSK’ to refer to the Strong Programme and related approaches, rather than the sociology of science in general.

124

A. Toon

References
Barnes, B., Bloor, D., & Henry, J. (1996). Scientific knowledge: A sociological analysis. London: Athlone. Berlin, B. (1992). Ethnobiological classification: Principles of classification of plants and animals in traditional societies. Princeton: Princeton University Press. Bloor, D. (1976). Knowledge and social imagery. London: Routledge and Kegan Paul. Bloor, D. (1982). Durkheim and Mauss revisited: Classification and the sociology of knowledge. Studies in History and Philosophy of Science, 13, 267– 297. Brown, M. (2011). Science as socially distributed cognition: Bridging philosophy and sociology of ¨ we, T. Mu ¨ ller, & B. van Kerkhove (Eds.), Foundations of the formal science. In K. Franc  ois, B. Lo sciences VII (pp. 17 –31). London: College Publications. Clark, A. (2010). Coupling, constitution and the cognitive kind: A reply to Adams and Aizawa. In R. Menary (Ed.), The extended mind (pp. 81– 100). Cambridge, MA: MIT Press. Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 58, 7 – 19. Collins, H. M. (1989). Computers and the sociology of scientific knowledge. Social Studies of Science, 19, 613– 624. Crane, T. (2003). The mechanical mind: A philosophical introduction to minds, machines and mental representation. London: Routledge. Dunbar, K. (1995). How scientists really reason: Scientific reasoning in real-world laboratories. In R. J. Sternberg & J. Davidson (Eds.), Mechanisms of insight (pp. 363– 395). Cambridge, MA: MIT Press. Giere, R. (1989). Computer discovery and human interests. Social Studies of Science, 19, 638– 643. Giere, R. (2002a). Scientific cognition as distributed cognition. In P. Carruthers, S. Stitch, & M. Siegal (Eds.), The cognitive basis of science (pp. 285– 299). Cambridge: Cambridge University Press. Giere, R. (2002b). Distributed cognition in epistemic cultures. Philosophy of Science, 69, 637– 644. Giere, R. (2002c). Models as parts of distributed cognitive systems. In N. J. Nersessian & L. Magnani (Eds.), Model-Based reasoning: Science, technology, values (pp. 227– 241). Dordrecht: Kluwer. Giere, R. (2004). The problem of agency in scientific distributed cognitive systems. Journal of Cognition and Culture, 4, 759– 774. Giere, R. (2006). Scientific perspectivism. Chicago: University of Chicago Press. Giere, R. (2007). Distributed cognition without distributed knowing. Social Epistemology, 21, 313 – 320. Giere, R. (2012). Scientific cognition: Human centered but not human bound. Philosophical Explorations, 15, 199– 206. Giere, R., & Moffatt, B. (2003). Distributed cognition: Where the cognitive and the social merge. Social Studies of Science, 33, 301– 310. Haugeland, J. (1985). Artificial intelligence: The very idea. Cambridge, MA: MIT Press. Hutchins, E. (1995). Cognition in the wild. Cambridge, MA: MIT Press. Kusch, M. (2002). Knowledge by agreement. Oxford: Oxford University Press. Langley, P., Bradshaw, G. L., & Simon, H. A. (1983). Rediscovering chemistry with the BACON system. In R. S. Michalski, J. G. Carbonell, & T. M. Mitchell (Eds.), Machine learning: An artificial intelligence approach (pp. 307– 330). San Mateo, CA: Morgan Kaufmann. Latour, B., & Woolgar, S. (1986). Laboratory life: The construction of scientific facts. Princeton: Princeton University Press. Longino, H. (1990). Science as social knowledge: Values and objectivity in scientific inquiry. Princeton: Princeton University Press. Magnus, P. D. (2007). Distributed cognition and the task of science. Social Studies of Science, 37, 297 – 310. Marr, D. (1982). Vision: A computational investigation into the human representation and processing of visual information. San Francisco, CA: Freeman.

Downloaded by [University of Exeter] at 03:39 18 February 2014

Philosophical Psychology

125

Downloaded by [University of Exeter] at 03:39 18 February 2014

Menary, R. (Ed.). (2010). The extended mind. Cambridge, MA: MIT Press. Nersessian, N. J. (2005). Interpreting scientific and engineering practices: Integrating the cognitive, social, and cultural dimensions. In M. Gorman, R. Tweney, D. Gooding, & A. Kincannon (Eds.), Scientific and technological thinking (pp. 17 –56). Mahwah, NJ: Erlbaum. Nersessian, N. J., Kurz-Milcke, E., Newstetter, W. C., & Davies, J. (2003a). Research laboratories as evolving distributed cognitive systems. In R. Alterman & D. Kirsh (Eds.), Proceedings of the 25th annual conference of the Cognitive Science Society (pp. 857– 862). Mahwah, NJ: Erlbaum. Nersessian, N. J., Newstetter, W. C., Kurz-Milcke, E., & Davies, J. (2003b). A mixed-method approach to studying distributed cognition in evolving environments. In P. Bell, R. Stevens, & T. Satwicz (Eds.), Proceedings of the fifth International Conference of the Learning Sciences (pp. 307– 314). Mahwah, NJ: Erlbaum. Osbeck, L. M., & Nersessian, N. (2006). The distribution of representation. Journal for the Theory of Social Behaviour, 36, 141– 160. Slezak, P. (1989). Scientific discovery by computer as empirical refutation of the strong programme. Social Studies of Science, 19, 563– 600. Solomon, M. (2001). Social empiricism. Cambridge, MA: MIT Press. Thagard, P. (1989). Welcome to the cognitive revolution. Social Studies of Science, 19, 653– 657. Thagard, P. (1992). Societies of minds: Science as distributed computing. Studies in History and Philosophy of Science, 24, 49– 67.

DYNAMICS OF SYSTEMIC THINKING

1

The Spatio-temporal Dynamics of Systemic Thinking

Gaëlle Vallée-Tourangeau and Frédéric Vallée-Tourangeau Kingston University

Author Note Correspondence concerning this article should be addressed to either Gaëlle Vallée-Tourangeau or Frédéric Vallée-Tourangeau, Department of Psychology, Kingston University, Kingston-upon-Thames. UNITED KINGDOM, KT1 2EE, g.vallee-tourangeau@kingston.ac.uk or f.valleetourangeau@kingston.ac.uk, tel: +44 (0)208 417 2000, fax: +44 (0)208 417 2388.

DYNAMICS OF SYSTEMIC THINKING Abstract Recent developments in cognitive science call for a reconceptualization of cognition as emerging in a system that encompasses the brain and the body in situ, and rejects the classical view that cognition is a fundamentally cerebral activity based on the rule-based processing of symbols represented in the mind. A radical anti-thesis to the classical information-processing approach goes as far as positing that cognition emerges directly from people’s actions in the world. This view comes with two corollaries. The first corollary is that representations are unnecessary to explain complex behaviours, which can be understood from the study of the dynamic relation between an agent and her environment. The second corollary is that there is a spatio-temporal dimension to cognition as it emerges from a continuous and fluid coupling of neural and physical activity. In this paper, we discuss the implications of the ecological view of cognition for higher cognitive functions such as problemsolving, judgement, and decision-making. We argue that a radical departure from the classical information-processing model is untenable, notably because higher-level cognition is fundamentally representation-based, thus rejecting the first corollary of the radical embodiment approach. However, we also argue that classical accounts of thinking put too great an emphasis on the role of internal representations and mental processing. This obscures the symbiotic relationship between thinking and acting and the role of spatiotemporal dynamics and ecological affordances on thinking, thus embracing the second corollary of a systemic view of cognition. Finally, we discuss current ecological accounts of higher cognition and show that they are, at best, interactional. To fully understand how people think, solve problems, and

2

DYNAMICS OF SYSTEMIC THINKING make decisions, we need to break from the traditional conception of thinking activities sequestered in a static mind, and instead study how thinking emerges in ecological space and ecological time from the transactional flow of action and representational opportunities outcropping from a dynamic agentenvironment interface.

3

DYNAMICS OF SYSTEMIC THINKING The Spatio-temporal Dynamics of Systemic Thinking

4

Recent developments in cognitive science call for a reconceptualization of cognition as emerging in a system that encompasses the brain and the body in situ, and rejects the classical view that cognition is a fundamentally cerebral activity based on the rule-based processing of symbols represented in the mind. A radical anti-thesis to the classical information-processing approach goes as far as positing that cognition emerges directly from people’s actions in the world. This view comes with two corollaries. The first corollary is that representations are unnecessary to explain complex cognition, which can be understood from the study of the dynamic relation between an agent and her environment. The second corollary is that there is a spatio-temporal dimension to cognition as it emerges from a continuous and fluid coupling of neural and physical activity. In this paper, we discuss the implications of the ecological view of cognition for higher cognitive functions such as problem-solving, judgement, and decision-making. We argue that a radical departure from the classical information-processing model is untenable to understand how individuals think, notably because thinking is fundamentally representation-based, thus rejecting the first corollary of the radical embodiment approach. However, we also argue that classical accounts of thinking put too great an emphasis on the role of internal representations and mental processing. This obscures the symbiotic relationship between thinking and acting. By contrast, we argue that spatio-temporal dynamics and ecological affordances play a fundamental role in thinking, thus embracing the second corollary of a systemic view of cognition. Finally, we discuss current ecological accounts of higher cognition

DYNAMICS OF SYSTEMIC THINKING and show that they are, at best, interactional. To fully understand how people think, solve problems, and make decisions, we need to break from the traditional conception of thinking activities sequestered in a static mind, and instead study how cognition emerges in ecological space and ecological time. Cognition Is an Emergent Property of Dynamical Systems The missing part of the Turing machine Thinking is an elusive concept. Classically, cognitive psychologists have conceived human thinking as the mental process of pondering on, or reasoning about, ideas and opinions produced by the mind. Current views sometimes adopt a dual-process perspective, distinguishing between intuitive and deliberative processes (e.g., Kahneman, 2003), but nevertheless conceive thinking is a mental activity aiming to acquire knowledge and understanding. Following influential thinkers such as Jerry Fodor who argued that “the character of a mental state is independent of its physical realization” (1981, p. 119), most cognitive psychologists view cognition as the product of a “Turing Machine”, named after the British mathematician who first imagined it. Turing (1936) compared a man to a computing machine, endowed with an initial configuration or “state of mind”, a limited set of operations that he can apply to information brought to its conscious awareness, and a limited set of procedures stored in memory dictating the order and numbers of operations to be applied. In this model, cognition results from the desultory, stepwise, processing of information inside one’s head. The coupling of an individual’s state of mind with the value of the information in his conscious awareness at any moment in time determines which procedure should be applied to transform that information. This results in a new state of mind, which can then

5

DYNAMICS OF SYSTEMIC THINKING be coupled with the next piece of information to enter conscious awareness and identify the next procedure to be applied, and so on. The Turing machine metaphor has had a long-lasting impact on cognitive scientists’ conception of cognition; they have since sought to discover and model the mental procedures linking informational inputs and outputs, independently of their physical realization. Outside the realms of cognitive psychology, however, philosophers (Clarke, 2008), anthropologists (Hutchins, 1995), and cognitive scientists (Kirsh, 2013) have challenged this conception of cognition, arguing instead that cognition emerges in complex systems that include, but cannot be fully specified by, neural computations. Turing’s machine was intended to model a mathematician who proceeds towards a computational goal. Yet, it leaves out the person, with her eyes, ears and hands, and the physical world in which she implements those computations, including, for example, her notebook and fountain pen. It leaves out the notebook’s annotations, strikeouts, mistakes, drawings, quotes, and so on. It misses out the dead-ends and insights occurring over time. The Turing machine reduces this rich process to the application of pre-determined rules to strings of symbols and dismisses the contribution of the mathematician’s interactions in space and time with the material world as mere implementational details. We are left with a sterilised model of human cognition without a human being in it. As such, the Turing machine epitomises models of the abstract computations that result from the processing of symbols, but it is not an adequate model of human cognition because it fails to explain how people’s interactions with the material world through their eyes, ears, and hands support the instantiation of these

6

DYNAMICS OF SYSTEMIC THINKING computations (Hutchins, 1995). Not all dissenters agree on what alternative model we should adopt to account for human cognition. Proponents of the distributed cognition approach continue to characterise cognitive processes as involving mechanisms operating upon representations but they reject the assumption that those mechanisms and those representations are necessarily uniquely mental: “Minds are not passive representational engines, whose primary function is to create internal models of the external world.” (Hollan, Hutchins, & Kirsh, 2000, p. 177). So, rather than conceive cognition as emerging from the activity of the brain alone, the distributed cognition approach conceives cognition as emerging from the coordination of people’s inner resources and mental processes with the resources present in their immediate material and social environments, and the processes taking place in these environmental spheres. As such, cognition is taken to arise from the transmission, transformation, and coordination of both mental representations (e.g., as people mentally rotate a geometric form) as well as physical ones (e.g., as people physically manipulate geometric forms in a game of Tetris to speed up identification, Kirsh & Maglio, 1994). Other dissenters, however, are calling for a more radical reconceptualisation of cognition. For proponents of the radical embodiment approach (e.g., Chemero, 2009; Shapiro, 2011; Wilson & Golonka, 2013; Van Gelder, 1995), the Turing machine is an inadequate model of human cognition, not only because it confines cognitive processes to those located inside the brain, but also because it assumes that cognition emerges from the processing of complex representations. These views come with two corollaries. The first corollary is that representations and complex

7

DYNAMICS OF SYSTEMIC THINKING computations are unnecessary to explain complex behaviours. The second corollary is that there is a spatio-temporal dimension to cognition. We review both corollaries in the next two sections.

8

Neither representations nor complex computations are necessary to account for cognition Understanding how people store knowledge in their mind is a core research topic for modern cognitive psychologists. This effort, in turn, calls for a specification of the manner in which such stored knowledge is represented in the human mind (e.g., Neisser, 1967). The Cognitive Revolution gave rise to amodal conceptions of knowledge representations defined as “data structures processed independently of the brain’s modal systems for perception, action, and introspection” (Barsalou, 2010, p. 717). An alternative conception of knowledge representations emerged from a grounded approach to cognition, which posits that internal representations “have a situated character, implemented via simulations in the brain’s modal systems, making them well suited for interfacing with external structures.” (ibid.). Perception, in the eye of classical constructivist theories of cognition, is thus cast as a product of mental processes that disambiguate impoverished environmental input to form modal representations (e.g., Gregory, 1980). Ecological psychologist Gibson (1986) criticised this view for it implicitly models the perceiver as an immobile, passive observer. He argued that perception was not an approximate, mental process; rather, it was a physical activity. From this perspective, perceiving organisms have access to a clear,

DYNAMICS OF SYSTEMIC THINKING definitive, and direct identification—that is, an identification that is not mediated by a mental representation—of environmental stimuli, through the movements of their head and body as well as their actions. Moreover, the purpose of such a direct perception is not to build a passive catalog of stimulus categories. Its purpose is to identify what the environment may “afford”; that is, what opportunities or possibilities for action it offers (Withagen & Chemero, 2011). Proponents of the radical embodiment approach propose that Gibson’s account of perception is applicable to cognition in general, thus recasting cognition as emerging online, in real time, in situ, and from the dynamical and continuous interactions of coupled systems made of brain, body, and world. From this perspective, meaning and knowledge are not invented through mental processes, nor are they mediated by representational states. Instead, they are discovered through the coupling of the cognising agents’ activities and the affordances available in their environment (Turvey & Shaw, 1999; Wilson & Golonka, 2013). The assumption that knowledge and understanding arise from the coupling of a knowing agent and the environment within which this agent functions calls for a shift in focus from the study of brain activity as a causal determinant of cognition towards the study of the causal interconnections between the agent and its environment. Cognition is assumed to self-organise over time and emerge “live” from perception-action couplings, in a nonlinear, thoughtless, but deterministic way. Constrained by their own biological attributes, the characteristics of the tasks they face, and the features of the immediate environment in which they operate, cognising organisms are ultimately channeled into more or less stable attractor states of mind (Thelen,

9

DYNAMICS OF SYSTEMIC THINKING 1992). To understand cognition is, therefore, first to understand what resources are at the disposal of the knowing agent to solve a specific task. These resources may be located in the body, in the environment, and, eventually, in the brain. Second, to understand cognition is to understand how such resources can be assembled to create a dynamical system that evolves over time towards a solution (Wilson & Gonlonka, 2013). Numerous examples from robotics and animal cognition show that complex behaviours can arise from the coupling between agents and their environment without the need to use representations or computations as mediators (e.g., see Barrett, 2011). Rat pups, for example, are born blind, deaf and without fur. To survive, they need to stay close to each other to keep their body temperature at a suitable level. Despite their “disabilities,” they exhibit a complex and characteristic pattern of behaviours known as “thigmotaxic behaviours” and which include wall following, corner burrowing and huddling. May, Schank, Joshi, Tran, Taylor and Scott (2006) were able to reproduce these behaviours in robot rats without sensors but simply programmed to select randomly from a limited number of movements (stop, move forward in a number of direction or move backward in a couple of directions) every two seconds. The key feature of these roborats was that their body was morphologically similar to that of real rats, and importantly, featured a pointy nose. Once they encountered a wall, the moving robot-rats’ pointy nose caused them to slide alongside it until they reached a corner, at which point they became stuck. As more robot-rats joined in, they all ended up huddling in a corner. This example illustrates how complex pattern of behaviours may emerge without the need for mental “action plans”. None of

10

DYNAMICS OF SYSTEMIC THINKING the robot-rats intended to huddle. Instead, their behaviour emerged from the interaction between their inner mechanisms (i.e., the different built-in types of motions available to them), their body (i.e., the shape of their nose), and the characteristics of their immediate environment (i.e., featuring walls and corners). Cognition results from a temporally continuous flow of activity A second corollary of the conception of cognition as emerging in real time from the continuous interactions of coupled systems made of brain, body, and world concerns the manner in which cognition evolves in time. The Turing machine metaphor for cognition suggests that cognitive processes unfold in a formally-specifiable sequence of discrete events. The order of these events is assumed to be dictated by a stored algorithm controlling the sequence of steps required to transform an informational input into an output. From this perspective, change is predictable and timing is irrelevant. In contrast, timing becomes an essential feature of cognitive systems if one views cognition as emerging from the real-time coupling of brain and body activities in the environment. The coupling of brain, body and environment in a cognitive system results in simultaneous unfoldings. Brain activity is changing while the knowing agent acts in the world while the world is changing. Brain activity, the agent’s actions in the environment, and the resulting changes in the landscape are continuously dependent on each other; they co-evolve in a cyclical pattern of causation and coupling. The temporal dimension to the emergence of cognition spans from milliseconds to weeks, months up to millennia. At the millisecond scale, realtime cognition can be shown to evolve through continuous and dynamic

11

DYNAMICS OF SYSTEMIC THINKING changes rather than “as a staccato series of abstract computer-like symbols” (Spivey & Dale, 2006, p. 207). EEG studies of insight in remote-associate tasks do not reveal a discrete, sudden, ‘epiphanic’ Aha! moment, but rather a continuous pattern of activation including prior preparatory brain states (Kounios & Beeman, 2009). Speech perception (e.g., distinguishing between competing phonemes such as ‘ba’ and ‘pa’), spoken word recognition (e.g., recognising ‘candle’ instead of ‘candy’) and semantic categorisation (e.g., categorising ‘whale’ as a mammal rather than a fish) are all examples of cognitive processes that do not involve a discrete, stepwise, resolution; rather they involve a temporally continuous flow of neural activity that is first attracted to competing “gravitational basins” until some “resolution” occurs, leading to a stable mental state (Spivey & Dale, 2006). Beyond the millisecond span, similar patterns of temporally continuous behaviours can be observed in infants who learn to reach for hidden objects over time. The classic A-not-B error occurs when 10-months old infants persist in reaching for an old hiding place to retrieve an attractive object after having seen the object being hidden in a new location. Traditional computational accounts for this error propose that it arises as motor experience takes precedence over a conscious representational system where behaviour is controlled by rule-based algorithms (Marcovitch & Zelazo, 1999). This ‘error’ can be explained without resorting to a representational system. Thelen, Smith, Schoner and Scheier (2001) showed that the perseverance error was determined by multiple causes—such as the history of reaching for the original location, the attention-grabbing of the hiding event or of the cover, the delay between the hiding event and the reaching

12

DYNAMICS OF SYSTEMIC THINKING behaviour, or the approach to reach—all potentially interacting over nested timescales. The error is not a failure of motor control planning at the neuronal level; rather it is the product of the interaction, in real-time, between the infant body, its mind, and the activities afforded by its immediate environment. By the age of 12-months, most infants stop persevering in the classical task; however, changing the environment can make the error re-appear in twoyear-olds (Butler, Berthier, & Clifton 2002). This suggests that cognitive development is not simply the product of brain maturation; it is the emergent product of the continuous unfolding of infants’ activities in their environment, where each action sets the stage for the next (Smith & Thelen, 2003). So, here again, behaviour cannot be accounted for by a timeless, orderly, plan of actions. Instead, time is of the essence: to predict what will happen at time 1 in the brain-body-world system, you need to know where everyone and everything were at, at time 0. This is true for the trajectory of cognition at the millisecond timescale, at the lifespan timescale, and the argument can be made that this also account for the development of homo sapiens’ cognitive capacities over millennia (Malafouris, 2013). To summarise, recent views, ranging from the distributed cognition approach to the radical embodiment approach recast cognition as emerging, in real-time, from the interactions of a knowing organism, acting in, and reacting to its immediate environment. While the radical embodiment approach is arguing that cognition can do without representations, the type of behaviours this approach aims to explain remain basic (usually implicating action-perception loops to guide adaptive movement in simple environments). Whereas one may ponder at the remarkable (in)ability of super-computers to

13

DYNAMICS OF SYSTEMIC THINKING model simple behaviours, it remains that humans are capable of achieving more than those simple behaviours. Humans can play chess, draw inferences, solve problems and make decisions. In the next section, we discuss the challenges these types of activities raise for both the traditional computational approaches as well as the radical embodiment approach. The Human Brain is not (Only) an Executive Controller of Actions Representations are the undeniable substrate of some human cognitive activities The assumption that cognition emerges from a system that never represents the world may be tenable from the perspective of robotics or animal psychology; but it is largely untenable from the perspective of cognitive psychology—the branch of psychology that is concerned with how humans acquire, transform, and produce knowledge. Notwithstanding David Hume’s (1740/1983, as cited in Turvey & Shaw, 1999) touchstone, some cognitive activities are unique to humans and may require an explanatory framework that is inapplicable to other organisms. For a start, there is ample empirical evidence that people represent the meaning of what they have read or heard. For example, Bransford and Franks (1971) showed that, after having studied a set of sentences such as “the ants in the kitchen ate the jelly” (a) and “the ants ate the sweet jelly, which was on the table” (b), people would later be equally likely to recognise an old sentence such as (a) or a new sentence combining propositions from studied sentences, such as (c): “the ants ate the sweet jelly.” However, they would not recognise a sentence that contained new units of abstract meaning such as “the ants ate the jelly beside the woods.” Such findings strongly suggest people construct a mental

14

DYNAMICS OF SYSTEMIC THINKING representation of the gist meaning of the sentences they read and later rely on those representations to support their recall or recognition. Besides meaning, people also represent and solve problems. Two broad types of problems have been subjected to laboratory scrutiny by psychologists: (i) analytic or transformation problems and (ii) insight problems. Transformation problems are knowledge lean and well-defined problems, presenting an initial situation or start state, and a final situation or goal state. To solve transformation problems, people are assumed to represent the problem’s start state and engage in a step-by-step mental transformation of this representation until they reach the problem’s goal state. Such transformations occur by applying operators, defined as more or less constrained means of changing one state into another. Simple transformation problems such as the Tower of London have been studied in the lab, primarily to observe, or in neuropsychological cases to diagnose, participants’ planning skills and move selection decisions. Insight problems, although they too tend to be knowledge lean, are less easily characterized although all involve an initial inability to envisage the operators one could apply to achieve the goal state. It is this initial impasse that must be overcome for a solution to be conceived. Insight problems interest psychologists largely because of the ‘Aha’ phenomenology associated with the sudden eureka clarity with which reasoners envisage the solution. Take the following problem: How do you throw a ping pong ball in such a manner that it comes to a complete stop and reverses direction without coming into contact with anything? An initial representation of the problem involving a horizontal throwing motion will encourage possibly creative but futile solutions. However, representing a

15

DYNAMICS OF SYSTEMIC THINKING vertical throwing motion quickly ushers in a satisfying resolution to this ‘problem’. What interest psychologists are the processes by which the initial representation of the problem is restructured to yield a more productive perspective from which the solution can be conceived easily. Once again, the fact that people can solve both classes of problems, simply from reading a problem statement written on a piece of paper (e.g., Metcalfe & Weibe, 1987) suggests that people construct mental representations of problem states and rely on those representations to support their problem solving activities. Representations of meaning and problem states are two among many more examples of higher cognitive activities. Similar arguments can be made with regards to other higher cognitive activities such as mathematical reasoning (e.g., Ashcraft & Battaglia, 1978), analogical reasoning (e.g., Gick & Holyoak, 1983), deductive and inductive reasoning (Johnson-Laird, 1983), or judgement and decision-making (Villejoubert & Mandel, 2002). The fact is that humans are able to think off-line. Put a hungry human in a new maze (e.g., a new city). Typically, he may hunt for a grocery store or restaurant randomly walking along city blocks. He might remember his way around initially through trials and errors but will soon develop a cognitive map of his surroundings. Early orientation behaviours may be closely coupled to perceptual inputs, but soon he will be able to plan his whereabouts or figure out ways to overcome changes in his environment off-line; that is, in the absence of concurrent dynamic interaction. This, as Clark (1997) aptly put it, points to “the difference between inner systems that operate only so as to control immediate environmental interactions and ones that use inner resources to model the world so as to obviate the need for such continual

16

DYNAMICS OF SYSTEMIC THINKING environmental interaction.” (p. 464). In other words, cognitive psychology, and more particularly higher cognitive functions, can hardly be conceived as unfolding solely from on-line perception-action couplings, as proponents of the radical embodiment approach sometimes argue. This raise the question of the part, if any, played by on-line perception-action couplings in higher cognition. We address this issue next. Is there a case for an objectivist account of cognition? Cognitive psychologists have traditionally embraced Fodor’s (1980) “methodological solipsism”, assuming higher cognitive functions take place inside the head, squeezed between perceptual inputs and behavioural outputs. Those who adopt this conception of cognition effectively ostracise the world outside the thinking head, and aim to characterise higher thinking through relatively static mental states and structures. From this perspective, perception-action couplings are not constitutive part of cognition, they are incidental by-products, implementation spandrels, of internal computation. The fact that something that is happening in the environment has some causal influence on cognitive activity does not warrant the conclusion, intracranialists argue, that such an external event is constitutive of cognition. From this perspective, tool use, for instance, is “typically a matter of cognitive processes interacting with portion of the non-cognitive environment” (Adams & Aizawa, 2009, p. 80). Moreover, the brain constitutes a natural kind and, consequently, its workings can be assumed to be bound upon by natural laws and its functions and processes are multiple but they are homogeneous because they all emerge from the same (neurological) substrate. These facts make a scientific account of brain cognition possible. By contrast, goes the

17

DYNAMICS OF SYSTEMIC THINKING argument, tools are not natural; they are artefactual and disparate. The processes involved in thinking with tools are not homogeneous, they are varied and indefinable; therefore, there cannot be a scientific account of braintool cognition that would lead to the discovery of general laws of behaviour. This position, it seems to us, underscores the deficiency of the current research methodology to capture how cognition may emerge from brain, body, and environment interactions rather than a genuine theoretical and ontological impasse. There are many examples throughout history where human understanding of natural phenomena shifted radically through methodological breakthroughs and the accumulation of empirical evidence. The brain itself was once considered such a trivial organ that it was simply removed in the preparation of bodies for mummification in ancient Egypt or conceived as a mere cooler for the heart’s passions in Aristotle writings (Finger, 1994). Admittedly facile, this diachronic contrast nevertheless illustrates how our current understanding of the workings of the brain was made possible once scholars believed in the importance of its study, developed and honed methods to study it, and started to accumulate empirical findings. To conclude a priori that the external props and aids are motley and irrelevant to cognition is unwarranted and risks missing a proverbial forest. Until an appropriate methodology is developed for studying how cognition may emerge from the dynamic configuration of extended cognitive systems, and until enough empirical evidence is accumulated, we are bound to see motley. Motley may seemingly preclude a scientific account of how transactions between brains, bodies and environments contribute to the

18

DYNAMICS OF SYSTEMIC THINKING emergence of higher cognition, but it does not necessarily entail that interactivity is devoid of meaningful structures, organizing patterns, and regularity. Only the empirical and scientific study of those systems can provide a meaningful answer to this question. This new programme of research may begin with a more qualitative and ethnographic research methodology (see Steffensen, 2013, for an excellent example) but could also borrow from established methods of applied behavior analysis in germane fields of enquiries such as developmental psychology, educational psychology and ethology (Bakeman & Quera, 2011; Sharpe & Koperwas, 2003). Such approaches are well poised to offer generalizable conjectures about how new ideas, understanding, and solutions emerge from an ontological substrate woven by interactivity. Note that this is not to say that we should treat artefact materiality as inconsequential: the evolution and transformation of an approximate number sense to exact counting and arithmetic in humans over millennia of cultural evolution is inextricably linked to the development and manufacturing of a wide range of artefacts—clay tokens, so-called ‘envelopes’, impressed tablets, and pictographic tablets (Malafouris, 2013). On the contrary, this is a call to embrace an epistemological approach and an empirical commitment to document how cognitive systems work. A related argument put forward against the proposition that external props and aids are constitutive of human cognitive processing is the fleeting nature of systems where brain and body interact with their immediate environment (Rupert, 2009). Cognitive psychology’s aim, admittedly, is to provide a nomothetic account of human cognitive processes through the

19

DYNAMICS OF SYSTEMIC THINKING discovery of general scientific laws that underpin human cognitive activities such as categorisation, memory, judgement and problem-solving, across contexts and situations. By contrast, situated systems where brain, body and environment interact seemingly epitomise the idiographic particular, with its essentially transient and concrete interactions bounded to a specific space and time. So in the face of the motley of environments, artefacts, and localised interaction, the brain appears as the only constant and thus the only suitable unit for studying human cognition. This argument, however, is fallacious. The complex patterns of neural cognitive activity unfolding inside one individual brain, at any point in space and time, are as particular, idiosyncratic and seemingly unruly as the motor activities unfolding outside the brain, at this very moment, in this very situation. Still, neuroscientists can formulate function-to-structure deductive and structure-to-function inductive inferences (Henson, 2005) despite heterogeneous patterns of brain activity. So, once again, the fleetness argument does not hold up as an argument against the possibility of formulating generalizable inductive inferences about the spatio-temporal dynamic regularities of systemic cognition. To summarise, the fact that individuals can think off-line does not take away the fact that they do not typically do so (Clark, 2010). People think everyday and yet, undeniably rarely come up with solutions to concrete or abstract problems through long streaks of off-line thinking. Instead, most of human thinking emerges online, through physical and verbal interactions with the immediate environment. For example, real-world problem solving is enacted through interactivity with people and things in a physical environment (Cowley & Nash, 2013; Steffensen, 2013). Scientists use artefacts and

20

DYNAMICS OF SYSTEMIC THINKING models to help them formulate and test hypotheses (Giere & Moffat, 2003; Watson, 1968), and people naturally recruit artefacts and tailor their immediate physical environments to augment and transform their problem solving activities. Once one recognises that cognition typically unfolds in a brain-body-environment system, the question of whether artefacts form a constitutive part of human cognition—e.g., to paraphrase Clark and Chalmers (1998), whether Otto’s notebook is included in his mind—or the question of whether or not cognitive processing only takes place inside brains both become peripheral to the concerns of the systemic cognitivist. The important issue is no longer where cognitive processing begins and where it ends. The important issue becomes how cognition emerges from the interactions of brain activity, motor actions, and artefacts. Systemic cognitivists do not question whether there can be a scientific study of real-time cognition. Actions are not unbounded: body and environment constraint the interactive trajectory. To adopt a systemic approach to cognition is to reject the a priori assumption that there are no general laws that can account for how cognition emerges in situ, and embark on a scientific journey of discovery, documentation and ultimately theoretical understanding of untapped psychological phenomena. The Spatio-temporal Dynamics of Systemic Thinking Interactivity plays a constitutive role in thinking (and in shaping the world that enacts it) As it disregard actions in the world as merely peripheral, traditional cognitive psychology has generally been reluctant to study thinking in problem-solving environments that permit and encourage interactivity (Vallée-

21

DYNAMICS OF SYSTEMIC THINKING Tourangeau & Villejoubert, 2013). Once that reluctance is shed, and modest efforts go toward implementing interactive thinking environments in the lab, there is much evidence that interactivity substantially transforms problem solving (Vallée-Tourangeau, 2013; Vallée-Tourangeau, Euden, & Hearns, 2011). Thus, for example, we observed that interactivity promoted a significantly higher rate of insight solutions in problems for which a false algebraic expression with Roman numerals can be transformed into a true one by moving a single feature of either a numeral or an operator (e.g., how to transform I = II + II into a true expression). Weller, Villejoubert and ValléeTourangeau (2011) demonstrated that participants were more likely to discover the solution (e.g., I = III – II) if the expressions themselves were presented as manipulable three dimensional objects—and participants were invited to manipulate them— than if they were presented on a piece of paper and participants announced their solution to the experimenter. Interactivity matters because action-perception cycles quickly unveil a range of potential solutions that help participants home in on the right one. Participants may experience a form of impasse, but the opportunity to manipulate the physical elements of the problem evinces a fluid set of action affordances, which encourage participants to tinker and continue their exploration. In turn, when participants are confronted with an unchanging algebraic expression, the perceptual feedback exerts a strong conceptual pull back to the false formulation: The participants must overcome this perceptual information by simulating the transformations mentally. To be sure, the problem solving exercise exacts a greater toll on working memory in a static than in an interactive condition. But interactivity does not simply augment working

22

DYNAMICS OF SYSTEMIC THINKING memory resources by letting the environment shoulder some of the computational burden; this indeed “hugely downplays what is going on” (Kirsh, 2013, p. 178). Interactivity transforms the terrain of cognition: the thinking that emerges from the intertwining of a reasoner’s internal resources and the external artefacts is qualitatively different. Performance on mental arithmetic problems provides another example of how interactivity transforms thinking. Asking college-age students to add a long series of single-digit numbers, without using pen and paper, does not exceed their acquired arithmetic skills. However, a long series of numbers arrayed in a random configuration puts a significant burden on working memory—participants must remember the interim total, what numbers were added, which one should be added next—and counting inaccuracies are likely to arise. Of greater interest, is the strategy employed in this situation: Participants are likely to adopt a simple strategy that conserves working memory resources, scanning single numbers in turn, keeping track of the running total. This slow and methodical strategy underexploits stored knowledge of simple sums which further handicaps the ability to identify congenial groupings and subtotal to enhance the speed and accuracy of the process. The same sum presented in terms of movable number tokens that participants are encouraged to move about will give rise to substantially different counting behavior. As Vallée-Tourangeau (2013) demonstrated, interactivity fosters more creative and efficient paths to solution in mental arithmetic: participants are less likely to adopt a conservative counting strategy, and better able to identify groupings that facilitate the creation of congenial interim totals (e.g., such as those that divide by 5 without a

23

DYNAMICS OF SYSTEMIC THINKING reminder). Interactivity thus enhances the robustness and efficiency of the ‘mental’ calculation process. Interactivity provides additional working memory resources, but the transformative impact on thinking cannot be predicted by the simple addition of external storage. Thinking emerges from transactions between online and offline cognition Classical accounts of thinking put too great an emphasis on the role of internal representations and mental processing. This is not to say that the impact of the environment has not been featured in such accounts, however. Core cognitive processes have often assumed to result from both inner processes and the situation within which people are embedded while they think (e.g., Neisser, 1967). The world, however, is assumed to be mirrored (more or less accurately) in people’s heads and the important question becomes “how an exploitative relationship between mind and environment has implication for the kind of cognitive machinery used by the mind” (Brighton & Todd, 2001, p. 324). The answer, according to proponents of the ecological rationality approach is that adaptive behaviour emerges when there is a match between information structure in the mind and in the world. Thus, this conception of cognition remain subordinate to the doctrine of animalenvironment dualism (Turvey & Shaw, 1999). It conceives the mind and the environment as disjointed entities and cognition as emerging from a “selfactional” brain, which has evolved its computational mechanisms in symbiosis with nature, and as a result may be more efficient in computing “natural” information (e.g., frequency information rather than probabilistic information; Gigerenzer & Hoffrage, 1995) but nevertheless computes information in a

24

DYNAMICS OF SYSTEMIC THINKING timeless manner, and independently of its immediate environment. This can be contrasted with a situated and interactional account of cognition, where the immediate environment has a direct impact on what is processed by the brain and how it is processed as well as where, conversely, what is processed by the brain can have a direct impact on the immediate environment (e.g., where the immediate environment is transformed the instantiation of a plan of action initially elaborated in the brain). Such an interactional account of cognition, however, is still limited by a mechanical conception of brain and environment as having independent influences on cognition. Intracranialists will readily grant the importance of the environment in shaping cognition and behavior: “the apparent complexity of our behavior (!) is largely a reflection of the complexity of the environment in which we find ourselves (Simon, 1969, p. 53). But the interaction between two independent entities can be parsed in terms of input-output segments, betraying an insidious dualism. This research programme misses the mark (see also Villejoubert & Vallée-Tourangeau, 2011). The dynamic context in which cognition and behavior emerge reflects transactional forces that coincidentally and reciprocally shape cognition and the world that enacts it. In our laboratory work on problem solving, the proto representations of a solution and the world that in turn reflects it, as well as offering the means to project its potential development, both co-evolve in space and time to a more complex representation of the solution. The only research methodology that can capture the genesis of insight in problem solving research is one that examines the thoughts and behaviours of an agent in real time in a dynamic and malleable physical environment. As such, Steffensen’s (2013) cognitive event analysis alongside other methods of

25

DYNAMICS OF SYSTEMIC THINKING applied behavior analysis offers a promising compass to guide these future efforts. Concluding Remarks We draw considerable inspiration and guidance from a more radical form of embodied cognition. However, the associated philosophical commitments that jettison mental representations in thinking require a herculean dismissal of voluminous evidence to the contrary. We also argue that traditional research methods in cognitive psychology reflect a dualism that keeps mind and environment in distinct ontological categories. Classical accounts of thinking put to great an emphasis on internal representation and mental processes and ignore the symbiotic relationship between thinking and acting. A more fruitful research methodology will map the spatio-temporal coordinates of the co-constitutive processes that enact, co-temporaneously, the mind and the world to which it is connected. Thinking emerges in ecological space and ecological time from the transactional flow of action and representational opportunities outcropping from a dynamic action-environment interface.

26

DYNAMICS OF SYSTEMIC THINKING References Adams, F., & Aizawa, K. (2009). Why the mind is still in the head. In P. Robbins and M. Aydede (Eds.), The Cambridge handbook of situated cognition (pp. 78-95). New York: Cambridge University Press. Bakeman, R., & Quera, V. (2011). Sequential analysis and observational methods for the behavioral sciences. New York: Cambridge University Press. Barrett, L. (2011). Beyond the brain: How body and environment shape animal and human minds. Woodstock, Oxfordshire: Princeton University Press. Bransford, J. D., & Franks, J. J. (1971). The abstraction of linguistic ideas. Cognitive Psychology, 2, 331-360. Brighton, H., & Todd, P. M. (2009). Situating rationality: Ecologically rational decision making with simple heuristics. In P. Robbins and M. Aydede (Eds.), The Cambridge handbook of situated cognition (pp. 322-346). New York: Cambridge University Press. Butler, S.C., Berthier, N. E., Clifton, R. K. (2002). Two-year olds’ search strategies and visual tracking in a hidden displacement task. Developmental Psychology, 38, 581-590. Chemero, A. (2009). Radical embodied cognitive science. London: The MIT Press. Clark, A. (1997). The dynamical challenge. Cognitive Science, 21, 461-481. Clark, A. (2008). Supersizing the mind: Embodiment, action, and cognitive extension. New York: Oxford University Press.

27

DYNAMICS OF SYSTEMIC THINKING Clark, A. (2010). Material surrogacy and the supernatural: Reflections on the role of artefacts in ‘off-line’ cognition. In L. Malafouris and C. Renfrew (Eds.), The cognitive life of things: Recasting the boundaries of the mind (pp. 23-28). Cambridge: McDonald Institute for Archaeological Research. Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 58, 7-19. Cowley, S. J., & Nash, L. (2013). Language, interactivity and solution probing: Repetition without repetition. Adaptive Behavior, 21, 187-198. Ashcraft, M.H., & Battaglia, J. (1978). Cognitive arithmetic: Evidence for retrieval and decision processes in mental addition. Journal of Experimental Psychology: Human Learning and Memory, 4, 527–538. Finger, S. (1994). Origins of neuroscience: A history of exploration into brain function. New York: Oxford University Press. Fodor, J. A. (1980). Methodological solipsism considered as a research strategy in cognitive psychology. Behavioral and Brain Sciences, 3, 63109. Fodor, J. A. (1981). The Mind-Body Problem. Scientific American, 244, 114– 123. doi:10.1038/scientificamerican0181-114 Gibson, J. J. (1986). The ecological approach to visual perception. Hove, East Sussex: The Psychology Press. Gick, M. L., & Holyoak, K. J. (1983). Schema induction and analogical transfer. Cognitive Psychology, 15, 1-38. Giere, R. N., & Moffatt, B. (2003). Distributed cognition: Where the cognitive and the social merge. Social Studies of Science, 33, 301-310.

28

DYNAMICS OF SYSTEMIC THINKING Gigerenzer, G., & Hoffrage, U. (1995). How to improve Bayesian reasoning without instruction: Frequency formats. Psychological Review, 102, 684704. Gregory, R. L. (1980). Perceptions as hypotheses. Philosophical Transactions of the Royal Society of London, B, 290, 181-197. Henson, R. (2005). What can functional neuroimaging tell the experimental psychologist? Quarterly Journal of Experimental Psychology, 58A, 193233. Hollan, J., Hutchins, E., & Kirsh, D. (2000). Distributed Cognition: Toward a new foundation for human-computer interaction research. ACM Transactions on Computer-Human Interaction, 7, 174-196. Hume, D. (1739/1983). A treatise of human nature. Oxford: Clarendon Press. Hutchins, E. (1995). Cognition in the wild. London: The MIT Press. Johnson-Laird, P. N. (1983). Mental models. Cambridge, MA: Harvard University Press. Kahneman, D. (2003). A perspective on judgment and choice: mapping bounded rationality. American Psychologist, 58, 697–720. doi:10.1037/0003-066X.58.9.697 Kirsh, D. (1995). The intelligence use of space. Artificial Intelligence, 73, 3168. Kirsh, D. (2013). Thinking with external representation. In S. J. Cowley, & F. Vallée-Tourangeau (Eds.), Cognition beyond the brain: Computation, interactivity and human artifice (pp. 171-194). London: Springer-Verlag. Kirsh, D., & Maglio, P. (1994). On distinguishing epistemic from pragmatic action. Cognitive Science, 18, 513-549.

29

DYNAMICS OF SYSTEMIC THINKING Kounios, J., & Beeman, M. (2009). The Aha! Moment: The cognitive neuroscience of insight. Current Directions in Psychological Science, 18, 210-216. Malafouris, L. (2013). How things shape the mind: A theory of material engagement. Cambridge, MA: The MIT Press. Marcovitch, S., & Zelazo, P. D. (1999). The A-Not-B error: Results from a logistic meta-analysis. Child Development, 70, 1297-1313. May, C. J., Schank, J. C., Joshi, S., Tran, J., Taylor, R. J., & Scott, I.-E. (2006). Rat pups and random robots generate similar self-organized and intentional behavior. Complexity, 12, 53-66. Metalfe, J., & Wiebe, D. (1987). Intuition in insight and noninsight problem solving. Memory & Cognition, 15, 238-246. Neisser, U. (1967). Cognitive psychology. East Norwalk, CT: AppletonCentury-Crofts. Rupert, R. D. (2009). Cognitive systems and the extended mind. New York: Oxford University Press. Shapiro, L. (2011). Embodied cognition. New York: Routledge. Sharpe, T., & Koperwas, J. (2003). Behavior and sequential analyses. Thousand Oaks, California: Sage Publications. Simon, H. A. (1969). The sciences of the artificial. Cambridge MA: The MIT Press Smith, L. B., & Thelen, E. (2003). Development as a dynamic system. TRENDS in Cognitive Sciences, 7, 343-348 Spivey, M. J., & Dale, R. (2006). Continuous dynamics in real-time cognition. Current Directions in Psychological Science, 15 , 207-211.

30

DYNAMICS OF SYSTEMIC THINKING Steffensen, S. V. (2013). Human interactivity: Problem-solving, solution probing and verbal patterns in the wild. In S. J. Cowley, & F. ValléeTourangeau (Eds.), Cognition beyond the brain: Computation, interactivity and human artifice (pp. 195-221). London: Springer-Verlag. Thelen, E. (1992). Development as a dynamic system. Current Directions in Psychological Science, 1, 189-193. Thelen, E., Schöner, G., Scheier, C., & Smith, L. B. (2001). The dynamics of embodiment: A field theory of infant perseverative reaching. Behavioral and Brain Sciences, 24, 1-86. Turing, A. M. (1936). On computable numbers, with an application to the Entscheidungsproblem. Proceedings of the London Mathematical Society, 42, 230–265. Turvey, M. T., & Shaw, R. E. (1999). Ecological foundations of cognition: I Symmetry and specificity of animal-environment systems. Journal of Consciousness Studies, 6 (11-12), 95-110. Vallée-Tourangeau, F. (2013). Interactivity, efficiency, and individual differences in mental arithmetic. Experimental Psychology, 60, 302-311. Vallée-Tourangeau, F., Euden, G., & Hearn, V. (2011). Einstellung defused: Interactivity and mental set. Quarterly Journal of Experimental Psychology, 64, 1889-1895. Vallée-Tourangeau, F. & Villejoubert, G. (2013). Naturalising problem solving. In S. J. Cowley, & F. Vallée-Tourangeau (Eds.), Cognition beyond the brain: Interactivity, cognition and human artifice (pp. 241-254). Dordrecht: Springer.

31

DYNAMICS OF SYSTEMIC THINKING Vallée-Tourangeau, F., & Wrightman, M. (2010). Interactive skills and individual differences in a word production task. AI & Society, 25, 433439. Van Gelder, T. (1995). What might cognition be, if not computation? Journal of Philosophy, 92, 345-381. Villejoubert, G., & Vallée-Tourangeau, F. (2011). Constructing preferences in the physical world: A distributed-cognition perspective on preferences and risky choices. Frontiers in Psychology, 2. doi:10.3389/fpsyg.2011.00302 Villejoubert, G., & Mandel, D. R. (2002). The inverse fallacy: An account of deviations from Bayes’s theorem and the additivity principle. Memory & Cognition, 30, 171–178. doi:10.3758/BF03195278 Watson, J. D. (1968). The double helix. London: Penguin. Weller, A., Villejoubert, G., Vallée-Tourangeau, F. (2011). Interactive insight problem solving. Thinking & Reasoning, 17, 429-439. Wilson, A.D., & Golonka, S (2013). Embodied cognition is not what you think it is. Frontiers in Psychology, 4, 58. Withagen, R., & Chemero, A. (2012). Affordances and classification: On the significance of a sidebar in James Gibson’s last book. Philosophical Psychology, 25, 521-537.

32

The Context-Dependent Nature of Action Knowledge
Nicholas J. Shipp (n.j.shipp@herts.ac.uk)
Department of Psychology, University of Hertfordshire Hatfield, AL10 9AB, UNITED KINGDOM

Frédéric Vallée-Tourangeau (f.vallee-tourangeau@kingston.ac.uk)
Department of Psychology, Kingston University, Kingston upon Thames, KT1 2EE, UNITED KINGDOM

Susan H. Anthony (s.h.1.anthony@herts.ac.uk)
Department of Psychology, University of Hertfordshire Hatfield, AL10 9AB, UNITED KINGDOM Abstract
Recent theories of semantic memory have proposed that concepts are grounded in sensorimotor activity and mediated by the context from which the knowledge is drawn (Barsalou, 1999, 2003, 2008). Conceptual knowledge draws upon information from all modalities and therefore includes knowledge of associated object actions linked with both function and general movement (Bub, Masson, & Cree, 2008). The following experiment examined the conditions under which action information exerts an influence on experimental tasks particularly when taxonomic information is present. The experiment used a forced-choice triad task giving participants the choice of selecting between items that shared either a taxonomic or an action based relation with the target. The results showed that when the objects were presented as images on a white background (context-lean condition), participants were more likely to select the taxonomically related item. In contrast, when the same triads were presented as images being used in a functional scene (context-rich condition) they were more likely to select the action-related item. The results show that action knowledge is not automatic but is context-dependent. In line with views on embodied semantics the motor cortex is activated and drawn upon when objects are viewed and this influences task performance despite being unnecessary for the task. Keywords: Action; Triads; Categorisation; Context; Embodied Semantics.

Introduction
The traditional view of conceptual knowledge suggested that concepts are formed from amodal abstractions of the context in which they were previously encountered (Barsalou, 1999; 2003; Yeh & Barsalou, 2006). However an increasing number of researchers have shown that concepts are embodied within sensorimotor activity (Barsalou, 1999, 2008; Wu & Barsalou, 2009; Yeh & Barsalou, 2006). Embodied semantics takes the view that concepts are much more than decontextualised feature lists and that they reside within the same sensory-motor circuits in which they were first established (Aziz-Zedah & Damasio, 2008; Fernandino & Iacoboni, 2010). According to Barsalou (1999, 2003, 2008) the conceptual system does not record the images seen of an entity, but registers the concomitant neural experience. When encountering or re-instantiating the concept at a later date the conceptual system partially reactivates those neuronal

patterns, thereby producing a simulation of the experience. Areas of the motor cortex that are active upon the initial object encounter will be reactivated and as such common actions associated with objects should readily come to mind when thinking of an object. Empirical work over the last decade has supported this view showing that semantic knowledge is embedded within physical actions which influences performance across a variety of cognitive tasks (Anelli, Nicoletti, & Borghi, 2010; Borghi, 2004; Borghi, Flumini, Natraj, & Wheaten, 2012; Bub & Masson, 2006; Bub, Masson, & Bukach, 2003; Chao & Martin, 2000; Creem & Proffitt, 2001; Iachini, Borghi, & Senese, 2008; Tipper, Paul, & Hayes, 2006; Tucker & Ellis, 1998, 2004; Vanio, Symes, Ellis, Tucker, & Ottoboni, 2008). What has been particularly evident from the research is that action can play a role even in tasks where knowledge of associated action is neither required nor asked for. Borghi (2004) used a property generation task to show that when participants are asked to simply think about an object such as a car the first parts of the car that they name are those related to direct human interaction—e.g., the gearstick, steering wheel. The same parts were named first when participants were given a direct context to think about such as building the object. This would be in line with the view that thinking about the car activated the motor cortex and as such direct interaction became an influential feature in this task. Helbig, Graf and Keifer (2006) showed how actions are drawn upon in object recognition using a priming task. In their experiment participants were asked to name both a prime and a target object and showed that participants were more accurate in naming the target object when the prime was congruent in its action manipulation. Here the prime had a clear effect of activating the motor system and its relevant action knowledge that remained active for the target and hence identification was quicker. Helbig, Steinwender, Graf and Keifer (2010) used video primes of an agent performing an action on an object that was blacked out. Following the prime participants then saw an image of an object followed by a word, they were asked to identify if the word matched the object. Participants were again more accurate in their responses when the action seen in the prime matched the action of the following object. Further priming studies have shown similar action-based effects (Bub & Masson, 2012; Vanio et al., 2008).

Research has also shown that when participants are asked to make an action-based response they are typically faster and more accurate when action-based knowledge is taken into account. Jax and Buxbaum (2010) presented participants with objects that they termed as being either conflict or non-conflict. Non-conflict objects, they suggest, are objects that require the same action to both use and transport them (e.g., drinking glass). In contrast when the action of use is different from the action of transport (e.g., as for a calculator) the object is referred to as a conflict object. When participants were asked to place their hand on the object as though they would either use it or hand it to another person, they were faster for the nonconflict items over the conflict items. In addition they also found that generally participants were faster at making hand placements to use the objects rather than to hand them to another person. Osiurak, Roche, Ramone and Chainay (2013) showed the reverse effect when participants were physically asked to either pick up the objects in order to hit a ping pong ball or pass it to the experimenter. Here participants were faster at making transport actions over use. The authors attribute this to additional information being activated such as weight and solidity, which would not be activated if participants are simply asked to place their hand on the objects. Yee, Chrysikou, Hoffman and Thompson-Schill (2013) found that performing concurrent actions along with a semantic decision task slowed down task performance. Their participants undertook a semantic decision task judging if spoken words were abstract or concrete while concurrently performing a three-step manual ‘patty-cake’ task. This involved participants placing two fingers, four fingers or the whole hand on its side onto a table top in a repeating manner while engaging in the semantic decision task. Yee et al. found a significant interference effect of performing such manual actions with slower decisions made when compared to participants who performed no concurrent task. The patty-cake task was found to have an increased interference effect on those objects that participants had greater levels of previous handling experience. No interference was found when participants performed a mental rotation task.

upon across different contexts. In supporting such findings Borghi et al. (2012) showed that activation of object affordances was contextually dependent. Participants were shown object pairs that were related either functionally (paper + scissors), spatially (stapler + scissors) or had no relation (bottle + scissors). In addition to this the pairs were also shown with either a hand with a functional grasp on one of the objects, a manipulative grasp, a hand present but not holding either item or with no hand present. The participants were quicker and more accurate at making decisions on the objects being related or unrelated when object pairs shared a functional context. Participants were also faster when pairs were presented with a functional hand grasp rather than a manipulative grasp, however overall responses were faster when the images were presented with no hand. Items that share the same function also share a related goal. As such it is possible that participants drew upon the related goal of the items that in turn decreased the reaction time between them. Since manipulative pairs share only a spatial context and no related goal they were slower than the functional pairs.

The Present Experiment
The first aim of the current experiment was to explore the role that action plays in category formation when it is presented in conjunction with category membership. It should be noted that for the purposes of this experiment action is defined as the direct interface between objects and the human body. For example the action of a rifle would be the grasp made by the hand around the handle, rather than action in terms of the act that can be carried out using a rifle. For example an orange and a banana both require a peeling action and also belong to the same category of fruit. In addition many items can share an action without sharing category membership such as a rifle and a water pistol which both require a grasp of the handle in the same fashion but are not both weapons. Therefore a task was designed in which category membership could be directly pitted against the action of interfacing with an object when using it in its functional capacity. Using this method it could be tested if participants would always choose category membership or if action knowledge would be drawn upon in line with recent views on embodied semantics. The experiment used a forced-choice triad task. This basic task has been used to demonstrate the influence of situational (thematic) information (Lin & Murphy, 2001; Murphy, 2001) when previously only taxonomic (shared property) information would have been predicted to guide choices. In Lin and Murphy’s (2001) studies, the choice items were selected to share either a taxonomic or a thematic relation to the target. For example the target bee was presented with wasp (taxonomic similarity) or honey (thematic similarity). In a similar manner to Lin and Murphy, participants in the present experiment were presented with a target and two choice options, only one of which shared an action with the target. Based on previous research showing the strong role of action knowledge in a range of tasks based on category knowledge, we predicted that participants would be more likely to select the actionrelated item over the taxonomic choice only when it too

The Influence of Context
Research has shown that the context in which information is presented influences what type of conceptual knowledge becomes activated. Barsalou (1982) showed that while certain types of information are contextually independent and activated irrespective of the context other types of information can be contextually dependent and activated under certain circumstances. Barsalou showed that in a property verification task participants were quicker to verify that a basketball can float when given a context requiring someone to use one as a floatation device compared to verifying whether a basketball can bounce or not. Therefore such information would be contextually dependent as it does not always come straight to mind for this task but varies according to context. In contrast participants showed no difference in verifying sentences regarding how skunks have an unpleasant smell across different conditions. This would be regarded as contextually independent being drawn

Figure 1: Examples of stimuli employed in the experiment. From left to right: Same Category Object triad, Different Category Object triad, and Perceptual Category Object triad in the context-lean condition (top panels) and in the contextrich condition (bottom panels). shared a taxonomic relation to the target. When the action-related item bore no taxonomic relation to target, we predicted that participants would be more likely to select the taxonomic choice. The second aim of the experiment was to investigate the nature of action information and whether its activation is contingent on the context of categorisation. The triads shown were manipulated between subjects based on context. Participants either saw the objects on a white background (context-lean condition) or shown within an action based scenario with the objects being used by an agent (context-rich condition). It was predicted that in line with Borghi et al. (2012) the action choice within the triads would be selected more often when the images are shown within a functional context. shares a motor action with orange. In the DCO triads, one choice item shared category membership with the target but not an action (rifle and sword). The remaining choice item shared a motor action with the target but not category membership (rifle and water pistol). In order to test the effect of context two sets of images were collected. The first set showed the objects against a white background (context-lean condition). The second set projected the objects being used in a functional context (context-rich condition, see Fig. 1). Twenty participants not used in the experiment took part in pilot work to ensure that the SCO and DCO triads were matched in terms of category membership and the action used to functionally interact with them. Fifteen of each triad set were initially designed and piloted. Using a Chronbach’s alpha level of .7 as a threshold criterion, the final sets of SCO and DCO items were composed 10 triads of each type. A third set of triads (PCO) was designed based on the results of experimental pilot work. It seemed possible that participants might select the choice item sharing an action not because they shared an action, but because they shared perceptual properties. For example pencils and paintbrushes share perceptual properties, in part as a function of the ergonomic constraints that guide their design. Using the triads described above it is not possible to ascertain whether such items are selected because of action or because they look the same. In the PCO triads neither of the choice items shared category membership with the target item. One of the choice items shared an action with the target but few perceptual features (nut and car key). The remaining choice item shared perceptual features with the target but not an action (nut and money). The PCO triads were again presented in the same contextlean/context-rich manner as the SCO and DCO triads (see Fig. 1). Twenty participants not used in the experiment took part in pilot work to ensure that the PCO triads were matched in terms of perceptually relevant features and the

Method
Participants
Fifty undergraduate students (36 females) from the University of Hertfordshire participated in return for course credit with a mean age of 25.3 years (SD = 7.9, Range = 18-49).

Materials
The triads were based on the standard design of a target item followed by two items from which a choice could be made. Since the aim of the research was to compare the effect of action knowledge both alongside and set against taxonomic information, two sets of triads were initially designed, namely same-category object (SCO) triads and different-category object (DCO) triads (see Fig. 1). In the SCO triads participants saw a target (e.g., orange) and two choice items (e.g., banana and strawberry) which all belonged to the same category (fruit) as confirmed by pilot work. Of these choice items both share category membership with the target and in addition banana also

action used to functionally interact with them. Fifteen PCO triads were initially designed and piloted. Using a Chronbach’s alpha level of .7 as a threshold, a final set of 10 PCO triads were selected. Thus the experimental material was composed of 30 triads consisting of 10 SCO, DCO and PCO triads.

triads than in the DCO triads (p = .001). The interaction between triad type and context was not significant, F (2, 96) = 2.33, p = .10, !2 = .05. ! ! Context Lean Context Rich
80%

Percentage of Action Choice

Procedure
The experiment employed a 3x2 mixed design where triad type was a repeated measure and context was a between subjects factor. The main dependent measure was the percentage of action choices calculated for each set of triad type in both context conditions. Stimuli and task instructions were presented on a 15” Macintosh laptop. Participants were instructed to “Please indicate which of the two items goes best with the item at the top of the screen”, developed from Lin and Murphy (2001). A fixation cue was presented on the screen for 1000ms after which the cue disappeared and the target word appeared along with the appropriate picture depending on which condition the participant was assigned to. After 1500ms the two choice options appeared beneath the target alongside the appropriate images. The triad remained on the screen while participants made their choice. Participants were instructed to press the ‘a’ key to choose the item on the left-hand side of the screen and the ‘l’ key for the item on the right-hand side of the screen. The choice items were counterbalanced across the triads so that in half the triads the action choice appeared on the left hand side while in the remaining half the action choice appeared on the right. After they had made their choice the triad disappeared and the fixation cue appeared again for the next triad.

70% 60% 50% 40% 30% 20% 10%

!

! !

0% ! SCO DCO PCO ! Triad Type ! "#$%&'!()!Mean percentage action choices with Same Category Object (SCO), Different Category Object (DCO), and Perceptual Category Object (PCO) triads in the context-lean condition (light grey bars) and in the context-rich condition (dark grey bars). Error bars are standard errors of the mean.!

Discussion
The experiment reported here sought to investigate the role of action in shaping categorical decisions. The first aim was to measure how action knowledge was used in the forced-choice triad task when pitted both against and alongside taxonomic information. The results from the different category object (DCO) triads showed that when action knowledge was pitted against taxonomic information participants primarily grouped items together based on taxonomic information. For example participants were more likely to put rifle with sword rather than water pistol. The finding that shared action could not overcome taxonomic constraints is perhaps not surprising given the central role that functional knowledge plays in category membership. Participants were most likely to select the action choice when it was combined with taxonomic information, as with the same-category objects (SCO), therefore showing that knowledge of action is perhaps insufficient on its own to act as a basis for category membership. Therefore while shared action may not be considered a sufficient basis in which to form categories as gauged by this task, it does appear to have an additive effect increasing the shared relations between two items. The perceptual-category object (PCO) triads were designed specifically to determine whether participants were selecting the items based on shared action or shared perceptual properties. If participants were drawing upon action knowledge then in such pairs where the choice comes down to an action or a perceptual choice they should pick the action. In contrast if perceptual information is driving choices then participants should pick the item that looks more similar. The results showed that action knowledge was more likely to be used on the PCO triads over perceptual similarity when shared category membership was removed. Therefore in line with views on embodied semantics we can infer that

Results
The mean percentage action choices for each of the three triad types in both the context-lean and context-rich conditions are illustrated in Figure 2. As can be seen, the SCO triads produced the highest percentage of actionrelated responses in both the context-lean (61%, SD = 15%) and the context-rich condition (70%, SD = 15%). The action choice was selected least often with the DCO triads, though the mean was greater in the context-rich condition (53%, SD = 21%) than in the context-lean condition (32%, SD = 13%). In the PCO triads participants chose the action item less often than the perceptual item in the context-lean condition (48%, SD = 20%) but more often in the context-rich condition (69%, SD = 14%). For all three triad types participants selected the action choice more frequently when contextualised. A 3x2 mixed analysis of variance revealed that the main effect of context was significant, F (1, 48) = 39.22, p < .001, !2 = .45. Participants were more likely to select the action item in the context condition when pictures of the objects were shown in a functional context. The main effect of triad type was also significant, F (2, 96) = 22.77, p < .001, !2 = .32. Post hoc analyses using the Bonferroni adjustment showed that participants selected more action choices overall on the SCO triads than in both the DCO triads (p < .001) and the PCO triads (p = .031). Participants also selected more action choices in the PCO

participants were drawing upon knowledge of how they interacted with the objects rather than how they looked. The second aim of the experiment was to see if action knowledge is drawn upon in all situations or whether its use in such tasks is context-dependent. The results showed that when participants saw the items in the context-rich condition they were more likely to select the action item. Since objects can cue a variety of potential actions, it is possible that when participants viewed the images in the context-lean condition, multiple actions were simulated and as such participants perhaps did not focus on the shared functional action between the objects. In contrast the context-rich condition clearly showcased the objects being used in their standard capacity and as such the shared actions inhibited the simulation of a broader range of potential actions. The relevance of action knowledge in driving categorisation intuitions is thus contingent on the context of presentation. In order to fully extend this research a new set of triads would need to be designed in which the items share an action along with taxonomic information, but do not share perceptual properties. While this would be the ideal condition there might be insurmountable constraints on designing the material required to run this experiment: In aiming to optimise the functionality of the human-artefact interface, objects that share an action will invariably share perceptual properties. For example pencils and paintbrushes look similar as they are designed to be used with a pinch grip and rest within the thenar space of the thumb and index fingers. Items sharing category membership further confines this problem as items become more similar to each other based on the ergonomics of design. Therefore it might prove impossible to find items that require the same method of interaction/operating but that did not share the perceptual properties linked with that action. The results of the experiment have brought attention to the circumstances under which action knowledge informs categorisation intuitions in a passive cognitive task. The results from the current experiment suggest that action knowledge plays an important role in categorisation. As revealed with the same-category object triads, participants were most likely to group items together when they shared category membership and functional action. The use of the perceptual-category object triads showed that this was not due to shared perceptual features between items that share an action. However when participants were asked to choose between either an action or category membership participants were most likely to choose the latter, until such items were shown in a functional context. When participants saw the items in use they were more likely to group the target with the item sharing a functional action. This suggests that viewing items without a context is not enough to instantiate action knowledge and is supported by previous research (Borghi, Bonfiglioi, Lugli, Ricciardelli, Rubichi, & Nicoletti, 2007; Borghi et al., 2012). Action responses are more frequent because such knowledge is accessed more readily upon presentation of objects in a functional context. This is most likely the result of activation of the motor cortex through both the mirror and the canonical neuron systems. Canonical neurons become active when a person sees an object as

well as interact with it, where mirror neurons also become active when they view another person interact with the object (Grèzes, Armony, Rowe & Passingham, 2003). In the present context-rich condition, participants not only saw the object but an agent interacting with it. Research shows that the motor cortex is activated upon viewing physically manipulable objects (Chao & Martin, 2000; Hauk, Johnsrude & Puvermuller, 2004) suggesting that some form of simulation is occurring allowing participants to draw upon action related knowledge. Under these conditions the associated actions are more salient and as such more likely drawn upon with the triads. This could also explain why action choices were lower in the context-lean condition as viewing these conditions should activate the canonical neuron system but not the mirror neuron system. Hence the action choices are less likely to be drawn upon in these conditions. The data here clearly show that action knowledge and motor affordances are more likely to be drawn upon when stimuli are presented within a functional context. In conclusion the data reported here indicate that actionrelated information is influential when participants are engaging in categorisation tasks that do not require any action to be made. This effect is made even more salient when the presentation of the objects is embedded in an action-relevant context. It has further been shown that while perceptual information plays a strong role in categorisation there are circumstances when action knowledge is chosen over perceptual information. The results are consistent with views on embodied semantics that the motor cortex is activated when objects are perceived and this activation influences task performance based on such shared action knowledge.

Acknowledgments
We are grateful to Lia Kvavilashvili for her advice throughout the preparation of this work and comments on earlier drafts.

References
Anelli, F., Nicoletti, R., & Borghi, A. M. (2010). Categorization and action: What about object consistence? Acta Psychologica, 133, 203-211 Aziz-Zadeh, L., & Damasio, A. (2008). Embodied semantics for actions: Findings from functional brain imaging. Journal of Physiology – Paris, 102, 35-39. Barsalou, L. W. (1982). Context-independent and contextdependent information in concepts. Memory & Cognition, 10, 82-93. Barsalou, L. W. (1999). Perceptual symbol systems. Behavioral and Brain Sciences, 22, 577-660. Barsalou, L. W. (2003). Situated simulation in the human conceptual system. Language and Cognitive Processes, 18, 513-562. Barsalou, L. W. (2008). Grounded cognition. Annual Review of Psychology , 59, 617-645. Borghi, A. M. (2004). Object concepts and action: Extracting affordances from objects parts. Acta Psychologica, 115, 69-96. Borghi, A. M., Bonfiglioli, C., Lugli, L., Ricciardelli, P.,

Rubichi, S., & Nicoletti, R. (2007). Are visual stimuli sufficient to evoke motor information? Studies with hand primes. Neuroscience Letters, 411, 17-21. Borghi, A. M., Flumini, A., Natraj, N., & Wheaton, L. A. (2012). One hand, two objects: Emergence of affordance in contexts. Brain and Cognition, 80, 64-73. Bub, D. N., & Masson, M. E. J. (2006). Gestural knowledge evoked by objects as part of conceptual representations. Aphasiology, 20, 1112-1124. Bub, D. N., & Masson, M. E. J. (2012). On the dynamics of action representations evoked by names of manipulable objects. Journal of Experimental Psychology: General, 141, 502-517. Bub, D. N., Masson, M. E. J., & Bukach, C. M. (2003). Gesturing and naming: The use of functional knowledge in object identification. Psychological Science, 14, 467-472. Bub, D. N., Masson, M. E. J., & Cree, G. S. (2008). Evocation of functional and volumetric gestural knowledge by objects and words. Cognition, 106, 2758. Chao, L. L., & Martin, A. (2000). Representation of manipulable man-made objects in the dorsal stream. Neuroimage, 12, 478-484. Creem, S. H., & Proffitt, D. R. (2001). Grasping objects by their handles: A necessary interaction between cognition and action. Journal of Experimental Psychology-Human Perception and Performance, 27, 218-228. Fernandino, L., & Iacoboni, M. (2010). Are cortical motor maps based on body parts or coordinated actions? Implications for embodied semantics. Brain & Language, 112, 44-53. Grèzes, J., Armony, J. L., Rowe, J., & Passingham, R. E. (2003). Activations related to “mirror” and “canonical” neurons I the human brain: An fMRI study. NeuroImage, 18, 928-937. Hauk, O., Johnsrude, I., & Pulvermüller, F. (2004). Somatotpic representation of action words in human motor and premotor cortex. Neuron, 41, 301-307. Helbig, H. B., Graf, M., & Keifer, M. (2006). The role of action representations in visual object recognition. Experimental Brain Research, 174, 221-228. Helbig, H. B., Steinwender, J., Graf, M., & Keifer, M. (2010). Action observation can prime visual object recognition. Experimental Brain Research, 200, 251258. Iachini, T., Borghi, A. M., & Senese, V. P. (2008). Categorization and sensorimotor interaction with objects. Brain and Cognition, 67, 31-43. Jax, S. A., & Buxbaum, L. J. (2010). Response interference between functional and structural actions linked to the same familiar object. Cognition, 115, 350355. Lin, E. L., & Murphy, G. L. (2001). Thematic relations in adults' concepts. Journal of Experimental PsychologyGeneral, 130, 3-28. Murphy, G. L. (2001). Causes of taxonomic sorting by adults: A test of the thematic-to-taxonomic shift. Psychonomic Bulletin & Review, 8, 834-839. Osiurak, F., Roche, K., Ramone, J., & Chainay, H. (2013). Handing a tool to someone can take more time

than using it. Cognition, 128, 76-81. Tipper, S. P., Paul, M. A., & Hayes, A. E. (2006). Visionfor-action: The effects of object property discrimination and action state on affordance compatibility effects. Psychonomic Bulletin & Review, 13, 493-498. Tucker, M., & Ellis, R. (1998). On the relations between seen objects and components of potential actions. Journal of Experimental Psychology: Human Perception and Performance, 24, 830-846. Tucker, M., & Ellis, R. (2004). Action priming by briefly presented objects. Acta Psychologica, 116, 185-203. Vainio, L., Symes, E., Ellis, R., Tucker, M., & Ottoboni, G. (2008). On the relations between action planning, object identification, and motor representations of observed actions and objects. Cognition, 108, 444-465. Wu, L. & Barsalou, L. W. (2009). Perceptual simulation in conceptual combination: Evidence from property generation. Acta Psychologia, 132, 173-189. Yee, E., Chrysikou, E. G., Hoffman, E., & ThompsonSchill, S. L. (2013). Manual experience shapes object representations. Psychological Science, 24, 909-919. Yeh, W., & Barsalou, L. W. (2006). The situated nature of concepts. American Journal of Psychology, 119, 349384.

HYPOTHESIS AND THEORY ARTICLE
published: 24 February 2014 doi: 10.3389/fpsyg.2014.00116

Tool use as distributed cognition: how tools help, hinder and define manual skill
Chris Baber 1 *, Manish Parekh1 and Tulin G. Cengiz 2
1 2

School of Electronic, Electrical and Computer Engineering, University of Birmingham, Birmingham, UK Department of Industrial Engineering, Uludag University, Bursa, Turkey

Edited by: François Osiurak, Université de Lyon, France Reviewed by: Blandine Bril, Ecole des Hautes Etudes en Sciences Sociales, France Lewis A. Wheaton, Georgia Tech, USA *Correspondence: Chris Baber, School of Electronic, Electrical and Computer Engineering, University of Birmingham, B15 2TT Birmingham, UK e-mail: c.baber@bham.ac.uk

Our thesis in this paper is that, in order to appreciate the interplay between cognitive (goal-directed) and physical performance in tool use, it is necessary to determine the role that representations play in the use of tools. We argue that rather being solely a matter of internal (mental) representation, tool use makes use of the external representations that define the human–environment–tool–object system. This requires the notion of Distributed Cognition to encompass not simply the manner in which artifacts represent concepts but also how they represent praxis. Our argument is that this can be extended to include how artifacts-in-context afford use and how this response to affordances constitutes a particular form of skilled performance. By artifacts-in-context, we do not mean solely the affordances offered by the physical dimensions of a tool but also the interaction between the tool and the object that it is being used on. From this, “affordance” does not simply relate to the physical appearance of the tool but anticipates subsequent actions by the user directed towards the goal of changing the state of the object and this is best understood in terms of the “complimentarity” in the system. This assertion raises two challenges which are explored in this paper. The first is to distinguish “affordance” from the adaptation that one might expect to see in descriptions of motor control; when we speak of “affordance” as a form of anticipation, don’t we just mean the ability to adjust movements in response to physical demands? The second is to distinguish “affordance” from a schema of the tool; when we talk about anticipation, don’t we just mean the ability to call on a schema representing a “recipe” for using that tool for that task? This question of representation, specifically what knowledge needs to be represented in tool use, is central to this paper.
Keywords: distributed cognition, tool use, affordances, representation, extended mind, systems dynamics

INTRODUCTION The central question for this paper is what representations are employed when using tools? In this paper, the term “representation” is taken to mean a set of parameters which describe an action (from goal to execution). In broad terms, one answer to this question might see the set of parameters as being specified prior to an action being performed, e.g., in the form of an action schema, or as being recruited in preparation of the action, e.g., in the form of activation of specific brain regions. In this case, the question becomes one of identifying what the representation might contain and where it might be stored. This is what we refer to as an “internal representation.” Alternatively, the parameters might arise from the performance of the action in response to constraints imposed by the environment, e.g., in the dynamic behavior of a system. This is what we refer to as an “external representation.” We argue that, while there is evidence to support the view that tool use can be guided by “internal representation,” this only provides a partial view of such activity and that the use of “external representation” can provide a viable alternative account. The position taken in this paper assumes that the physical behavior of the person can be viewed as part and parcel of their cognitive activity, and that there is a close coupling between a

person’s action and their perception of features of objects in the world. However, neither assumption fully captures human activity when using physical objects for goal-directed activity (which is the broad definition of tool-use employed in this paper). Thus, we argue for a broader appreciation of Gibson’s (1979) notion of complimentarity as an explanation of affordance at a“system”level. The notion of “system” here draws on Maravita and Iriki’s (2004) idea of the “hand-tool body schema” but we extend this to cover person–environment–tool–object. For us, this requires the notion of Distributed Cognition to encompass not simply the manner in which artifacts represent concepts but also how they represent praxis. In other words, the design of the tool (as a human-made artifact) reflects not only the manufacturing process but also a set of assumptions about how that tool should be grasped and manipulated, and how activities involving that tool can be performed “correctly.” This means that “tools” are distinct from other physical objects in the human environment because their use is defined not only by their appearance or the user’s goals but also by cultural constraints that have influenced their production (Baber, 2003, 2006; Burghardt et al., 2011). While there are instances in which other physical objects, such as sticks or stones, can fulfil tool functions, and while the neurological evidence suggests that images

www.frontiersin.org

February 2014 | Volume 5 | Article 116 | 1

Baber et al.

Tool use as distributed cognition

of these objects activate similar regions in the brain to images of tools, there is accumulating evidence that the pattern of brain activation for tools is somewhat different from that of physical objects per se.
“Human beings, viewed as behaving systems, are quite simple. The apparent complexity of our behaviour over time is largely a reflection of the environment in which we find ourselves” [Simon, 1969]

While Simon was not talking explicitly about Distributed Cognition, this quotation points to the need to understand human behavior in the environment in which it occurs. For us this implies a need to better understand how the environment makes an impact on our actions and decisions, and this suggests the benefit of an approach which studies human action as they occur in natural (or as near natural as possible) conditions. This raises challenges for “ecological validity” (Neisser, 1967) which takes us out of the laboratory (or, for that matter, the brain scanner) and into the settings in which activity is performed. A primary reason for this quest is the assumption that the relations between human, environment, tool, and object are fundamental to the study of perception and action (Gibson, 1979; Beek and Bingham, 1991; Newell, 1991). A study of the activities of tool use away from typical environments runs the risk of ignoring the constraints that the environment places on the performance of these activities. Thus, it is vital to ensure that enough of the characteristics of the person–environment–tool–object system are reflected in the design of studies (even if these are conducted in laboratories). We are interested in ways in which we might be able to capture data from the tool using actions of people in work environments, through analyzing video of their activity (and discussing these videos with them) or through putting sensors on the tools that they use. For this paper, the focus will be on the use of data collected from sensors on tools. Two areas of activity will be used in this paper: using hand-tools in jewelery and eating with cutlery. In both areas, the concern will be to compare experienced and less experienced users of the tools. The comparison will be qualitative rather than quantitative, i.e., examples of the data collected during our studies will be presented but more detailed analyses of these data will be found in other papers.
WHAT NEEDS TO BE REPRESENTED IN TOOL USE?

moves the person towards their goal. This strikes us as an elegant reformulation of the notion of affordance as a goal-directed, physical response to the environment. The difference between this view and the one presented in this paper is simply (we believe) a matter of scale: rather than considering problem solving in the broad terms that Osiurak et al. (2010) offer, our focus is on the interface between tool and object (or, rather, we propose that the “problem” that concerns tool users is how to modify the object in ways that satisfies a goal, given the constraints that the tool (and the tool-users’ ability to wield that tool) might impose on their action). In order to explore further the question of representation in tool use, it is important to consider what needs to be represented in order to use a tool. Tool use is not only a matter of recognizing that an object is a tool but also of knowing how to hold and manipulate that particular tool. It is also a matter of understanding the consequences of a particular way of using a particular tool. Knowing that a piercing saw (used by jewelers to cut metals) is held vertically for cutting (with the wrist more or less locked and most of the motion about the elbow), and has teeth which cut in one direction, leads to an understanding that the cut is made on the downstroke (not the upstroke), and helps define a set of possible actions when using this tool. From this it might appear that we are arguing for (at least) some representations of the tool and the actions associated to be internal to the person. Does this mean that these representations are stored in the brain?

By way of a definition of the word “tool,” we propose that a tool is a physical object which lends itself to manipulation by a human (or animal) in order to solve a problem presented by objects in the physical environment. This notion of tool-use as a form of problem-solving not only emphasizes the goal-directed aspect of using tools but also the need to respond to, and overcome, constraints. This definition allows us to combine both the physical action of manipulating the tool with the cognitive aspects of goal-directed, purposeful behavior. Following a similar line of argument (tool use as problem solving), Osiurak et al. (2010) suggest that the coordination of the physical actions involved in using tools represent a problem to be solved. They view cognition and physical activity in a dialectic in which a particular goal encourages the perception of particular affordances in the world and serves to influence the bodily action to perform, which, in turn,

INTERNAL REPRESENTATION: NEURAL ACTIVATION IN TOOL USE The suggestion that the use of tools depends on “internal models” is nicely encapsulated in a recent paper by Imamizu and Kawato (2012). They review literature and report studies which indicate the existence of both a feed-forward model, taking efference copies of motor commands to enable motion dynamics, and inverse models used to manage these dynamics. During learning, changes in cerebellar activity indicate the acquisition and refinement of such models. As we argue in this paper, the notion that brain-based “internal models” are causal represents a particular view of tool use, and we are proposing that it is possible to explain much of the activity involved in tool use through a combination of Distributed Cognition and dynamics which might not be represented in the brain per se. However, before exploring this proposal further, we consider some of the neuropsychological evidence relating to tool use. Imamizu and Kawato (2012) review neuropsychological studies of tool use and suggest that, “[A]lthough the brain regions related to each type of component cannot be uniquely determined. . .” (p. 325) there are two distinct functional regions of the brain related to tool use: one related to the physical skills involved in dextrous tool manipulation, and one related to the semantic and conceptual knowledge relating to the functions of tools (see also Lewis, 2006; Higuchi et al., 2007, 2009). These distinct regions are discussed in more detail in the rest of this section. In their now classic study, Chao and Martin (2000) used functional magnetic resonance imaging (fMRI) to show that viewing and naming of tools led to activation of the left ventral premotor cortex, suggesting a strong relationship between the physical

Frontiers in Psychology | Cognition

February 2014 | Volume 5 | Article 116 | 2

Baber et al.

Tool use as distributed cognition

appearance of objects and the fact these objects could be acted upon. Grafton et al. (1997) used positron emission tomography (PET) scanning of participants asked to observe or (silently) name tools and their use. Observation of tools resulted in strong activation of the left dorsal premotor cortex, and (silent) naming of these tools resulted in additional activation of Broca’s area. However, naming the use of the tools led to activation in Broca’s area, together with activation in left dorsal premotor cortex, left ventral premotor cortex, and left supplementary motor area. This implies that naming the use of a tool (even when the action is not performed with it) has motor valence which is additional to that obtained when looking at the tool. It also suggests that the physical appearance and name of a tool activates slightly different areas than the use of the tool. Taken together these, and related, studies imply that brain activation relates to specific properties of the tool-as-form and tool-as-function, and that these properties are not solely related to a tool’s physical appearance but also to how it moves or how it is used (Johnson-Frey, 2004; Martin, 2007). One suggestion is that representations of tools are held in specific regions of the brain and become activated during activities in which similar objects are used. According to Gallivan et al. (2013) the distributed coding of different actions associated with hand movement and tool use imply that these actions are represented separately and then integrated in the frontoparietal cortex. As Yee et al. (2013) show, in an ingenious experiment, asking people to think about manipulable objects when they are performing manual actions which are incompatible with those objects is difficult (but it easy to think about non-manipulable objects during the performance of such actions). This suggests that the meaning of objects (specifically in terms of their properties which support manipulation) is recruited during action, and that incompatible action interferes with this. Furthermore, work by Hoeren et al. (2013) points to the suggestion that the recognition of action (performed by other people) is processed using distinct streams: the dorso-dorsal stream focusing on movement determined by the properties of the objects being used, and the dorso-ventral stream focusing on functional appropriateness and dexterity of task performance.
KNOWLEDGE OF (FAMILIAR) TOOL USE

The discussion so far points to the need to draw on knowledge of the appropriateness of a given tool for a given task and how to wield that tool to achieve the most effective result. Riddoch et al. (2006) presented patients (manifesting visual extinction) with images of pairs of objects. The pairs showed objects which people are likely to have experienced being used together (e.g., a bottle and a glass), or objects which could plausibly be used together, although might not have been experienced as such (e.g., a bottle and a bucket), or were randomly paired in order to, as far as possible, produce pairs which had no association. The results showed that commonly paired objects were identified more quickly than plausibly paired objects which, in turn, were identified more quickly than the randomly paired objects (although this latter finding only held when the image showed the objects being used together rather than having them presented side by side). One implication of this work (which could be applied to normals as well as patients) is that

the common and plausible pairs activate familiar routines in tool use. In contrast with this observation, Vingerhoets (2008) found that presentation of images of “familiar” or “unfamiliar” tools activated the same brain regions, with “unfamiliar” tools generating more activation in the left hemispheric medial posterior occipital and inferior posterior temporal areas (in comparison to images of “familiar” tools) and more activation around the supramarginal gyrus for the familiar tools. While these results showed strong individual differences, they also imply that the activation in response to “familiar” tools can be associated with knowledge of the appropriate hand position for the use of the tool (as opposed to simply whether or not the tool could be grasped). A similar line of argument comes from studies in which participants are asked to pick up handled objects (such as cups) when the handle faces either towards or away from the hand that they are instructed to use (Tucker and Ellis, 2001). For example, Bub et al. (2012) presented images of everyday objects together with images of hands in different orientations. The objects all had handles which were either oriented horizontally, e.g., pliers, frying pan, or vertically, e.g., beer mug, hairdryer. Participants were asked to name the object. Reaction (naming) time was significantly faster when both hand and wrist orientation matched the type of handle, or when neither hand and wrist orientation matched the handle, but much slower when either hand or wrist orientation was incongruent. Relating this to the previous discussion of neural imaging, one can assume that the photographs of the hands and the objects might have activated different regions, with a combination occurring prior to response. The suggestion that there might be preparative neural activity which corresponds to different types of action (Rizzolatti et al., 1988) could provide evidence for the recruitment of a set of representations determining task performance. Certainly the movement-related cortical potential (MRCP) recorded from electroencephalography (EEG) begins 2–3 s before the onset of movement (Toma and Hallett, 2003; Wheaton et al., 2005). Furthermore, onset seems to be proportional to complexity of movement, with more complex movements having longer onset times. Such activity, typically in the left posterior parietal cortex, is taken to indicate the need to manage complex motor activity and, as Wheaton et al. (2005) propose may include “. . .imagining executing such movements; the goal of the movement; determining the natural position and setting required for proper performance; sequence of motor acts and comprehension of the task.” (p. 535). While we have every reason to accept that complex movements involve recruitment of appropriate muscle groupings and specification of appropriate control parameters, we do not see why this necessarily involves the definition of specific representations of the task context. Thus, our debate is not with the neurological evidence per se but with the assumptions that these must point to internal representations which drive behavior. What is interesting in the Bub et al. (2012) study is less the reinforcing of activation of congruent images (or, indeed, the effect of incongruence) than the problems caused when one of the hand images did not match the other image or the object. Bub et al. (2012) suggest that this reflected disruption of the plan

www.frontiersin.org

February 2014 | Volume 5 | Article 116 | 3

Baber et al.

Tool use as distributed cognition

being developed in working memory (with the images activating particular judgments about using tools). However, the images presented in these studies serve as the (external) representations about which people are asked to make judgments. As such the idea that they would need to create corresponding internal representations in order to make such judgments seems a little odd. The images that are presented provided sufficient information to make a judgment and the need is to determine whether these are “true” or “possible.” On the one hand, it seems plausible to assume that prior experience provides the “grounding” (Mizelle and Wheaton, 2010) of a tool in terms of its usage, but on the other hand, it is equally plausible that this could be part of the person’s action repertoire (e.g., in terms of Bernstein’s (1967) idea of coordinative structures) as it is activation of specific regions of the brain. For an action in which participants had to use different tools to touch a target, precuing the target had no benefit on performance, but precuing the tool to use had significant benefits (Massen and Prinz, 2007). We take this to suggest that the precuing of the tool enabled the recruitment of the appropriate “coordinative structure,” to use Bernstein’s (1967) phrase describing combinations of muscle enervation and limb movement, to perform the task with a given tool. What is interesting about this interpretation of their findings is that “representation” need not be same for different tasks (and, we would argue, shows how it can shift to outside the brain per se ). Hermsdörfer et al. (2006) compared performance of apraxic patients with a control group of normals on a sawing task. Participants were asked to demonstrate sawing under three conditions: when they were shown a photograph of a saw and asked to pantomime sawing; when they were shown the photograph, given a piece of wood (the same size as the saw’s handle) to hold and asked to pantomime sawing; when they were given the actual saw to hold. While the controls showed fairly consistent performance across the three conditions, apraxic patients showed motion errors (deduced from 3D motion tracking) in the first two conditions. Typically, these errors involved substituting mediolateral motion for the anteposterial motion expected. Interestingly, these errors were not apparent when the apraxic patients were given the actual saw to use. On the one hand, this supports a common finding in apraxic studies (that providing people with the physical object seems to enable them to perform tasks more effectively and reliably than when they do not have the object to hand). On the other hand, we believe it tells us something about the need for internal representation when using tools. Hermsdörfer et al. (2006) conclude that “. . . pantomiming the use of a tool and actually using the tool are facilitated by largely different neural processes which differ in demands and goals.” [p. 1651]. We would argue further that these differences arise because the use of the tool involves the control of the person– environment–tool–object system and need not depend on internal representation.
CONCLUSION

influenced by the action which one intends to perform with the tool. We have a repertoire of appropriate grasps for manipulable objects, and we adapt these grasps according to contextual demands. The adaptation often occurs with sufficient fluency and speed to make it unlikely that we have simply retrieved a particular piece of “motor schema” from memory and applied this; indeed, the very notion of a “motor schema” (with its attendant implication of stored sequences of action) has been called into question (Sherwood and Lee, 2003; Shea and Wulf, 2005). Thus, we argue the tool user is, partly, using the tool to make changes to objects in the environment, but also partly using the tool to help create further opportunities in the environment for using the tool. In other words, tool use is an interplay between seeking a defined goal and managing the affordances arising from changes in the object in the environment (resulting from the ongoing use of tools). Before discussing the collection of data and their analysis, the next section describes the particular stance taken in this paper: Distributed Cognition.

Just because the tool-using behaviors have neural correlates does not mean that these are the only places in which representations for the behaviors exist. Clearly, the type of grasp is likely to be

EXTERNAL REPRESENTATION: DISTRIBUTED COGNITION AND THE EXTENDED MIND As the phrase implies, Distributed Cognition addresses situations in which the processing of information occurs outside the brain. For some writers, this is the proposal that the environment and the objects it contains can shape the way in which cognition is performed (Zhang and Norman, 1994; Hutchins, 1995a,b; Scaife and Rogers, 1996). While this position could be seen as paraphrasing the well-known observation that the representation of a problem space influences the strategy that problem solvers apply (Chase and Simon, 1973; Larkin et al., 1980; Chi et al., 1981), e.g., changing the layout of a puzzle can make it easier or harder to solve, it also points to the importance of interactivity in behavior. For example, people playing Tetris or Scrabble can benefit (in some situations) by being allowed to manipulate and rearrange the playing pieces (Kirsh and Maglio, 1994; Maglio et al., 1999). This points to the need to not simply focus on the arrangement and design of the problem representation, but also on the nature of the interaction between person and objects. From this point of view, “embodiment” becomes an essential feature of acting not only on the objects but also on the cognitive tasks involved in problem solving. In other words, rearranging the pieces is not simply performed in order to assist thinking, it is thinking. This is taken to mean that the relationships within the human–environment–tool–object system not only supports (or affords) different actions but also shapes cognition (Wilson, 2002). The reason for this is that activity within this system is often time-limited, in that the actions are performed at speed, in real-time and offer little opportunity for planning (what Clark, 1997, has termed “mind on the hoof ”). From this, the main purpose of cognition (in tool use) is to support action in as situation-appropriate manner as possible. It also suggests that, rather than needing to construct “internal representations” of the environment, it is sufficient to respond to the appearance of the environment. From his work with robots, Brooks (1990) pointed out that robot performance could be more efficient if they spent less time “planning” and creating representations, and more

Frontiers in Psychology | Cognition

February 2014 | Volume 5 | Article 116 | 4

Baber et al.

Tool use as distributed cognition

time “acting” because “the world is its own best model” [Brooks, 1990, 12].
CONSIDERING AFFORDANCE

Our reading of Gibson’s (1977, 1979) concept of affordance lies in his notion of “complimentarity” in which the properties of objects in the environment are responded to by the animal. Turvey (1992) offers the term“effectivity”as a way of capturing these properties of the animal. So, one way of seeing affordance lies in the complimentarity between the object’s properties and the animal’s effectivity. One of the problems that the idea of “effectivity” and “properties” raises is the suggestion that these are separate aspects which are brought together during the performance of a task, which implies that they are independent, autonomous features which become coupled during task performance. Indeed, Gibson (1979) suggests that the “affordance” exists whether or not the observer perceives or attends to it. If this is the case, then it makes sense to assume that one aspect of the “effectivity” of the task performer would be the neural representations of the actions involved in performing the task (as well as morphological features and motor skills). There are many situations in which the observer cannot but attend to the affordance, e.g., perseveration in the behavior of stroke patients, or response to “fake” cues by animals. Rather than implying (as Gibson seems to) that the “affordance” is an invariant property of the environment, the fact that perseveration is an unusual state of affairs suggests that humans (and some animals) are able to choose to respond to affordances (and by implication, to see affordances in different situations). This implies that what is essential to “affordance” is this combination of the property of the object in the environment and the effectivity of the specific individual (with specific knowledge, skills, abilities, and goals) in that environment. While the properties of the object in the environment may well be invariant, the actual affordance arises from the complimentarity of environment and actor. Affordance is partly a matter of perception-action coupling and partly a matter of intention (goal) – action coupling. Perhaps a better way of putting this is that perception-action coupling is mediated by the intentionality of the actor. However, as Chemero (2009) points out, the idea of separable components that can be coupled runs counter to the notion of complimentarity; taking affordance as the result of the system created by person–environment–tool–object (as we do in this paper) leads to the conclusion that this is a system which is non-decomposable and which exists only during the performance of the given task. From this, we suggest that the goal, in the person–environment–tool–object system is partly held by the person (in terms of the effect that they intend to produce on the object) and partly situated (Suchman, 1987) in the ongoing interactivity in the system. This assumption echoes the earlier assertion of van Leeuwen et al. (1994) that tool use can be “. . . defined as performing an action on a target by performing an action on a tool. The action on the tool is embedded in the action on the target.” [van Leeuwen et al. (1994), p. 188–189]. For van Leeuwen et al. (1994) this embedding reflected a “higher-order affordance structure” of “mutually constraining complimentarities.” van Leeuwen et al. (1994) argued that it was important to understand the role of context in task performance in terms of a sufficiency principle, i.e., “if an affordance has already been realized,

there is no need to take it into account.” [van Leeuwen et al. (1994), p. 190]. To take this a little further, Turvey (1992) suggests that affordance might play a role in “predictive control” of activity and, while the analysis (and indeed use of the term), in this paper might differ from his, the idea that affordance refers not only to immediate action but to future actions is central to the ideas in this paper. Additionally, Mizelle et al. (2013) discuss the notion of functional affordance, in which there is an optimal manner in which a given object can be used to achieve a desired goal. For example, Mizelle et al. (2013) note that a hammer can be held a variety of grasps (some involving the handle, some involving the head, for instance) but that there is a grasp which “. . . best affords the action of driving a nail. . .” (p. 280). This can be seen as taking the predictive control further, in that there is a goal state against action can be optimized. While these notions of affordance could be represented internally (in terms of specific neural correlates of functional affordance that can be adapted to contextual demands), the notion of complimentarity followed in this paper offers a more parsimonious explanation. In other words, the Gibsonian notion of affordance is taken in this paper to describe a particular form of complimentarity in the person–environment–tool–object system, and it is the “system” as a whole which can be said to optimize the tool-using activity.
TASKONOMY AND HOW THE ENVIRONMENT AFFORDS SKILLED ACTION

One way in which the environment can be created to provide affordances for future action is in the ways in which experts lay out their workspace. In their discussion of blacksmiths, Keller and Keller (1996) use the term “taskonomy” to refer to the ways in which an expert’s knowledge of the tasks to be performed help create the arrangement (taxonomy) of tools in their space. This arrangement is not simply a matter of having particular types of tools kept near each other, but arises through a combination of tools and actions. A similar pattern can be seen in the workspace of the jeweler (Figure 1). As the jeweler performs a particular task, so a tool is picked up, used and then laid down in the workspace; as work progresses so tools are either reused or new ones introduced. However, the expert is often able to describe what work had been completed in a particular workspace by looking at the collection of tools in the immediate vicinity. In some cases, specialized tools will be brought to the workspace with the intention of supporting a particular goal. Thus, the workspace becomes managed to provide particular affordances (in terms of available tools and the position in which these tools are placed to support particular types of grasp). This suggests the anticipation of tasks and the arrangement of the workspace in line with these anticipations. In these ways, the movement of tools in the workspace (as the result of deliberately selecting these in preparation of a specific job, or as the result of picking up and putting down the tools during the performance of the job, or as the result of moving tools which are no longer needed further away from the central point of reaching) becomes part of the structuring of the workspace. Rather than simply reflecting the ebb and flow of actions in the workspace, we argue that this reflects the management of potential affordances and, as such, is a form of Distributed Cognition. The suggestion

www.frontiersin.org

February 2014 | Volume 5 | Article 116 | 5

Baber et al.

Tool use as distributed cognition

FIGURE 1 | “Taskonomy” in jewelers’ workspaces.

that moving tools around the workspace is a form of “cognition” is logically the same as the suggestion that presentation of a problem will “frame” the approach to the solution and that manipulating pieces in a puzzle might be a form of thinking. In other words, layout of the workspace will frame the actions which are most likely to be performed and this framing is the result of deliberate choices made to retain, discard or move a tool after it has been used (rather than merely a consequence of moving tools around).

WORKING WITH TOOLS In his discussion of craftwork, David Pye draws the useful distinction between “certainty” and “risk” in craftwork. He argues that in “the workmanship of certainty” there is an impetus to design work to ensure consistency, repeatability, and minimize variation or ambiguity. Such work involves heavily proscribed procedures and measures of quality and could be interpreted in terms of industrialized production processes. In this approach, the artifact being produced will be tightly specified prior to production and the resulting artifact will be considered in terms of this specification. Anyone who has constructed flat-pack or selfassembly furniture will have encountered a situation in which the manufacturer has sought to encourage workmanship of certainty. However, anyone who has built self-assembly furniture will also recognize the challenges that this poses. Misreading the instructions or believing that you know what you are doing so don’t need to read the instructions can lead to results which differ from the goal. This could be quite minor (a handful of left over components) or quite major (the door which doesn’t open, the shelf which drops out when the unit is stood up). This variation illustrates the workmanship of risk. This, in turn, reflects the variability in outcome which can arise from decisions made by the worker during the performance of the tasks. The decisions could reflect a choice of tool, or knowledge/skill in the use of the tool, but they could equally reflect responses to the opportunities presented (or constraints created) by the materials being used. For example, the knot in a piece of wood, or the finish on one side of the self-assembly wardrobe, could constraint the actions which are possible or could suggest an appropriate action to perform. In contrast, the “workmanship of risk” does

not involve such tight specification, i.e., “. . . the quality of the result is not predetermined, but depends on the judgment, dexterity and care which the maker exercises as he works.” (Pye, 1968, p. 20). Rather than the intent or purpose being predetermined, it is now something which crystalizes through the developing interaction between craftworker, tools, and materials being worked. This is something which we noted in our study of jewellery making (Baber and Saini, 1995): the jeweler worked to very sketchy “plans” but adapted these plans to suit the resulting state of the material, often modifying a particular ring or brooch to capitalize on a particular facet that they noticed as the metal was being worked.
“First, the experienced worker usually employs “smoother ” and more consistent movements…Secondly, the experienced worker operates more rhythmically, indicating that a higher degree of temporal organization has been achieved. Thirdly, the experienced worker makes better use of the sensory data. . .Fourthly, the experienced worker reacts in an integrated way to groups of sensory signals, and makes organized grouped responses to them” [Seymour, 1972, 35–36]

The quotation from Seymour (1972) indicates how the output of the human–environment–tool–object system is being optimized, but not necessarily how the dynamics of the system relate changes in input to output. In order to consider this, we turn our attention to series of studies conducted by Bril and her colleagues, focusing on tasks involving hammering (either stone hammers to knap flint or metal hammers to shape stone or glass beads).
SYSTEM DYNAMICS: TRANSFORMATIONS IN TOOL USE

Our actions, when using tools, involve the coordination of a set of transformations (Biryukova and Bril, 2012). We transform kinetic energy into tool motion – but need to appreciate how much energy to exert in order to produce the desired motion of the tool (and in order to produce the desired effect on the object from the tool’s motion). We manage dynamic transformations, balancing the movement of the tool in the air and on the object with our own motions and with the outcome of the tool’s activity. We anticipate what effect the tool’s motions will produce and relate these to the outcomes that we desire. As Ingold

Frontiers in Psychology | Cognition

February 2014 | Volume 5 | Article 116 | 6

Baber et al.

Tool use as distributed cognition

(2000) points out, “Intentionality and functionality are . . . immanent in the activity itself, in the gestural synergy of human being, tool and environment.” (Ingold, 2000, p. 352.) The ability to both anticipate the outcome of the tool’s action and manage the functionality of the tool are an integral part of the use of the tool. The dynamics of using the tool thus becomes far more important than might be implied by the neurological imaging work which concentrates on the form and function of the tool. Given that these (and related) transformations need to be managed during the use of tools, it is worth asking where these transformations might be represented? If they are “represented” simply during the performance of an action, and arise from the moment-to-moment correction of the action, then one might not expect to see anticipatory effects. On the other hand, if there is evidence of anticipation (of the consequences of any of these transformations) then this implies a need to represent the consequence and the question remains, where does this representation reside and what form does it take? Before considering the questions of transformations, it is worth repeating some of the observations from these studies regarding expertise. For example, Roux et al. (1995) showed that expert craftsmen (making stone or glass beads) showed significantly less inter- and intra-individual variations in performance than less experienced workers. Similarly, Biryukova and Bril (2008) showed that expert knappers used a larger repertoire of joint angle combinations than their less qualified colleagues (who tended to demonstrate more rigid behavior), and Bril et al. (2010) showed that experts showed a lower variability in kinetic energy compared to intermediates and novices. In related work, Vernooij et al. (2012) explored learning in a task involving the use of a 300-g hammer-stone. Analysis of motion tracked during the performance of this task showed inter-individual differences in the ways in which joint angles were combined to strike a particular type of blow and that these combinations changed during the course of the study. The analysis of learning to use such a hammer suggested that participants were only able to modify one parameter (relating to joint angles or impact force) but not both at the same time, until they had gained proficiency in the task. In a series of experiments comparing expert, intermediate and novice users of stone hammers (in flint-knapping tasks conducted in the laboratory), Bril et al. (2010) identify three primary parameters that seem to contribute to the dynamics of tool use in this context. The first are Control Parameters, such as the velocity with which the hammer stone approached the target. The study showed that, in general, novices appreciated the need to control velocity but were not able to control this efficiently (this finding is supported by the work on Vernooij et al., 2012, discussed above). Thus, we would expect greater variability in the novice performance on these control parameters; as Seymour (1972) put it, the expert actions would be performed in a “smoother,”“more consistent,”“more rhythmical[ly]”manner. The second set of parameters considered by Bril et al. (2010) are regulatory parameters, such as the trajectory followed by the hammer stone and the potential energy applied. Experts tended to show shorter trajectories and smaller ratios between parameters. In Seymour’s (1972) terms, this shows how experts are able to use a “higher degree of temporal

organization” and also to make “better use of the sensory data” in managing their actions. As Bril et al. (2010) note, “In the present task, the velocity of the hammer had to be controlled to produce the required kinetic energy in relation to the mass of the hammer. This was achieved by concurrently changing the trajectory, the amplitude of the movement, and the muscular force. In this perspective, the movement became meaningful only in relation to the production of functional parameters at the level of the task, which allowed for movement flexibility as long as the task requirements were fulfilled.” (Bril et al., 2010, p. 837). This quotation introduces the third parameter, the Functional parameter, such as kinetic energy, which experts appear to hold constant and aim to apply the lowest kinetic energy that is sufficient for the task. As the experiments involved presenting participants with hammers of different weights and requiring them to produce flakes of different sizes, one can assume that all participants would be able to discern changes in hammer weight or task demands (in terms of flake size), but the results suggest that a characteristic of expertise (which was not available to the novices) was the ability to respond to “nested relationships” (Wagman and Carello, 2003) between weight of hammer and size of flake to produce. The ability to appreciate these “nested relationships” allowed the experts to interpret the constraints placed on them by the person–environment–tool–object relationships and respond to these in ways that the novices could not. So, we return to the question of where these constraints might be represented? One possibility (implied by Seymour, 1972 and mooted by Bril et al., 2010) is that the initial representations involve Functional parameters which are learned and then adapted to changes in context. In his discussion of dexterity, Bernstein (1967) highlighted that the main determinant was not bodily movement so much as the capability to respond to changes in the conditions surrounding the person. Bernstein’s (1967) notions of tool use, in terms of dexterity, relate to the quotation from Simon (1969) at the start of this paper. The manipulation of tools is rarely an end in itself but is performed with the intention of shaping objects in the environment. The actions performed lead to changes in the objects but also indicate the intentionality of the tool user (providing they have sufficiently dexterity in their use of the tools). The expert tool user thinks through the tools that are used because the actions performed with the tools shape the environment in such a way as to solve the problems that it presents and in such a way as to produce the results that the tool user desires. The action performed with the tool also creates the opportunities for the next action; and this, in turn, reflects the type of grip and posture which the tool user adopts. In this way, grip and posture (in holding and using a tool) indicate the chosen solution to the problem that the tool user is solving. In much the same way that Rosenbaum et al. (2012) speaks of end-state comfort (and the ways in which a posture anticipates a particular end-state following the movement), so we can think of the ways in which the tool user is continually seeking to adapt their current motions in anticipation of subsequent motions and states of the object. For this paper, we take this to mean that the skilled tool-user is better able to coordinate the person–environment–tool–object system and to anticipate how changes in this system require adaptation of activity. This realtime adaptation need not imply internal representation of either

www.frontiersin.org

February 2014 | Volume 5 | Article 116 | 7

Baber et al.

Tool use as distributed cognition

FIGURE 2 | (A) An instrumented handle and (B) using a file in an instrumented handle to remove paint from a piece of wood.

FIGURE 3 | Example of data collected from experienced silversmith using a file. The data were sampled at 120 Hz. Velocity is derived from accelerometer data, de-trended using a moving average of 100 samples and grip force is the average output from the Analog to Digital

Converter (ADC). The Y velocity line describes anteposterial motion, the Z velocity line describes vertical displacement, and the top grip describes the force applied to the top of the handle (pressing down on to the metal).

task dynamics or some form of “motor program.” Rather, the expert is able to produce movements which are coordinated to task goals (being more efficient and economical in terms of energy use). In a sense, expertise is the practiced adaptation of intrinsic dynamics to task dynamics (where task dynamics are defined by the person–environment–tool–object system) so that changes in task constraints and affordances can be appropriately responded to through subtle tuning of actions. This implies that experts are able to modify the pattern of activity without necessarily impairing the

functional impact of the activity. We do not believe that experts need to possess, or even represent, these various patterns of activity but rather these arise on-the-fly during the coordinated control of limbs holding and controlling tools.

STUDIES USING SENSORS FITTED TO THE HANDLES OF TOOLS In order to explore these questions of dynamics, we have been exploring ways in which to capture behavior in the field (or, at

Frontiers in Psychology | Cognition

February 2014 | Volume 5 | Article 116 | 8

Baber et al.

Tool use as distributed cognition

FIGURE 4 | Spectrograms from three participants performing one of the “paint removal” tasks. The spectrogram of frequencies 0–20 Hz over time for y -movement (i.e., back and forth motion of the file across the wood), moving between the 1st and 3rd dots on the file with the goal of removing two layers of paint.

least, in laboratory and workshop settings which are as close to the field as possible). This has involved designing and developing handles which combine different types of sensor to capture the actions a person performs. Often such data are collected from using camera systems with markers on the person. While these can be very accurate, they are not easy to use in the field. Thus, it makes more sense to instrument the person or their tools in order to collect data in situ. We have taken the lead from Bril and her colleagues (discussed in the previous section) to instrument our tools (Figure 2). In our work, strain gages are used to capture force applied to the handle and a three-axis accelerometer is used to capture motion (Parekh and Baber, 2010 for a description of the design of these handles).

In order to appreciate how experience in using a given tool can shape activity, Figure 3 presents an extract of recordings (from a three-axis accelerometer and strain gages integrated into the handle of a jeweler’s file) of an experienced silversmith filing the edge of a metal strip. Figure 3 shows three filing strokes over the course of 2.5 s. Each stroke (occurring at approximately 11.6, 12, and 12.9 s) is indicated by an increase in the y -velocity data. There are two types of stroke here: rapid (at 11.6 and 12.9 s) in which the file in moved rapidly across the metal, and slow (12 s) in which the file is drawn more slowly over the metal. During each stroke, there is downward pressure on the file (indicated by the decrease in z -velocity data and increase in “top grip” force applied to the top of the handle). Immediately following the stroke, the file is lifted up (increase in z -velocity) and brought back to the starting point. Prior to the next stroke the file is adjusted and aligned with the metal (which takes around 1 s), which involves little change in grip force applied and z-velocity. The top grip loosens as the file position is reset for lifting the file off the object (movement in the z direction); the expert user only applies force on the forwards motion. This action is partly dictated by the file being used and partly by the results that the tool user intends. As the expert said, you can remove metal easily enough but you can’t put it back. So filing is about removing sufficient (but not too much) of the metal. Furthermore, the metal being worked (copper in this instance) could easily be dulled if too much of the upper surface was removed, and so filing was also a matter of retaining the luster of the metal. Such knowledge can affect the way in which the tool is wielded and influence the outcomes that one might expect when using the tools. In another study, we asked novice users of a file to remove paint from a piece of wood. Figure 2B shows the task being performed. There are three dots painted on the top of the file and participant was instructed to ensure that the file was kept between the first and second, or the first and third dots. Contrasting three people performing the filing task (Figure 4), we can see that while the main activity (yellow on the spectrographs) occurs at similar frequencies, the harmonics vary. These variations might reflect differences in strategy. We would expect to see harmonics from these data due to the periodicity of the repetitive motions employed. This also suggests that differences in performance can be captured through a better appreciation of dynamics and, potentially, following the lead of Bernstein (1967), can be reflected in the conservation of energy of the tool users. The raw accelerometer data were integrated to produce velocity, on which we applied a Fourier transform to determine the fundamental frequency of the filing motion. Table 1 suggests that the main determinant of this fundamental frequency is not the tool-specific goal to keep the two dots inside the wood, but the task-specific goal to remove one or two layers of paint.

CULTURAL AFFORDANCES

In this section, we turn our attention to the broader question of cultural effects in tool use. For the sake of the discussion, we restrict ourselves to the simple assumption that cultural constraints can

www.frontiersin.org

February 2014 | Volume 5 | Article 116 | 9

Baber et al.

Tool use as distributed cognition

Table 1 | Comparing fundamental frequency of filing for task and tool-directed goals. Task goal Tool-directed goal Filing white paint down to show red paint Keep file between dot a and dot b F0 6.201 Hz Keep file between dot a and dot c 5.518 Hz Filing white paint down to show bare wood Keep file between dot a and dot b 3.174 Hz Keep file between dot a and dot c 3.467 Hz

have a bearing of the experiences that people might have with specific types of tools and, in particular, can serve to define acceptable or proper ways in which particular artifacts are used. Thus, one question that can be used to address the issue of “culture” in tool use is to ask how should one properly use cutlery, such as a spoon, knife or fork? In their study of eating (kale or water) with a spoon, van der Kamp and Steenbergen (1999) used video-based motion tracking to record arm motion. The likelihood of spilling the contents of the spoon (kale or water) when it was moved from bowl to mouth increased the number of corrective sub-movements made during the action which affected the kinematic profile of the movement. The contents of the spoon also affected head motion. Participants were more likely to move their head towards the spoon when it contained water which, in turn, shows how the coordination of the motion system (i.e., contents–spoon–hand–arm–head) changes in response to task demands. Interestingly, the study also hinted at variation in “eating styles” which reflected individual differences in performance. We are interested in how these “eating styles” might also reflect cultural responses to cutlery and how culture defines the “proper” way to use an item of cutlery. Of course, the use of the word “properly” is deliberately provocative and culturally loaded. At one level, “proper” use could simply mean that food is moved from plate to mouth in a controlled manner, in sufficient quantities to make it easy to eat. At another level, “proper” use could relate to various social mores and rules of etiquette in terms of how the knife and fork are held and moved, and how much food is held on the fork or put into the mouth. For example, in her discussion of using forks, Visser (1991) contrasts the “English” style (of eating from the back of the fork tines and holding the knife in the other hand) with the “American” style (of eating from the bowl of the fork and swapping, or “zig-zagging” fork and knife). We asked participants, using a knife and a fork (fitted to our instrumented handles), to perform a somewhat unusual version of “eating.” The task goal required participants to lift a forkful of sweet-corn to their chin. This breaks down into: “load fork’, “lift fork’, and “terminate” (e.g., most participants simply tipped the fork to drop the sweet-corn back onto the plate). The “English” or “American” styles outlined above are illustrated by Figure 5. In order to consider variation, we selected one participant who was familiar with the “English” style (Figure 6) and one of the participants who had never used cutlery in this manner (Figure 7). Figures 6 and 7 show that variability in the data from the inexperienced user are consistently higher than the experienced user for both grip and accelerometer data. This echoes the earlier findings relating the variability in “skilled” performance. Rather than the “skill’, in this case, being the result of instruction, training and practice (as one might expect in the use of hand-tools), these

results hint that enculturation and exposure to particular beliefs about appropriate use of cutlery can have an impact on the ease with which these artifacts are manipulated in different ways.

DISCUSSION
We use tools to solve the problems that objects in the environment present to us. This is an obvious statement but hides a couple of points which are worth noting. The first is that intention which underlies the use of the tool combines a task goal with the affordances of the tool–object interface, and the constraints of the person–environment–tool–object system. This means that “cognition” becomes the active response to the affordances of the interaction between tool and object in terms of the task goal that the user is seeking to achieve. Taking Gibson’s notion of complimentarity, we can say that the dynamic aspect of this activity continually shapes the actions of the person as much as it shapes the state of the object. In other words, the states of the object, environment, tool, and person become combined to form the focus of action and, by implication, to help frame and reframe the task goal. One might expect the task goal to be kept constant during the performance of the task. However, our discussions with, and observations of, expert jewelers suggests that this not entirely the case. While the high-level objective might remain the same (e.g., produce a ring of a particular size set with a particular stone), the development of the “plan” to achieve this goal adapts to the state of the metal and the performance of the task. Thus, the task goal would appear to follow the notion of “situated action” (Suchman, 1987) which changes with context. This raises the second point, that, the focus of action is context-dependent and the context is continually changing. So, tool use is enactive, embedded and embodied. The comparisons of experienced and inexperienced users of tools (and cutlery) considered in this paper show that expertise not only involves less variability in physical performance but also better control of energy expended in the performance of a given task with a given tool. We believe that this points to the well-known assertion that the expert develops a “feel” for the tool, and often prefer to use their own tools for particular tasks because these have become very familiar to them. Indeed, a potential problem that we face with the instrumented handles that we use is that these feel different from those that the experienced tool users prefer. Anecdotally, only the experienced tool users commented on the feel (weight, balance, material) of these handles during the data collection. The skilled craft-worker will often speak of the tool becoming part of the body, and the feeling of manipulating the tool being akin to simply moving the hand in which the tool held. For some writers, this implies that the tool can be considered as a physical

Frontiers in Psychology | Cognition

February 2014 | Volume 5 | Article 116 | 10

Baber et al.

Tool use as distributed cognition

FIGURE 5 | Comparing English (left) and American (right) cutlery use.

FIGURE 6 | Consistent “English” use (over six separate attempts as indicated by thick lines between each attempt). The pattern of grip force applied (particularly to the fork handle) and the

smoothness of the fork’s accelerometer trace show how the experienced participant’s repetitions are consistent and reflect a well-practiced motion.

www.frontiersin.org

February 2014 | Volume 5 | Article 116 | 11

Baber et al.

Tool use as distributed cognition

FIGURE 7 | Variable “English” use (over six separate attempts). The inexperienced participant shows large variation in grip force and accelerometer trace for the fork. During the task, his preferred approach was to tilt the fork on its side and move it towards the kernels, using the knife as a stop. He then held the fork at an angle and used the knife to keep the kernals

pressed to the tines as he lifted both knife and fork. There is less correlation shown between grip and activity from the knife, which is being pushed on to the top of the fork, and, particularly towards the right of the graphs, the fork is held with force primarily only on two sides of the handle as opposed to a full grip.

extension of the person and that, therefore, motor control becomes a matter of adapting to the added potential of the “extended-limb.” However, rather than simply being a matter of planning movement with the addition of the tool, it is plausible to suggest that the tool changes the perception of space around the tool user (Maravita et al., 2002). “People who use tools. . .build an increasingly rich implicit understanding of the world in which they use the tools. . .” [Cutler, 1994, p. 80]. In her discussion of representations in tooluse, Massen (2013) emphasizes the need to appreciate how tools become part of the peripersonal space of the user, such that “there is no need to distinguish between external goal locations to which the tool has to be moved and the locations to which the bodily effector has to be moved.” (p. 2). While this makes sense when considering movements with tools, it overlooks an equally important aspect of the skilled craft-worker. The reason that the tool feels as if it is part of the person is because it “disappears” from attention which becomes more and more focused on the object being worked on. This suggests that, rather than the tool being an extension of the body, it makes more sense that the tool creates a focus of attention – with the sense that the tool’s movement becomes so central to attention that the control of the limb operating it becomes less important. This suggests that, rather than considering the tool-hand combination, it is more important to consider the toolobject combination because this is where the skilled practitioner is attending. The use of tools, by experts, seems to involve anticipatory, feedforward control of movement (as well as rapid and efficient use of feed-back through all of their senses) in which subtle adjustments in the manipulation of the tool are performed in order to effect desired changes in the object being worked on. Not only does this explain the minimal variability but also highlights the central

question of this paper; if so much of the activity of the expert tool user is anticipatory, how are these anticipations represented? We propose that it is not sufficient to only look in the brain of the expert tool user to discover these representations. Even if there are regions which are active under specific conditions, the skill of the expert tool user comes from the ability to control their activity with sufficient spare capacity to cope with future demands and to respond to the changing context in which they are using the tools to effect changes in the object being worked on. The idea that the environment (and the objects it contains) can be interpreted in different ways, suggests that these become “external representations” to which the person responds. Response is partly a matter of knowledge, skill and ability of the person, partly a matter of fit between action and environment and partly a matter of the nature of the environment and the objects it contains. As the person focuses on specific aspects, which are relevant to the task (of shaping a piece of metal or arranging tools in a workspace) so these aspects become the cognitive space in which subsequent decisions are made. Tool use, as a form of problem solving, becomes a matter of making these decisions as the cognitive space changes; and a means of acting upon the cognitive space to create new opportunities. This further suggests that much of the activity which is assumed to be “feed-forward” (in the sense that there needs to be a model which guides behavior) could be explained by fast-acting, negative feedback loops (integrated across several sensory modalities) which support moment-by-moment correction through solving the inverse kinematics problems of positioning a given tool in a given position in order to effect change in the object being worked on. We believe that much of the “representation” drawn upon in the use of tools can be in the form of external representations (the

Frontiers in Psychology | Cognition

February 2014 | Volume 5 | Article 116 | 12

Baber et al.

Tool use as distributed cognition

objects and tools in a given environment, particularly in support of the situated action of ongoing planning in tool use) and in the form of coordinative structures (the control and management of physical activity, particularly in terms of feed-forward control of movement and use of feed-back from the results of the movement). In other words, following the lead of Riccio (1993) and the more detailed arguments of Chemero (2009) an internal representation is not necessary for the control, coordination and (we propose) planning of tool use because it is sufficient for the tool user to have the ability to perceive the state of the object on which she works and to manipulate the tool in order to produce a particular pattern of perceptions (and, in this case, we suggest that these patterns are equally as likely to be olfactory, haptic, and auditory as visual). This ability becomes manifest only during the performance of a person–environment–tool–object system (echoing Butler’s claim that “strictly speaking, nothing is a tool except during use”) and this system can be described using System Dynamics, in which the systems goal is the optimization of specific movement parameters in order to produce an effect on a given object. This reduces the need for there to be internal representations per se (see also Barrett, 2011). Furthermore, any“representation”that the tool user employs is likely to spread across the entire nervous system rather than solely in regions of the brain. From this, the strong and compelling evidence accumulated from the activation of specific regions in the brain is taken to indicate the result rather than the cause of tool using behavior (whether observed, imagined, or performed) which arises from the recruitment and activation of coordinative structures (Bernstein, 1967) through task-specific devices (Beek and Bingham, 1991). While our paper has not sought to present evidence in support of this claim, we believe that this statement helps to bring together the ongoing work that we have reviewed and raises the opportunity to develop testable hypotheses for future exploration of the ways in which people use tools.

ACKNOWLEDGMENTS
An early draft of this paper was much improved through comments from Anthony Chemero and two anonymous reviewers.

REFERENCES
Baber, C. (2003). Cognition and Tool Use: Forms of Engagement in Animal and Human Use of Tools. London: Taylor and Francis. Baber, C. (2006). Cognitive aspects of tool use. Appl. Ergon. 37, 3–15. doi: 10.1016/j.apergo.2005.06.004 Baber, C., and Saini, M. (1995). The Role of Tacit Skills in Jewellery Manufacture, Contemporary Ergonomics. London: Taylor and Francis, 92. Barrett, L. (2011). Beyond the Brain: How Body and Environment Shape Animal and Human Minds. Princeton, NJ: Princeton University Press. Beek, P. J., and Bingham, G. P. (1991). Task-specific dynamics and the study of perception and action: a reaction to von Hofsten (1989). Ecol. Psychol. 3, 35–54. doi: 10.1207/s15326969eco0301_3 Bernstein, N. A. (1967). The Coordination and Regulation of Movements. Oxford: Pergamon. Biryukova, E. V., and Bril, B. (2008). Organization of goal directed action at a high-level of motor skill: the case of stone-knapping in India. Motor Control 12, 181–209. Biryukova, E. V., and Bril, B. (2012). Biomechanical analysis of tool use: a return to Bernstein’s tradition. Zeitshcr. Psychol. 220, 53–54. Bril, B., Rein, R., Nonaka, T., Wenban-Smith, F., and Dietrich, G. (2010). The role of expertise in tool use: skill differences in functional action adaptations

to task constraints. J. Exp. Psychol. Hum. Percept. Perform. 36, 825–839. doi: 10.1037/a0018171 Brooks, R. A. (1990). Elephants don’t play chess. Robot. Autonom. Syst. 6, 3–15. doi: 10.1016/S0921-8890(05)80025-9 Bub, D. N., Masson, M. E. J., and Lin, T. (2012). Features of planned hand actions influence identification of graspable objects. Psychol. Sci. 24, 1269–1276. Burghardt, G. M., Shumaker, R. W., Walkup, K. R., and Beck, B. B. (2011). Animal Tool Behavior: The Use and Manufacture of Tools by Animals. Baltimore, MD: The John Hopkins University Press. Chase, W. G., and Simon, H. A. (1973). “The mind’s eye in chess,” in Visual Information Processing, ed. W. G. Chase (New York: Academic Press), 215–281. Chao, L. L., and Martin, A. (2000). Representation of manipulable manmade objects in the dorsal stream. Neuroimage 12, 478–484. doi: 10.1006/nimg.2000.0635 Chemero, A. (2009). Radical Embodied Cognitive Systems. Cambridge, MA: MIT Press. Chi, M. T. H., Feltovich, P. J., and Glaser, R. (1981). Categorization and representation of physics problems by experts and novices, Cogn. Sci. 5, 121–152. doi: 10.1207/s15516709cog0502_2 Clark, A. (1997). Being There: Putting Body, Brain, and World Together Again. Cambridge, MA: MIT Press. Cutler, A. (1994). The Hand of the Master: Craftsmanship, Ivory and Society in Byzantium, Princeton, NJ: Princeton University Press. Gallivan, J. P., McLean, D. A., Valyear, K. F., and Culham, J. C. (2013). Decoding the neural mechanisms of tool use. eLife e00425, 1–29. Gibson, J. J. (1977). “The theory of affordances,” in Perceiving, Acting and Knowing: toward an ecological psychology, eds R. Shaw and J. Bransford (Hillsdale, NJ: LEA), 67–82. Gibson, J. J. (1979). The Ecological Approach to Visual Perception. Boston, MA: Houghton-Mifflin. Grafton, S. T., Fadiga, L., Arbib, M. A., and Rizzolatti, G. (1997). Premotor cortex activation during observation and naming of familiar tools. Neuoroimage 6, 231–236. doi: 10.1006/nimg.1997.0293 Higuchi, S., Chaminade, T., Imamizu, H., and Kawato, M. (2009). Shared neural correlates for language and tool use in Broca’s area. Neuroreport 20, 1376–1381. doi: 10.1097/WNR.0b013e3283315570 Higuchi, S., Imamizu, H., and Kawato, M. (2007). Cerebellar activity evoked by common tool-use execution and imagery tasks: an fMRI study. Cortex 43, 350–358. doi: 10.1016/S0010-9452(08)70460-X Hermsdörfer, J., Hentze, S., and Goldenberg, G. (2006). Spatial and kinematic features of apraxic movement depenn on the mode of execution. Neuropschologica 44, 1642–1652. doi: 10.1016/j.neuropsychologia.2006.03.023 Hoeren, M., Kaller, C. P., Glauche, V., Vry, M.-S., Rijntjes, M., Hamzei, F., et al. (2013). Action semantics and movement characteristics engage distinct processing streams during the observation of tool use. Exp. Brain Res. 229, 243–260. doi: 10.1007/s00221-013-3610-5 Hutchins, E. (1995a). Cognition in the Wild. Cambridge, MA: MIT Press. Hutchins, E. (1995b). How a cockpit remembers its speeds. Cogn. Sci. 19, 265–288. doi: 10.1207/s15516709cog1903_1 Ingold, T. (2000). The textility of making. Cambridge J. Econ. 34, 91–102. doi: 10.1093/cje/bep042 Imamizu, H., and Kawato, M. (2012). Cerebellar internal models: implications for the dextrous use of tools. Cerebellum 11, 325–335. doi: 10.1007/s12311-0100241-2 Johnson-Frey, S. H. (2004). The neural bases of complex tool use in humans. Trends Cogn. Sci. 8, 71–78. doi: 10.1016/j.tics.2003.12.002 Keller, C. M., and Keller, J. D. (1996). Cognition and Tool Use: The Blacksmith at Work. Cambridge: Cambridge University Press. Kirsh, D., and Maglio, P. (1994). On distinguishing epistemic from pragmatic action. Cogn. Sci. 19, 513–549. doi: 10.1207/s15516709cog1804_1 Larkin, K., McDermott, J., Simon, D., and Simon, H. (1980). Expert and novice performance in solving physics problems. Science 208, 1335–1342. doi: 10.1126/science.208.4450.1335 Lewis, J. W. (2006). Cortical networks related to human use of tools. Neuroscientist 12, 211–231. doi: 10.1177/1073858406288327 Maglio, P. P., Matlock, T., Raphaely, D., Chernicky, B., and Kirsh, D. (1999). “Interactive skill in scrabble,” in Proceedings of the Twenty-First Annual Conference of the

www.frontiersin.org

February 2014 | Volume 5 | Article 116 | 13

Baber et al.

Tool use as distributed cognition

Cognitive Science Society (Englewood Cliffs, NJ: Lawrence Erlbaum Associates), 326–330. Maravita, A., and Iriki, A. (2004). Tools for the body (schema). Trends Cogn. Sci. 8, 79–86. doi: 10.1016/j.tics.2003.12.008 Maravita, A., Spence, C., Kennett, S., and Driver, J. (2002). Tool-use changes multimodal spatial interactions between vision and touch in normal humans. Cognition 83, 25–34. doi: 10.1016/S0010-0277(02)00003-3 Martin, A. (2007). The representation of object concepts in the brain. Annu. Rev. Psychol. 58, 25–45. doi: 10.1146/annurev.psych.57.102904.190143 Massen, C. (2013). Cognitive representations of tool-use interactions. New Ideas Psychol. 31, 239–246. doi: 10.1016/j.newideapsych.2012.12.002 Massen, C., and Prinz, W. (2007). Programming tool-use actions. J. Exp. Psychol. Hum. Percept. Perform. 33, 692–704. doi: 10.1037/0096-1523.33.3.692 Mizelle, J. C., and Wheaton, L. A. (2010). The neuroscience of storing and molding tool action concepts: how “plastic” is grounded cognition? Front. Psychol. 1:195. doi: 10.3389/fpsyg.2010.00195 Mizelle, J. C., Kelly, R. L., and Wheaton, L. A. (2013). Ventral encoding of functional affordances: a neural pathway for identifying errors in action. Brain Cogn. 82, 274–282. doi: 10.1016/j.bandc.2013.05.002 Neisser, U. (1967). Cognitive Psychology. New York: Appleton-Century-Crofts. Newell, K. M. (1991). Motor skill acquisition. Annu. Rev. Psychol. 42, 213–237. doi: 10.1146/annurev.ps.42.020191.001241 Osiurak, F., Jarry, C., and Le Gall, D. (2010). Grasping the affordances, understanding the reasoning: towards a dialectical theory of human tool use. Psychol. Rev. 117, 517–540. doi: 10.1037/a0019004 Parekh, M., and Baber, C. (2010). “Tool use as gesture: new challenges for maintenance and rehabilitation,” Proceedings of the 24th BCS Interaction Specialist Group Conference (Swindon: British Computer Society), 241–249. Pye, D. (1968). The Nature and Art of Workmanship. Cambridge: Cambridge University Press. Riccio, G. E. (1993). “Information in movement variability about the qualitative dynamics of posture and orientation,” in Variability and Motor Control, eds K. M. Newell and D. M. Corcos (Champaign, IL: Human Kinetics), 317–359. Riddoch, M. J., Humphreys, G. W., Hickman, M., Clift, J., Daly, A., and Colin, J. (2006). I can see what you are doing: action familiarity and affordance promote recovery from extinction. Cogn. Neuropsychol. 23, 583–605. doi: 10.1080/02643290500310962 Rizzolatti, G., Camarda, R., Fogassi, L., Gentilucci, M., Luppino, G., and Matelli, M. (1988). Functional organization of inferior area 6 in the macaque monkey. II. Area F5 and the control of distal movements. Exp. Brain Res. 71, 491–507. doi: 10.1007/BF00248742 Rosenbaum, D. A., Chapman, K. M., Weigelt, M., Weiss, D. J., and van der Wel, R. (2012). Cognition, action and object manipulation. Psychol. Bull. 138, 924–646. doi: 10.1037/a0027839 Roux, V., Bril, B., and Dietrich, G. (1995). Skills and learning difficulties involved in stone knapping: the case of stone-bead knapping in Khambhat, India. World Archaeol. 27, 63–87. doi: 10.1080/00438243.1995.9980293 Scaife, M., and Rogers, Y. (1996). External cognition: how do graphical representations work? Int. J. Hum. Comp. Stud. 45, 185–213. doi: 10.1006/ijhc.1996.0048 Seymour, W. D. (1972). Industrial Skills. London: Pitman. Shea, C. H., and Wulf, G. (2005). Schema theory: a critical appraisal and reevaluation. J. Motor Behav. 37, 85–102. doi: 10.3200/JMBR.37.2.85-102 Sherwood, D. E., and Lee, T. D. (2003). Schema theory: critical review and implications for the role of cognition in a new theory of motor learning. Res. Q. Exer. Sport 74, 376–382. doi: 10.1080/02701367.2003.10609107

Simon, H. (1969). The Sciences of the Artificial. Cambridge, MA: MIT Press. Suchman, L. A. (1987). Plans and Situated Actions: The Problem of Human-Machine Communication. Cambridge: Cambridge University Press. Toma, K., and Hallett, M. (2003). “Generators of the movement-related cortical potentials and dipole source analysis,” in The Bereitschaftspotential: Movementrelated cortical potentials, eds M. Jahanshahi and M. Hallett (Berlin: Springer), 113–130. Tucker, M., and Ellis, R. (2001). The potentiation of grasp types during visual object categorization. Vis. Cogn. 8, 769–800. doi: 10.1080/135062800420 00144 Turvey, M. (1992). Affordance and prospective control: an outline of the ontology. Ecol. Psychol. 4, 173–187. doi: 10.1207/s15326969eco0403_3 van der Kamp, J., and Steenbergen, B. (1999). The kinematics of eating with a spoon: bringing the food to the mouth, or the mouth to the food? Exp. Brain Res. 129, 68–76. doi: 10.1007/s002210050937 van Leeuwen, L., Smitsman, A., and van Leeuwen, C. (1994). Affordances, perceptual complexity and the development of tool use. J. Exp. Psychol. Hum. Percept. Perform. 20, 174–191. doi: 10.1037/0096-1523.20. 1.174 Vernooij, C. A., Mouton, L. J., and Bongers, R. M. (2012). Learning to control orientation and force in a hammering task. Zeitschr. Psychol. 220, 29–36. doi: 10.1027/2151-2604/a000088 Vingerhoets, G. (2008). Knowing about tools: neural correlates of tool familiarity and experience. Neuroimage 40, 1380–1391. doi: 10.1016/j.neuroimage.2007.12.058 Visser, M. (1991). The Rituals of Dinner. London: Penguin. Wagman, J. B., and Carello, C. (2003). Haptically creating affordances: the user– tool interface. J. Exp. Psychol. Appl. 9, 175–186. doi: 10.1037/1076-898X. 9.3.175 Wheaton, L. A., Yakota, S., and Hallett, M. (2005). Posterior parietal negativity preceding self-paced praxis movements. Exp. Brain Res. 163, 535–539. doi: 10.1007/s00221-005-2314-x Wilson, M. (2002). Six views of embodied cognition. Psychon. Bull. Rev. 9, 625–636. doi: 10.3758/BF03196322 Yee, E., Chrysikou, E. G., Hoffman, E., and Thompson-Schill, S. L. (2013). Manual experience shapes object recognition. Psychol. Sci. 24, 909–919. doi: 10.1177/0956797612464658 Zhang, J., and Norman, D. A. (1994). Representations in distributed cognitive tasks. Cogn. Sci. 18, 87–122. doi: 10.1207/s15516709cog1801_3 Conflict of Interest Statement: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Received: 28 November 2013; accepted: 27 January 2014; published online: 24 February 2014. Citation: Baber C, Parekh M and Cengiz TG (2014) Tool use as distributed cognition: how tools help, hinder and define manual skill. Front. Psychol. 5:116. doi: 10.3389/fpsyg.2014.00116 This article was submitted to Cognition, a section of the journal Frontiers in Psychology. Copyright ©2014 Baber, Parekh and Cengiz. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

Frontiers in Psychology | Cognition

February 2014 | Volume 5 | Article 116 | 14

The Hands That Guide the Thinking: Interactivity in Mental Arithmetic
Lisa G. Guthrie (l.guthrie@kingston.ac.uk)
Department of Psychology, Kingston University, Kingston upon Thames, KT1 2EE, UNITED KINGDOM

Julia K. Mayer (julia.mayer@alumni.maastrichtuniversity.nl)
Faculty of Psychology and Neuroscience, Maastricht University 6200 MD Maastricht, THE NETHERLANDS

Frédéric Vallée-Tourangeau (f.vallee-tourangeau@kingston.ac.uk)
Department of Psychology, Kingston University, Kingston upon Thames, KT1 2EE, UNITED KINGDOM

Abstract
Whether it is in mining distal cultural influences or using more proximal artefacts, problem solving in the wild routinely scaffolds on the basis of interacting with resources outside the head. Individuals often gesture, point or use objects as an aid to solving quotidian arithmetic problems. Interactivity has been linked to better performance in problem solving, possibly due to a more efficient allocation of attentional resources and better distribution of cognitive load. Previous research suggests an interplay between the cognitive and motor system whereby the later can lighten the strain on working memory capacity (GoldinMeadow, Nusbaum, Kelly & Wagner, 2001; Carlson, Avraamedes, Cary & Strasberg, 2007; Vallée-Tourangeau, 2013). In attempting to simulate these moves made in the world, different levels of interactivity were examined with a series of mental arithmetic problems. Participants were also profiled in terms of attitude to varying problem presentations as an assessment of their engagement in the task. The integration of artefacts, such as tokens or a pen, provided individuals with the possibility to explore the opportunities afforded by a dynamic modification of the problem. Mental arithmetic performance was more accurate and more efficient under these conditions. Participants also felt more positive about and better engaged with the task when they could reconfigure the problem presentation through interactivity. These findings underscore the importance of engineering task environments that support distributed problem representation and adequate levels of interactivity that creates a dynamically shifting topography of action affordances. Keywords: Interactivity; Mental arithmetic; Problem solving; Distributed Representation; Task engagement.

pose a relatively high cognitive load on an individual’s internal resources, such as working memory (Ashcraft, 1995; DeStefano & Lefevre, 2004). Numbers are held, added and manipulated in order to solve the problem employing different working memory subsystems, including storage, retrieval and allocation of attentional resources. Dependent on the complexity and length of a mental arithmetic task, the demands of finding a solution may impose a relatively low or high cognitive load, potentially imposing substantial demands on working memory capacity. This capacity may, however, be stretched or reduced by certain internal or external factors, which can subsequently paint a misleading profile of an individual’s true arithmetic capabilities (Ashcraft & Moore, 2009).

Interactivity
The internal cognitive and physical resources deployed to tackle a problem may be taxed by various features of the task—such as time pressure, level of difficulty, fatigue. Reasoners naturally recruit artefacts and use the physical space to make thinking easier and more efficient. Increased levels of interactivity have been linked to better performance, possibly due to a stronger focus of attention and better distribution of cognitive load (Goldin-Meadow, Nusbaum, Kelly & Wagner, 2001; Carlson, Avraamides, Cary & Strasberg, 2007; Vallée-Tourangeau, Sirota & Villejoubert, 2013; Weller, Villejoubert, & ValléeTourangeau 2011). Previous research implicates an interplay between the cognitive and motor system whereby the later can lighten the strain on working memory capacity reducing the expense of resources to the task (Goldin-Meadow, Alibali, & Church, 1993). Improved effectiveness, indicated by increased accuracy and speed, has also been related to movement execution, such as nodding and pointing (Goldin-Meadow et al., 2001), as well as manipulations of the problem’s spatial arrangement (Vallée-Tourangeau, 2013). So it seems that the shaping and re-shaping of the problem presentation can help surpass the original limitations of working memory capacity by lowering the expense of internal resources necessary to solve the task and guide attention

Introduction
Mathematical problems are embedded in everyday life in a variety of different shapes and forms. When confronted with an arithmetic task, people often rearrange the physical display by interacting with the environment. They might move coins while counting their money, note subtotals with a pen or use their hands to gesture, point or count (Kirsh, 1995; Neth & Payne, 2001). Mental arithmetic tasks often entail strategic thinking and deliberate information processing, which require time and effort (Vallée-Tourangeau, 2013). Besides basic, well-rehearsed sums, computations are generally said to

Figure 1: The board on the top left is an example of a standard template used for all four conditions. The board on the top right shows the participant undertaking the pen-paper condition. The board on the bottom left is an example of the wooden tokens in preparation for the participant. On the bottom right the same board after the participant has completed the task. Note the congenial groups of the numbers. (Weller et al., 2011). This could subsequently increase efficiency. Thus, interacting with the environment and utilizing artefacts can increase efficiency by distributing the storage and computational demands of the task across resources internal and external to the reasoner. Such distributed cognitive processes shift the cognitive load from the reasoner onto a system in which she is embedded (Vallée-Tourangeau, 2013). Mathematical tasks are frequently assessed in terms of accuracy and efficiency. Accuracy measures the precision of the calculated solution in relation to the correct answer. Efficiency, on the other hand, involves a relation between invested effort and resulting performance (ValléeTourangeau, 2013). Yet, it is not only the problem itself or its complexity that impacts how accurately or efficiently an individual performs in a mathematical task. The presentation of a problem can guide behaviours and strategic choices in the path to a solution (ValléeTourangeau, Euden, & Hearn, 2011). Embedded in this problem presentation are the varying possibilities for interaction, the dynamic loop of information and action flowing between a person and the outside world, the nature of these interactions having the potential to direct strategic choices (Neth & Payne, 2001; Kirsh 2013). Kirsh (1995) describes an organizing activity that recruits external elements such as the hands, coins and pen and paper to reduce cognitive load as a complimentary strategy to the internal processes of cognition. In turn this coupling of the internal mind with that which is external to the skull generates a distributed system of thinking. suggestion that the activity by which learning is experienced may provide a stimulus for this engagement (Shernoff, Csikszentmihalyi, Schneider, & Shernoff, 2003). It is also possible that a task that offers a student a sense of connection to the real world is more likely to maximize student engagement (Newmann, Wehlage, & Lamborn, 1992). Furthermore Schiefele and Csikszentmihaly (1995) discuss the importance of the affective experience on performance, while engaging in mathematics in the classroom. Positive emotions elicited by the task experience may contribute to increased problem-solving capacities (Shernoff et al., 2003).

The Current Experiment
This experiment explored the role of interactivity in adult participants using tangible artefacts with which the participants can modify the problem presentation as they attempt to complete the arithmetic task. Thus the external problem presentation tracks the dynamic interface between the agent’s internal representation and the world. Previous research on the role of interactivity in mathematical reasoning and learning has generally presented material either on paper or a computer display. Interactivity and the potential to re-shape the problem presentation was manipulated in terms of four conditions. In the first, participants added a sequence of single-digit numbers with their hands down and in a second they were allowed to point at the numbers. Thus in these two conditions, the problem presentation can not be modified, but participants can engage in some complementary actions (Kirsh, 1995) in the latter. In the other conditions participants could re-shape the problem presentation: in the third, they were given a pencil and could recast the sum as they saw fit in arriving at a total. In the fourth, the

Attitude Toward the Task
Student engagement in performing academic tasks may be an influential factor in learning and achievement, with the

sums were presented as a randomly arrayed set of wooden tokens that participants were invited to move to arrive at the correct sum. Across these four different levels of interactivity, performance was measured in terms of accuracy and efficiency. Not only did we expect accuracy to be influenced by interactivity, but efficiency should be related to the degree to which participants can modify the problem presentation as they compute the totals. By efficiency we mean, the degree of accuracy relative to the resources invested in completing the sum. We operationalized resources as the time taken to do the sum. We expected that interactivity conditions that made it possible for participants to manipulate the problem presentation in a manner that reflected and complemented their internal processing to yield the highest level of accuracy and efficiency.

Method
Participants
Sixty participants (40 females, mean age 23.32, SD = 4.41) were recruited for this experiment.

proportion of time invested in solving that set (out of the longest time the slowest participants invested in solving that set). For each of the four conditions, participants were first ranked according to their averaged latencies. The average of the slowest 25% served as a reference point and represented the maximum effort one could expend in that condition. Thus the efficiency ratio denominator was a given participant’s latency over the average latency for the slowest quartile; the numerator was that participant’s proportion correct solutions in that condition. For example, a participant in a given condition may have solved three out of the five sums, for a proportion .6 correct. In turn, the participant’s average latency for completing the five sums in that condition might have been 30 seconds. If the average latency for the slowest quartile was 40 seconds, then that participants invested 75% (30/40) of the total possible time for completing the sums in that condition. The efficiency ratio for that participant would then be .6/.75, or .8. Level of Interactivity. Interactivity was manipulated in terms of four experimental conditions; namely (i) static, (ii) pointing, (iii) pen-paper and (iv) token. In the static condition, participants were asked to compute the sum mentally with their hands flat on the table. In the pointing condition, there were no restrictions on movement, other than to exclude the use of the pen to make notes. Hence, participants were allowed to use their fingers to point to the numbers that composed the sum. In the pen and paper condition, participants were given a pen and were allowed to write on the sheet provided by the experimenter containing the number set. Finally, in the token condition, the sums were presented in the form of round numbered wooden tokens (2cm in diameter), which could be moved by the participants. The format of the presentation was visually constant and the material was always presented on the same surface. Attitude Toward Task Assessment (ATTA). Shernoff et al. (2003) used the Effective Sample Method (ESM) to measure a number of factors including affective experiences. This affective experiences component of the ESM questionnaire was used as the basis for a scale, Attitude Toward Task Assessment (ATTA) designed to assess the engagement of participants in the tasks undertaken in this study. A scale composed of eight items was created to assess an individual’s attitude towards completing the sums in each experimental condition. The eight items asked participants to rate how easy, pleasurable, fun, threatening, stressful, tiresome or effortful the task was and how motivated they were to perform well in the task. Each item was scored on an 8-point Likert scale, labeled from zero (definitely not) to seven (definitely yes). Total scores could range from zero to 56 - the higher the score the more positive the attitude toward the task. Each participant completed the same ATTA scale four times once following each of the four conditions. The alpha reliability of the eight-item scale for each experimental condition indicated that the scale had good reliability (Static, Cronbach’s " = .80; Pen-paper, Cronbach’s " = .77; Pointing, Cronbach’s " = .78; Tokens, Cronbach’s " = .77).

Materials and Measures
Arithmetic Task. All participants were presented with five sets of numbers in four conditions and asked to calculate the sum of the numbers. Therefore each participant calculated 20 sums over the experimental session. They were requested to calculate each set as quickly and accurately as possible. Each set consisted of 11 single-digit numbers. For the purpose of the present study, single-digit numbers between one and nine were first categorized as low (1-4) or high (5-9) in order to generate the range of possible sums in a more principled manner. Four groups of sums were created: Group I (5 low, 6 high), Group II (only high), Group III (3 low, 8 high) and Group IV (4 low, 7 high). Each of these groups was assigned to one of the four interactivity conditions, and this assignment was counterbalanced across participants. Sums were presented in the form of templates which consist of 11 circles (2.2cm) covering between half and ! of a side of A4, that was delineated by a varnished cutting board on which participants carried out all the additions (see Fig 1). By altering the order of the templates within each group, the visual presentation of the sums was additionally randomized for each set. For the token condition, templates of tracing paper were created with the same configurations of the constituent numbers as the paper version of the other conditions. Wooden tokens were placed in the corresponding position and the tracing paper was removed before the start of the task. The numbers were not revealed to the participant until all tokens were in place. Maths performance was measured in terms of accuracy (the correct answer), absolute error (the absolute deviation from the correct answer) latency (time taken until answer verbalised) and efficiency. Participants were not given feedback concerning the accuracy of their answers. Efficiency was calculated as a ratio of the proportion of correct answers for a given problem set over the

80% 75%

Accuracy

3.5

Deviation

3.0

Mean Percent Correct

70% 65% 60% 55%

Mean Absolute Error
Static P&P Pointing Tokens

2.5

2.0

1.5

50%

1.0
Static 1.4 P&P Pointing Tokens

30

Latency

Efficiency

28

Mean Latency (s)

1.3

26

Mean Effieicncy
Static P&P Pointing Tokens

1.2

24

22

1.1

20

1.0 Static P&P Pointing Tokens

Figure 2: Mean percentage correct sums (top right panel), absolute calculation error (top left panel), latency to solution (bottom left panel) and mean calculation efficiency (bottom right panel) in the four experimental conditions. Error bars are standard errors of the mean.

Results
Accuracy
The mean number of correct answers, as shown in the top left panel of Figure 2, was greatest in the token (M = .69, SD = .22) and the pen-paper (M = .69, SD = .23) conditions. The pointing condition (M = .66, SD = .26) indicated slightly less accurate calculations, with the static condition resulting in the weakest performance (M = .60, SD = .30). A one-factor repeated measures analysis of variance (ANOVA) indicated a significant difference between the conditions, F(3,177) = 3.12, p = .027, !2 = .050. Post-hoc tests revealed a significant difference between the static and the pen-paper conditions (p = .006) and the static and token conditions (p = .020). Absolute Error Deviation from the correct answer was greatest in the static condition (M = 2.64, SD = 2.39); the pointing (M = 1.90, SD = 2.43) and pen-paper (M = 1.61, SD = 1.65) conditions produced lower deviations than the static condition while the lowest deviations from the correct answer were observed in the token condition (M = 1.41, SD = 1.69; see top left panel of Fig. 2). The one-factor repeated measures ANOVA revealed a significant difference between the tasks, F(3,177) = 6.34, p < .001, !2 = .097, with post-hoc tests indicating a significant difference between the pen-paper and the token conditions when compared to the static condition (p = .005, p < .001 respectively).

Latency The latency data are shown in the bottom left panel of Figure 2. Participants generally took about the same amount of time to complete the task across the four conditions (static M = 26.79, SD = 9.88; pen-paper M = 27.26, SD = 9.73; pointing M = 25.70, SD = 10.09; token M = 26.58, SD = 10.41). The main effect of interactivity in the one-way repeated measures ANOVA was not significant, F < 1. Efficiency As illustrated in the bottom right panel of Figure 2, performance was most efficient in the token (M= 1.20, SD = .62) and the pen-paper conditions (M = 1.15, SD = .60) with the static (M = 1.05, SD = .71) and the pointing (M = 1.12, SD = .59) conditions being least efficient. However, these differences failed to reach significance, F (3,177) = 1.39, p = .247.

Attitude Toward the Task
The attitude of participants was more positive toward the pen-paper (M = 37.98, SD = 8.38) and the token (M = 37.78, SD = 8.94) conditions, than the pointing condition (M = 34.12, SD = 8.76) and least favourable for the static condition (M = 31.63, SD = 9.13). An overall main effect was found for attitude toward the task, F(3,117) = 17.07, p < .001, !2 = .231. Post-hoc tests further identified highly significant differences between the static and the token conditions and the static and pen-paper conditions (p < .001 for both conditions). The static and pointing conditions were also significantly different, p = .025.

Feelings toward the pointing condition differed significantly from those in the pen-paper (p < .001) and token conditions (p = .008). Therefore, these results indicated that participants preferred to use artefacts either pen and paper or tokens in calculating the solutions.

Discussion
The present experiment was designed to explore the effects of different levels of interactivity on arithmetic performances for single-digit additions. The degree of engagement with and attitude towards completing the task as a function of the level and nature of interactivity was also investigated. The results indicated that the use of artefacts enhanced performance in simple arithmetic problems, supporting the hypothesis that interactivity benefits performance. When participants were given the opportunity to use artefacts such as tokens, accuracy improved and deviation from the correct answer decreased. The increase in interactivity generally required no more time to announce an answer, confirming similar findings in previous research (Neth & Payne, 2011; Vallée-Tourangeau, 2013). Since accuracy increased with interactivity but latencies remained unchanged, interactivity therefore promoted more efficient problem solving. Conversely, when participants were asked to rely primarily on internal cognitive resources, as in the static condition, accuracy was impaired. In the higher interactivity conditions participants were given the opportunity to recruit external resources to aid in calculating the answer. The opportunity to engage with the environment enabled the distribution of cognitive load, augmented working memory resources and delegated the control of attention in part to the dynamic environment that cued the next action. In addition, the possibility of modifying the physical presentation of the problem, enabled participants to reconfigure the problem. This improved the cognitive congeniality of the problem (Kirsh, 1995), but also provided a more dynamic set of action affordances that supported more efficient problem solving. Thus the beneficial effect of interactivity on reasoning is not simply a function of off-loading working memory, but also reflects better executive function skills that are cued and prompted by the shifting affordances offered by a dynamic problem environment. The data on the participants’ attitude towards completing the sums in the different conditions paralleled the impact of interactivity on performance. Conditions involving external resources, pens or tokens, seemed to elicit a more positive, engaged attitude towards the simple arithmetic problems, than the restricted, static condition. Of course, participants were also more accurate in the interactive conditions. But the more positive attitudes towards the problems cannot be attributed to task success as such since participants were never given feedback about their performance, that is, after announcing each sum, the experimenter did not tell the participants whether their answer was right or wrong. Those findings are in line with the suggestion that higher levels of personal involvement positively affect performance (Shernoff et al., 2003). Also, changing the visual display may ease the task and thereby lighten the cognitive load, which

increases effectiveness and alters attitudes (ValléeTourangeau et al., 2013). In calculating simple arithmetic sums, an individual presented with the opportunity to use a complimentary strategy, such as manipulating tokens, is embedded in a distributed cognitive environment. Studying systems rather than individuals poses theoretical and methodological challenges. Theoretically, the nature of the problem representation and the trajectory of the solution as it evolves from an embryonic to a fully formed answer, should perhaps be understood as being distributed and configured in terms of a transaction between the participants’ internal resources and the shape and nature of the resources in the external environment. What a participant is ‘thinking’ is not independent of the state of the environment, and as the environment is shaped by the participants, understanding that environment is not independent from the participant. The methodological implications of this transactional perspective are important. Of course, systems can be more complex, and composed of a much wider range of functional elements, which challenge the traditional toolkit of experimental cognitive psychologists designed to deal with a cognitively sequestered individual in a laboratory environment that generally prevents interactivity. But beyond issues of complexity and computational promiscuity (Wilson & Clark, 2009), the dynamic meshwork of internal and external resources encourages a more qualitative idiographic cognitive science supported by an observational toolkit that can code at a much smaller time scale the evolution of a problem representation and its solution (for an excellent example of how such a toolkit can be developed, see Steffensen, 2013). Finally, adapting the cognitive psychologist’s laboratory to permit the physical manipulation of a problem presentation offers a more representative window onto thinking outside the laboratory. To be sure, people can simulate and think in their head without physically interacting with the outside world (although this internal cogitation may well reflect the internalization of much interactivity); but they often “go to extraordinary lengths to avoid having to resort to (…) fully environmentally detached reflection(s)” (Clark, 2010, p. 24, emphasis in the original). The data presented here reveals the importance of engineering task environments in the lab that support distributed problem representations to better understand the engagement of individuals as they explore and manipulate the external world to solve problems.

References
Ashcraft, M. H. (1995). Cognitive psychology and simple arithmetic: A review and summary of new directions. Mathematical Cognition, 1, 3-34. Ashcraft, M. H., & Moore, A. M. (2009). Mathematics anxiety and the affective drop in performance. Journal of Psychoeducational Assessment, 27, 197-205. Carlson, R. A., Avraamides, M. N., Cary, M., & Strasberg, S. (2007). What do the hands externalize in simple arithmetic? Journal of Experimental Psychology: Learning, Memory, and Cognition, 33, 747–756.

Clark, A. (2010). Material surrogacy and the supernatural: Reflections on the role of artefacts in ‘off-line’ cognition. In L. Malafouris and C. Renfrew (Eds.), The cognitive life of things: Recasting the boundaries of the mind (pp. 23-28). Cambridge: McDonald Institute for Archaeological Research. DeStefano, D., & LeFevre, J.-A. (2004). The role of working memory in mental arithmetic. European Journal of Cognitive Psychology, 16, 353386.European Journal of Cognitive Psychology, 16, 353-386. Goldin-Meadow, S., Alibali, M. W., & Church, R. B. (1993). Transitions in concept acquisition: Using the hand to read the mind. Psychological Review, 100, 279 - 297. Goldin-Meadow, S., Nusbaum, H., Kelly, S. D., & Wagner, S. (2001). Explaining math: Gesturing lightens the load. Psychological Science,12, 516-522. Kirsh, D. (1995). The intelligent use of space. Artificial Intelligence, 73, 31–68. Kirsh, D. (2013). Thinking with external representation. In S. J. Cowley, & F. Vallée-Tourangeau (Eds.), Cognition beyond the brain: Computation, interactivity and human artifice (pp. 171-194). London: Springer-Verlag. Neth, H., & Payne, S. J. (2001). Addition as Interactive Problem Solving. In J.D. Moore, & K. Stenning (Eds.), Proceedings of the Twenty-third Annual Conference of the Cognitive Science Society.(pp. 698703). Austin, TX: Cognitive Science Society. Neth, H., & Payne, S. J. (2011). Interactive coin addition : How hands can help us think. In J.D. Moore, & K. Stenning (Eds.), Proceedings of the Thirty-third Annual Conference of the Cognitive Science Society.(pp. 279-284). Austin, TX: Cognitive Science Society Newmann, F. M., Wehlage, G. G., & Lamborn, S. D. (1992). The signifiance and sources of student engagement. F. M. (Ed.) Student engagement and

schievment in American secondary schools (pp. 1139). New York : Teachers College Press Columbia University. Schiefele, U., & Csikszentmihaly, M. (1995). Motivation and ability as factors in mathematics experience and achievement. Journal for Research into Mathematics Education, 26, 163-181. Shernoff, D. J., Csikszentmihaly, M., Schneider, B., & Shernoff, E. S. (2003). Student engagement in high school classrooms from the perspective of flow theory. School Psychology Quarterly, 18, 158-176. Steffensen, S. V. (2013). Human interactivity: Problemsolving, solution probing and verbal patterns in the wild. In S. J. Cowley, & F. Vallée-Tourangeau (Eds.), Cognition beyond the brain: Computation, interactivity and human artifice (pp. 195-221). London: Springer-Verlag. Vallée-Tourangeau, F., Sirota, M., & Villejoubert, G. (2013). Reducing the impact of math anxiety on mental arithmetic: The importance of distributed cognition. Proceedings of the Thirty-Fifth Annual Conference of the Cognitive Science Society (pp. 3615-3620). Austin, TX: Cognitive Science Society. Vallée-Tourangeau, F. (2013). Interactivity, Efficiency, and individual differences in mental arithmetic. Experimental Psychology, 60, 302-311. Vallée-Tourangeau, F., Euden, G., & Hearn, V. (2011). Einstellung defused: Interactivity and mental set. The Quarterly Journal of Experimental Psychology, 64, 1889–1895. Weller, A., Villejoubert, G., & Vallée-Tourangeau, F. (2011). Interactive insight problem solving. Thinking & Reasoning, 17, 429–439. Wilson, R. A., & Clark, A. (2009). How to situate cognition: Letting nature take its course. In P. Robbins and M. Aydede (Eds.), The Cambridge handbook of situated cognition (pp. 55-77). New York: Cambridge University Press.

! (forthcoming in Synthese)

"!

KNOWLEDGE AND COGNITIVE INTEGRATION
S. Orestis Palermos University of Edinburgh

Abstract: Cognitive integration is a defining yet overlooked feature of our intellect that may nevertheless have substantial effects on the process of knowledge-acquisition. To bring those effects to the fore, I explore the topic of cognitive integration both from the perspective of virtue reliabilism within externalist epistemology and the perspective of extended cognition within externalist philosophy of mind and cognitive science. On the basis of this interdisciplinary focus, I argue that cognitive integration can provide a minimalist yet adequate epistemic norm of subjective justification: so long as the agent’s belief-forming process has been integrated in his cognitive character, the agent can be justified in holding the resulting beliefs merely by lacking any doubts there was something wrong in the way he arrived at them. Moreover, since both externalist philosophy of mind and externalist epistemology treat the process of cognitive integration in the same way, we can claim that epistemic cognitive characters may extend beyond our organismic cognitive capacities to the artifacts we employ or even to other agents we interact with. This move is not only necessary for accounting for advanced cases of knowledge that is the product of the operation of epistemic artifacts or the interactive activity of research teams, but it can further lead to interesting ramifications both for social epistemology and philosophy of science.

1. INTRODUCTION Cognitive integration is an overlooked yet defining feature of our intellect. It is the reason why, in contrast to mere stimulus-response automata, we entertain advanced beliefs in an epistemically responsible way, and why we can do so even in the complete absence of any reasons to back those beliefs up. Nevertheless, within philosophy of mind and cognitive science, it is only recently that the topic of cognitive integration was brought into focus. This recent change of focus, however, has not been and is still not guided by an attempt to understand how our complex brain capacities intertwine with each other. Due to Fodor’s (1983) persuasive understanding of our intellectual

#! ! architecture as modular, in-the-head cognition has largely thought to be for the most part domain specific and informationally encapsulated.1 Consequently, cognitive scientists who take the brain to be the primary cognitive explanandum have so far been indisposed to explore and understand the phenomenon of cognitive integration. Instead, only after the hypothesis of extended cognition (Clark & Chalmers 1998) was proposed (according to which external elements can be proper parts of our cognitive systems), did the topic of cognitive integration start to attract attention; claiming that artifacts can be parts of an agent’s cognitive system presupposes an account of how such external elements can be properly integrated into our cognitive loops. Accordingly, over the past fifteen years, several attempts have been made to account for the process of cognitive integration, most of which rely either on a common-sense functionalist understanding of our minds (Clark & Chalmers 1998; Clark 2010), or on a mathematically inspired approach (Cark 2008; Chemero 2009, Palermos 2014) that focuses on the dynamical nature of cognitive processes. Interestingly, however, the topic of cognitive integration has also recently emerged within epistemology in reference to the concept of our epistemic cognitive characters. Even within epistemology, however, the discussion of cognitive integration has not been entirely unrelated to the idea of cognitive extension. Admittedly, Greco (2010)—the first to discuss (in passing) the idea of cognitive integration in epistemological terms—does not commit himself to the possibility of cognitive extension. Both Pritchard (2010) and I (Palermos 2011), however, have noted that the notion of the epistemic agent’s cognitive character, to which all of the agent’s knowledge-conducive belief-forming processes must have been properly integrated, is open to an interpretation along the lines suggested by the extended cognition hypothesis. However, and despite these few efforts to elucidate the process of cognitive integration within epistemology, more needs to be said since the effects of this process on knowledge could turn out to be surprisingly substantial. To bring these effects to the fore, we need to approach the idea of cognitive integration both from the perspective of virtue reliabilism in externalist epistemology (section 2.2) and the perspective of extended !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Despite the influence of Fodor’s work within philosophy of mind and cognitive science, his modular understanding of the mind has met some considerable resistance by equally influential philosophers. One clear example is the work of Churchland (1979, 1988, 1989).
1

! cognition in externalist philosophy of mind (section 2.3). This interdisciplinary focus will help us make apparent, in section 2.3, that both

$!

views treat cognitive integration in essentially the same way—an observation that we can then use to bring both perspectives together and thereby demonstrate how our epistemic cognitive characters can extend beyond our organismic cognitive capacities to the epistemic artifacts we employ. This combination of externalist epistemology with externalist philosophy of mind, however, as I will argue in section 3, is not just an available option, but is actually necessary for accounting for advanced cases of knowledge whereby one’s true believing is the product of the operation of epistemic artifacts. Furthermore, this approach can generate interesting ramifications both for social epistemology and philosophy of science, wherein the pursuit of knowledge has been traditionally associated with the use of artifacts, carefully tailored labs, and the combined efforts of epistemic agents working in research teams. To start with, however, a few introductory remarks are first in order. 2. THE PROCESS OF COGNITIVE INTEGRATION 2.1 Introduction “John is sad” is an assertion that can be true or false. In producing this claim, of course, perception will always be foundational in a certain way, but in order to make, as well as check, the validity of the claim we combine several other processes as well. In the background of our minds we have a biological as well as a primordial (animalistic) conception of what a human being is and we also have a socio-contextual sense of who “John” is; we have a theory of mind that enables us to understand other people with thoughts and emotions like ours, and we may even need a more personal theory of John’s character: John has spent the entire night socializing at the pub and he just made a witty joke, but the downwards motion of his eyes accompanied by a melancholic smile at the end of his remark are enough to indicate to his close friends that his mind is secretly occupied with his recent loss. It is in a sense like this that theories (commonsensical as well as scientific ones) have a top-down effect on the bottom-up input we receive

%! ! from our sense-apparatuses. This overriding effect, known as the theoryladeness of observations, has been well-studied both within philosophy of science (Kuhn 1962; Hanson 1961; 1969) and philosophy of mind (Churchland 1979; 1988; 1989; Fodor 1984; 1988), and indicates that perceptual beliefs are never ‘purely perceptual’, but they instead emerge out of the cooperation of several intellectual capacities operating in tandem. Now, considerations like these indicate that perception (traditionally thought of as the most foundational aspect of our belief systems) may not be all that basic, and have thereby hinted towards a relativistic picture of science—and possibly epistemology as well—whose most prominent proponents are thought to be Feyerabend (1975) and Kuhn (1962) (the latter quite possibly unjustly though). Fortunately, however, at least as far as the theory-ladeness of observation is concerned, recent studies within cognitive psychology not only point away from relativism, but they also demonstrate how the interaction between our theoretical beliefs and our sensory apparatus has a facilitatory effect on the overall process of perception. For example, exploring the literature on relevant experiments, Brewer and Lambert (2001) concede that “perception is determined by the interaction of top-down theory information and bottom-up sensory information” (178, my emphasis):
However, note that in all of the above cases the stimuli were either ambiguous, degraded, or required a difficult perceptual judgment. In these cases the weak bottom-up information allowed the top-down influences to have a strong impact on perceptual experience. It seems likely that strong bottom-up information will override top-down information. [...] Thus, the topdown/ bottom-up analysis allows one to have cases of theory-laden perception, but does not necessarily lead down the slippery slope of relativism (ibid.).2

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
2

Similarly, Estany (2001, 208) holds that the beliefs of the higher or more fundamental level influence how perceptual units are interpreted by the lower levels [...] Humans use both types of processes in perception because each have characteristic advantages and disadvantages. Thanks to top-down processes we can recognize patterns with incomplete or degraded information. Moreover, top-down processes make perception faster, but they can induce us to make mistakes in a perception by relying on previous knowledge.

Nevertheless, Estany further notes that even though our perceptual systems get guidance from higher-order expectations, when attention is caused by the mismatches between expectation and reality, the inputs from the arousal system constitute a “reset wave” making it possible to avoid arbitrary relativistic errors of perception (Estany 2001, 213).

! Therefore, we should neither be misguided nor feel intimidated by the possibly false impression of relativistic bias that the integration of our

&!

intellectual capacities into one overarching belief-generating system may have initially created. To the contrary, we should focus on the phenomenon of cognitive integration because, as we shall see, it actually has a very important facilitatory effect on our epistemic standing. In particular, the interconnectedness of our cognitive capacities allows us to be subjectively justified even if we lack explicit reasons for holding our beliefs. Provided that we act in a conscientious mode—i.e., provided we are motivated to believe what is true—cognitive integration allows us to trust the deliverances of our cognitive abilities by merely lacking any negative reasons against (as opposed to possessing positive reasons for) our beliefs. This sense of epistemically adequate—yet unreflective—cognitive responsibility can only be achieved by agents like us, whose intellectual capacities are appropriately interconnected such that in cases where there is something wrong with the way we form our beliefs or with the beliefs themselves, we will be able to notice this and respond appropriately. Otherwise—if there is nothing wrong—we can go on about with our daily activities without questioning our epistemic standing with respect to every single of the millions (possibly billions?) of beliefs we enjoy in the course of our days. On a first pass, this probably sounds sketchy, but focusing on contemporary epistemology should allow us to both understand what this unreflective sense of cognitive responsibility amounts to and why it is so important. 2.2 Cognitive integration in epistemology To start with, consider epistemic internalism, which takes an approach to epistemic responsibility that is very different from the one suggested here. According to traditional forms of internalism, one should always be able, at least in principle, to access the reasons that justify one’s beliefs, by reflection alone.3 This may initially sound as a reasonable demand, but the problem is that it creates serious complications with respect to our perceptual and empirical beliefs. Specifically, it poses the requirement that there be necessary
3 For classical defenses of this view see Chisholm (1977) and Bonjour (1985, ch. 2). See also Steup (1999), Pryor (2001, §3), Bonjour (2002), Pappas (2005), and Poston (2008).

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

'! ! support relations between one’s empirical and perceptual beliefs and one’s evidence for holding them (such that one can be in a position to justify one’s empirical and perceptual beliefs by reflection alone). As Hume’s problem of induction demonstrates, however, this is impossible. Accordingly, it has been traditionally assumed that Hume’s arguments lead to skepticism about our empirical knowledge.4 Contemplating on the Humean problematic, however, Greco (1999) argues that this is too fast. Hume’s arguments should not be directly considered as skeptical ones. Instead, the immediate conclusion to be drawn from them is that there are no necessary support relations between our empirical beliefs and their evidence; that if the evidence for our empirical beliefs is reliable, then it is at most contingently reliable. This realization alone, however, cannot automatically lead to skepticism. Only after we embrace the internalist understanding of knowledge, such that there always be necessary support relations between one’s evidence and one’s beliefs do we face skepticism. In other words, in order to avoid skepticism about empirical and perceptual knowledge, we must allow knowledge to be grounded on evidence that is merely contingently reliable, and so we must give up the requirement that one’s beliefs should always be internally—i.e., by reflection alone—justified. Any adequate epistemology must be able to account for the fact that merely contingently reliable evidence can give rise to knowledge (Greco 1999, 273). Now, in order to accommodate the above realization, contemporary epistemologists have put forward process reliabilism; viz., the idea that knowledge is true belief that is the product of reliable belief-forming processes, where a reliable process is a process that results in a
The problem of induction is well known. We form our beliefs about unobserved matters of fact and the external world on the basis of evidence provided by past and present observations and sensory appearances, respectively. In order, however, for the support relations between our empirical and perceptual beliefs and the evidence offered in their support to be necessary, we also need the further assumptions that the future will resemble the past and that sensory appearances are reliable indications to reality, respectively. The problem, however, is that both of these assumptions rely for their support on what they assert. Consequently, given that circular reasoning is invalid, there are no necessary support relations between our empirical beliefs and the evidence offered in their support. Accordingly, the conclusion that has been traditionally drawn is that our empirical and perceptual beliefs cannot amount to knowledge. For more details on a reconstruction of Hume’s skepticism along these lines, see (Greco 1999).

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
4

! preponderance of true over false beliefs. Moreover, in direct response to the Humean problematic, this approach “denies that one must know that one’s evidence is reliable”, by making “de facto reliability the grounds of positive epistemic status” (Greco 1999, 284-5). Accordingly, process reliabilism is an externalist approach to knowledge, because—contrary to the traditional

(!

account of knowledge as internally justified true belief—on this view, in order to know, one does not need to know, be justified in believing (by reflection alone, or any other means), or even believe that one’s beliefs are formed in a reliable fashion. So long as one employs an objectively reliable process, one is justified in holding the resulting belief. Process reliabilism, therefore, has the resources to overcome the Humean skepticism. There are, however, two serious complications with the view. The first one is that process reliabilism, as it stands, is too weak a condition on knowledge because it allows any reliable belief-forming process to count as knowledge-conducive, and this, as we shall see, is intuitively incorrect. The second complication is that by making “de facto reliability the grounds of positive epistemic status” process reliabilism misses a very important dimension of our epistemic nature. While it is true that in order to know we do need the way of forming our beliefs to be objectively reliable, this sort of objective justification is not sufficient in its own. What we further need is that we be subjectively justified in the sense that we must be somehow sensitive to the reliability of our evidence.5 Process reliabilism, however, ignores this dimension of our epistemically sentient nature altogether, to the extent that it has been even criticized that it equates us to mere stimulusresponse automata (Fuller 2012).6 We can better appreciate these two problems by taking a look at a few (eccentric, yet informative) examples. Consider Hercules first:

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Remember, however, that if, as Hume’s skeptical arguments demonstrate, the relation between evidence and belief is not necessary (see also fn. 4), then it is far from obvious how a person can be subjectively justified, especially in externalist approaches such as process reliabilism. If a condition of ‘subjective sensitivity to the reliability of one’s evidence’ must be satisfied, then this should better be accomplished in a way that will not require knowledge of or even beliefs about the said reliability (otherwise Hume’s skepticism will strike back). 6 “Epistemic zombies” would probably be the name that David Chalmers would give to such creatures. Given, however, the present discussion I don’t think they could really exist.
5

)! !
Hercules (Adapted from Pritchard’s Temp (2009, 48)) Hercules tosses a drachma whenever he wants to form a belief about the weather outside. If it is heads, he forms the belief that it is sunny; if it is tails he believes it is cloudy; and if it balances in between, he believes it is rainy. As it happens, Hercules’ way of forming his weather beliefs is perfectly reliable, because Zeus, who wants to save Hercules from the embarrassment of forming false weather beliefs, has an eye on him; every time he sees Hercules tossing the coin arranges the world accordingly.

Hercules’ beliefs are formed in a highly reliable way. So, according to process reliabilism, Hercules has knowledge of the weather conditions. Intuitively, however, this is incorrect. There is a problem with the direction of fit between his beliefs and the facts. In cases of knowledge, we want our beliefs to be true because they correspond to the facts, and not because the facts comply with our beliefs; when one knows, one’s true beliefs are about the world, not the other way around. In Hercules’ case, however, his beliefs are not true because they are formed in a way that detects the facts. Instead, he first forms his beliefs in an arbitrary way—he makes no efforts to ensure they will come out true—and then Zeus takes over so that the facts will comply with Hercules’ beliefs. This, however, is not knowledge; it is the ‘luck of the gods’. If one day Zeus had a fight with Hera, Hercules’ beliefs would cease coming out true. Notice, however, that if Hercules used his cognitive abilities—say by taking a look at the sky—to form his weather beliefs, then he would not run into any such problems. If he didn’t form his beliefs in an arbitrary way, but on the basis of his cognitive abilities, he would not need Zeus to tweak the world so that his beliefs could systematically turn out true. If one’s beliefs are the product of one’s cognitive abilities, then if they turn out to be true it will be because they are sensitive to the facts; the direction of fit will be the correct one. So, it may be proposed that the way to restrict the reliable beliefforming processes to those that get the direction of fit correctly—such that they can be knowledge-conducive—is to identify them with one’s cognitive abilities, or, in other words, with those processes that can be intuitively thought of as cognitive ones. But can all prima facie cognitive processes count as cognitive abilities and thereby produce knowledge? The answer, as we shall now see, must be a negative one, because there are reliable processes that we might be inclined to categorize as cognitive ones, but which fail to

! deliver knowledge, exactly because they disallow the agent to be subjectively

*!

justified in employing them. Think about the Serendipitous Brain Lesion first:
Serendipitous Brain Lesion (Greco 2010, 149) Suppose that S has a rare brain lesion, one effect of which is to reliably cause the true belief that one has a brain lesion. Even if the process is perfectly reliable, it seems wrong that one can come to have knowledge that one has a brain lesion on this basis.

Again, the unfortunate agent’s way of forming his true belief about the brain lesion on the basis of his brain lesion is reliable. Process reliabilists, therefore, must accept that he can gain knowledge in this way. As Greco claims, however, this does not sound correct. Why not? Mainly because the way the agent forms his belief is so strange from his point of view that he cannot accept he can gain knowledge in this way (Greco 1999, 2010). Accordingly, there might be reliable in-the-head processes (such that we may be inclined to call them cognitive ones) that we wouldn’t like to claim they are knowledgeconducive cognitive abilities, because from the agent’s point of view they are strange. More precisely, the underlying intuition here is that for a process to be eligible to count as a cognitive ability it must not be strange, in the sense that it must not be at odds with the rest of the agent’s cognitive system.7 The reason is that if the process is strange, then, in light of the rest of his cognitive system, the agent will reject both the process and its deliverances despite the fact that they are in fact reliable—from the agent’s point of view, they aren’t. So, in order for a process to be a candidate for qualifying as a cognitive ability such that it can be knowledge-conducive it must not be inconsistent with the rest of the agent’s beliefs and his methods of producing them. In other words, it must be such that it can become part of, or be integrated into, the rest of the !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

7 An anonymous referee points out that strangeness is description-relative. Take vision for example. We are all familiar with acquiring knowledge through seeing things. But learning about the physiological and neural underpinnings of vision will surely seem strange to some; couldn't such a person say "This is really strange, and I don't really see how it works, but, I guess, this is how I know the color of my shoes"? I think this is right, but this example wouldn’t be problematic for the following two reasons. First, even though the explanation may seem ‘strange’ to the agent (in the sense of being difficult to understand) it is not ‘at odds with the rest of the agent’s cognitive system’. Second, the requirement that the process not be strange does not refer to a reflective-explanatory understanding of the process (as in the referee’s example), but to the presence of the process itself. Think about the analogy of a strange (i.e., eccentric) person who is nevertheless not a stranger: the requirement that the process not be strange allows for the process to be ‘strange’ in the first sense, but not in the sense of being a ‘stranger’. I am thankful to the referee for bringing this ambiguity to my attention.

"+! ! agent’s cognitive system (Greco 2010, 152). And clearly, this is not the case with the serendipitous brain lesion. The process is a cognitive malfunction, and even more crucially, its output is so odd that no epistemic agent could accept as true. In other words, the serendipitous brain lesion cannot count as knowledge-conducive because it is so strange that it cannot become part of the rest of the agent’s cognitive system and so cannot count as a cognitive ability.8 Nevertheless, even a reliable process that is normal enough to become part of the agent’s cognitive system cannot yet count as a cognitive ability that can produce knowledge. Consider a further example:

Careless Math Student (Greco 2010, 149) Suppose that S is taking a math test and adopts a correct algorithm for solving a problem. But suppose that S has no understanding that the algorithm is the correct one to use for this problem. Rather, S chooses it on a whim, but could just as well have chosen one that is incorrect. By hypothesis, the algorithm is the right one, and so using it to solve the problem constitutes a reliable process. It seems wrong to say that S thereby knows the answer to the problem, however.

The careless student’s algorithm for solving the problem is also reliable. But again, we cannot attribute knowledge to her. Why not? The reason is that she employed the right method on a whim, such that she could have very easily employed another, incorrect method. Her reliable process is a fleeting one. It is not a habit or a disposition of hers. Given the same circumstances, she could have employed an inappropriate method, thereby, ending up with a falsehood. If, however, the student had habitually invoked the correct
8 An anonymous referee insists that the brain lesion can yield knowledge, thereby implying that process reliabilism is a sufficient condition on knowledge. While it may be possible to make the case for the sufficiency of process reliabilism, the orthodox view within mainstream epistemology goes against this prospect. For classical rejections of the sufficiency of process reliabilism on the basis of thought experiments very similar to the Serendipitous Brain Lesion, see Bonjour (1980), Lehrer (1990) and Plantinga (1993b). The main idea is that reliability might be necessary for knowledge but what is further required is satisfaction of the internalist intuitions with respect to the possession of subjective justification (as we mentioned above, one of the problems for process reliabilism is that by making de facto reliability the grounds of positive epistemic status, it fails to capture the intuition that, somehow, we must also be sensitive to the reliability of our evidence). Internalists, typically require the possession of reflectively accessible reasons for said reliability. Here, following Greco’s intuitions (1999; 2010, 149-155), we opt for a weaker condition of subjective justification according to which the agent must lack beliefs against the reliability of his beliefforming process, where the process being strange from the agent’s point of view would count as just one such defeating belief. For very similar intuitions on how the strangeness of the origin of the relevant beliefs acts as a defeater in Bonjour (1980) and Lehrer’s (1990) thought experiments see Goldman (1986, 111–112).

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

!

""!

algorithm when the problems called for it, then we would indeed be inclined to claim that she could gain knowledge on its basis. The reason for this is that if a process is a disposition or a habit of the agent, then the agent will be able to become aware of the circumstances in which it can become unreliable. Otherwise, it seems arbitrary that the agent employed it in an appropriate, but isolated case, and so cannot gain knowledge on its basis. In other words, a reliable process that is normal—such that it can, in principle, become part of the rest of one’s cognitive system—won’t be a candidate for qualifying as a cognitive ability, unless it is also a disposition or a habit of the agent. Why is this so? The intuition is that abilities, in general, are habits or dispositions possessed by agents.9 But apart from such intuitions we have also noted that in order for a reliable process to count as a cognitive ability it must be such that it can become part of (or be integrated into) the rest of the agent’s cognitive system. One requirement for this, we have noted, is that the process not be strange such that it won’t be inconsistent with the rest of the agent’s cognitive system. What is further required, however, is that it also be coherent with her cognitive system in the following sense: The agent must be able to become aware that the process is unreliable in certain circumstances, because this will allow her to non-accidentally endorse its deliverances in the rest of the circumstances, even if she lacks any positive beliefs for its reliability. And in the absence of any explicit reasons that are accessible through reflection alone— recall the Humean problematic—the only realistic way for the agent to so

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
An anonymous referee is worried that I should not use ‘dispositions’ and ‘habits’ as synonyms. Specifically, not all dispositions are habits; someone or something, for example, may be disposed to act in a certain way—should the appropriate conditions obtain—even if the relevant person or thing has never behaved in that way before. Accordingly, the worry further goes, the fact that a cognitive ability is a disposition does not mean it will also be a habit. In response, even though it is true that in one sense of the term, ‘dispositions’ are not always going to be habits, there is another sense of the term that they are; according to this second sense of the term, to claim that cognitive abilities are dispositions means that abilities are character traits, or habitual behaviors that the agent tends to exhibit. A strong indication that this is how we should understand the dispositional nature of cognitive abilities has to do with the fact that abilities can only be acquired and sustained through practice, whereas dispositions, in the other meaning of the term, can be possessed by an entity even if they are never actually manifested (e.g., a vase may be fragile even if it has never been broken). As we shall see below, Greco appears to concur with this understanding of abilities as he claims they are the stable traits of the agent’s cognitive character; a behavior can be in character only if it is habitually manifested. See also (Greco 2010, 150). I am thankful to the referee for pressing this point.
9

"#! ! become epistemically responsible in employing a process is that it be a disposition or habit of hers.10,11 So to summarize what we have gathered from the three examples above, in cases of knowledge, we want one’s beliefs to be responsive to the facts. Accordingly, we claimed that only prima facie cognitive processes can be knowledge-conducive, but not all of them will do. The process must be a cognitive ability, meaning that the agent will be able to be conscientious and thereby subjectively justified in employing it. And in the absence of positive reasons (that are accessible through reflection alone) for the reliability of the cognitive process, this last condition can be satisfied in a realistic way only if the relevant process is a normal disposition or habit of the agent such that it can become part of (or be integrated into) the rest of her cognitive system. Notably then, the general idea, which all the above considerations are alluding to, is that for a reliable process to be knowledge-conducive it must be a cognitive ability. This idea has also appeared in the literature as the ability intuition on knowledge and can be summarized by stating that knowledge is belief that is true in virtue of cognitive ability.12 Now the end result of the above considerations is that they fill in the details of which reliable processes may plausibly count as cognitive abilities by demonstrating that it is only normal, dispositional or habitual cognitive processes that the agent can be subjectively justified in employing (and thereby able to gain knowledge from). Now, building on considerations very similar to the above ones, Greco (1999) has proposed a virtue reliabilist account of knowledge, which emphasizes that when we assess whether some agent knows, we shouldn’t be focusing on the reliability of isolated (cognitive) belief-forming processes, but on the reliability of the overall agent, conceived of as a stable, interconnected system of such belief-forming processes. 13 It is this interwoven totality of
I here say ‘the only realistic way’ because we can imagine, for instance, a case of a benevolent mentalist who hypnotizes the agent to trust a newly acquired process, and trust it only in the appropriate conditions (thereby allowing him to be epistemically responsible in employing it), despite the fact that the process is not a disposition of hers. 11 For further discussion of the above intuitions on the Serendipitous Brain Lesion and Careless Math Student cases see (Greco 1999) and (Greco 2010, 149-155). For the discussion of similar thought experiments and intuitions see (Bonjour 1980), (Goldman 1986), (Lehrer 1990), and (Pantinga 1993b). 12 The idea that knowledge must be grounded in cognitive abilities can be traced back to the writings of Sosa (1988; 1993) and Plantinga (1993a). For more recent approaches to this intuition, see Greco (1999; 2004; 2007) and Pritchard (2009; forthcoming; 2010a; 2010b). 13 Any theory of knowledge that places in its center the ability intuition on knowledge will fall under the general trend of Virtue Reliabilism (abilities are normally understood as virtues

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
10

!

"$!

cognitive abilities that give rise to one’s sense of epistemic self (or in Greco’s terms to one’s ‘cognitive character’), and which should be the focus of our epistemic assessment. To make apparent the motivation for this change of epistemic focus, remember that in order to avoid the Humean problematic we must accommodate subjective justification in a way that does not involve knowledge or even beliefs about reliability. At this point, Greco (1999, 289) suggests that a promising strategy for doing so is to claim that “a belief p is subjectively justified for a person S (in the sense relevant for having knowledge) if and only if S’s believing p is grounded in the cognitive dispositions that S manifests when S is thinking conscientiously” (i.e., when S is motivated to believe what is true). In this way, the agent will employ his reliable cognitive processes only in circumstances that have not been problematic in the past, and he will be able to do so without even having any beliefs about their reliability.14 Greco, then, goes on to further claim that the dispositions/habits that a person manifests when she is thinking conscientiously intertwine with each other and give rise to what we may call one’s cognitive character (Greco 1999, 290). So, overall, “a belief p has a positive epistemic status for a person S just in case S’s believing p results from the stable and reliable dispositions that make up S’s cognitive character” (ibid., 287-8). !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
and vice versa). To accentuate the features of his account, Greco calls his view Agent Reliabilism, but it is clearly a version of Virtue Reliabilism—one that emphasizes the importance of the overall agent in the manifestation of the relevant intellectual virtues. For alternative, robust as well as weaker formulations of Virtue Reliabilism see Sosa (1993; 2007) and Pritchard (forthcoming; 2010a; 2010b), respectively. 14 The fact that people manifest highly specific, finely tuned dispositions to form their beliefs in certain ways but not in others amounts to an implicit awareness of the reliability of those dispositions. For example suppose that it seems visually to a person that a cat is sleeping on the couch, and on this basis she believes that there is a sleeping cat on the couch. Suppose also that this belief manifests a disposition that the person has, to trust this sort of experience under these sorts of conditions, when motivated to believe the truth. Now, suppose that much less clearly, it seems visually to the person that a mouse has run across the floor. Not being disposed to trust this kind of fleeting experience, the person refrains from believing until further evidence comes in. The fact that the person, properly motivated, is disposed to trust one kind of experience but not the other, constitutes sensitivity on her part that the former is reliable. There is a clear sense in which she takes the former experience to be adequate to her goal of believing the truth, and takes the latter experience not to be. And this is so even if she has no beliefs about her goals, her reliability, or her experience (Greco 1999, 290 ) . A similar argument can be found in (Sosa 1993, 60-63).

"%! ! In this way, Greco can do away both with strange and fleeting processes. Strange processes cannot be part of the agent’s cognitive character because they are not the kind of processes that a conscientious agent would employ. Fleeting processes are also excluded. First, because they are not dispositions or habits—so they cannot really count as character traits. And, second (I may add), because, in the absence of reasons to believe that the relevant process is reliable, it is only dispositions or habits that one can become aware they are unreliable in certain circumstances and, so—without relying on any beliefs about their reliability—use them conscientiously in the rest of the circumstances.15 So in order to gain knowledge on the basis of a process the agent must be able to employ a conscientious attitude towards that process. And in order for that to be the case, the relevant process must be a cognitive ability of the agent, meaning that it must have been integrated into the agent’s cognitive character. Now, despite the previous points on the importance of the normality and dispositionality of the relevant process in order for it to count as a genuine part of one’s cognitive character, Greco attempts to further accentuate and shed some light on the integrated nature of our cognitive characters by noting that the process of “cognitive integration is a function of cooperation and interaction, or cooperative interaction with other aspects of the cognitive system” (2010, 152). So, how exactly should we think about the required conditions for a process to count as knowledge-conducive? In general, every knowledge-conducive process must be a cognitive ability such that the agent will be subjectively justified in employing it, which requires that the process be integrated into the agent’s cognitive character by cooperatively interacting with it. Accordingly, we may say that the only necessary and sufficient condition for a process to count as knowledgeconducive is that it cooperatively interacts with the rest of the agent’s cognitive character. Now, apparently, this makes the normality and dispositionality criteria seem redundant—which strictly speaking they are— but they may still have a role to play; normality and dispositionality of the relevant process seem to be practical preconditions for the agent to be able to !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Apart from the example given in the previous footnote, Greco has not attempted to provide an account of how the process of subjective justification works. I assume, however, that he wouldn’t reject this falsificationist approach, as I cannot see how else subjective justification could be accommodated in an externalist way.
15

!

"&!

cooperatively interact with it. The extent, however, to which each one of these criteria may need to be satisfied will differ from case to case. An agent, for example, may be subjectively justified in employing a process in the appropriate conditions not because it is a normal disposition of hers but because a benevolent mentalist hypnotized her to do so. In most realistic cases, however, normality and dispositionality will still have a significant guiding effect. The decisive effect of cognitive integration, however—on the basis of which the agent can be conscientious and thereby subjectively justified—will only be ensured if the agent’s cognitive character mutually interacts with the relevant process. So, all we need to accept that a process is knowledge-conducive is that it be integrated into the agent’s cognitive character by cooperatively interacting with it. On the basis of this mutual interaction with the rest of his cognitive system the agent will be able to employ the relevant process conscientiously by merely lacking any beliefs that it is unreliable (at least not in the circumstances in which he employs it), or if the employment is involuntary, conscientiously accept its deliverances when he lacks any beliefs that the conditions that gave rise to them were unreliable. Moreover, as we saw previously in the discussion of the three examples, this process of cognitive integration gives rise to a coherentist effect both on the level of processes (how the beliefs are generated) and on the level of content (how the beliefs themselves combine). Also, it ensures that at least directly related belief-generating mechanisms and their resulting beliefs will be consistent. Overall then, the epistemic importance of cognitive integration is that it allows epistemic agents to satisfy the condition of subjective justification in a minimalist way, which is nevertheless sufficient for acquiring knowledge even if one cannot—not even in principle—offer any explicit reasons in favor of one’s beliefs. Stated explicitly, this minimalist condition of subjective justification is that conscientious epistemic agents ought to accept the deliverances of, and employ their cognitive abilities, only when they lack any doubts that they are unreliable, given the conditions they employ them in.16 In !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
As an anonymous referee has pointed out I should make clear that this condition should be restricted to reliable processes. We should not allow, for example, to an agent who forms his beliefs on the basis of astronomical considerations or wishful thinking to count as subjectively justified merely by lacking any doubts about the unreliability of his belief-forming processes.
16

"'! ! other words, cognitive integration allows epistemic agents to be conscientious in the sense that in cases where there is something wrong with the way they form their beliefs they will be able to spot this and respond appropriately. Otherwise, they can go on formulating their beliefs, without worrying whether they can actually offer reasons for every single one of them or for their reasons for holding them.17 2.3. Cognitive integration in epistemology and philosophy of mind In the introduction, I mentioned that the topic of cognitive integration has been explored in epistemology also in reference to the possibility of cognitive extension. The general idea is that the knowledge-conducive cognitive abilities and their relation to one’s cognitive character as discussed within virtue epistemology is particularly apt for an interpretation along the lines suggested by the hypothesis of extended cognition. To make this clear it should be helpful to repeat, for the last time, what are the important features of a knowledge-conducive belief-forming process. In general, the process must be a cognitive ability. In order for that to be the case the process must be a cognitive process. This will guarantee the correct direction of fit between the belief and the fact. Also, we want the belief-forming process to be objectively reliable, where a reliable process is !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Given, however, that the epistemic agent must be conscientious (i.e., motivated to believe what is true) this qualification is actually redundant. If the agent is motivated to believe what is true, he will not employ astronomical considerations or wishful thinking because he will have noticed that such processes were notably unreliable in the past. For the same reason, they won’t even be parts of his (conscientious) cognitive character. 17 Further to footnotes 14 and 15, in providing this sort of account of subjective justification, I have relied for the most part on phenomenological intuitions about how we seem to go about our beliefs in everyday life. Nevertheless, such phenomenological intuitions seem to already entertain a certain degree of scientific support. Specifically, within cognitive psychology, there have been several studies indicating that subjects engage in analytic reasoning only when they experience the metacognitive effect of the lack of ‘fluency’: “Fluency is not a cognitive operation in and of itself but, rather, a feeling of ease associated with a cognitive operation, it can be generated by nearly any form of thinking. If a percept is blurry, we are aware that it was hard to see. If a word is phonemically irregular, we recognize the challenge in processing it. We know whether we had to struggle to bring a memory to mind and whether we had a hard or easy time solving a riddle. Because the metacognitive experience of fluency can be generated by so many cognitive processes and is nearly effortless to access, it can serve as a cue toward judgments in virtually any situation”. (Oppenheimer forthcoming). For an overview on the metacognitive feeling of fluency see (Oppenheimer forthcoming), (Alter & Oppenheimer 2007) and (Unkelbach & Greifeneder 2013).

!

"(!

one that tends to produce true rather than false beliefs. Recall, however, that according to reliabilism, the agent does not need his evidence to be necessarily reliable, such that he can be internally justified in holding his belief; if forming a belief on a certain kind of evidence constitutes a reliable belief-forming process, it does not matter that one’s evidence is only contingently reliable; the agent, on his part, does not need to know or even have any beliefs about the reliability of his way of forming beliefs. Instead, the agent can be subjectively justified simply by forming his belief on the basis of a process that is integrated into his cognitive character, which he employs when he is thinking conscientiously. Now, in order for a process to be a candidate for inclusion to the agent’s conscientious cognitive character, we noted that it will probably have to be neither strange nor fleeting. In most realistic scenarios the process (1) will have to be normal so that the agent won’t reject it when conscientious and (2) will have to be a disposition or a habit of the agent, because (barring scenarios such as the mentalist case) it is only dispositions or habits that one can become aware they are unreliable in certain circumstances, and so—without relying on any beliefs about their reliability—be able to employ them conscientiously in the rest of the circumstances. As we further noted, however, even though normality and dispositionality will, in most cases, be practical preconditions, they are neither necessary nor sufficient for a process to count as integrated into the agent’s cognitive character. Instead, the only thing that is required is that the process be integrated into the agent’s cognitive character, by engaging in cooperative interaction with the rest of the agent’s cognitive system. Accordingly, no matter what the practical preconditions for this interactive process to be achieved are, once it is in place, it will guarantee both that the relevant beliefforming process is a cognitive process, and that it is indeed part of the agent’s cognitive character such that he can be conscientious in employing it. So putting all the above points together: a belief-forming process counts as a cognitive ability and thereby as knowledge-conducive if and only if it is a reliable belief-forming process that is integrated into the agent’s cognitive character, on the basis of a process of cooperative interaction with it.18 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Many externalist epistemologists would reject the above biconditional on the grounds that in order for a process to be knowledge-conducive it should also be safe (where a safe process is one that could not have easily being wrong). Consider for example Anti-Luck Virtue Epistemology: S knows that p if and only if S’s safe belief that p is the product of her relevant cognitive
18

")! ! Now to see why we might get the impression of a close fit between virtue epistemology and the hypothesis of extended cognition, here are three common-sense functionalist criteria, which Clark (2010) suggests must be satisfied by non-biological candidates in order to be included into an individual’s cognitive system:
1) “That the process be reliably available and typically invoked”.

That is, the agent should habitually and easily invoke the external resource. In other words, its employment must be a disposition/habit of the agent’s overall cognitive mechanism. !
2) “That any information thus retrieved be more-or-less automatically endorsed. It should not usually be subject to critical scrutiny. […] It should be deemed about as trustworthy as something retrieved clearly from biological memory”.

That is, the information in the resource must be regarded as normal and reliable and not be necessarily reliable. It suffices that its employment result into an equally trustworthy belief-forming process as the one of forming beliefs on the basis of one’s own biological memory.19 At this point, however, one might object that being reliable is not the same as being trustworthy (i.e., being regarded as reliable). But, in response, notice first that Clark identifies the notion of trustworthiness of a process with the idea of being “more-or-less automatically endorsed” or in other words “not usually subject to critical scrutiny”. That is, the target process must not have been (for the most part) problematic in the past. Moreover, the processes under consideration are also supposed to be cognitive dispositions or habits of the agent that he has repeatedly employed in the past, and so had they been problematic the agent would have noticed that and responded appropriately. Accordingly, a trustworthy belief-forming process in Clark’s account, will be one that tends to produce true rather than false beliefs, which !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
abilities (such that her safe cognitive success is to a significant degree creditable to her cognitive agency) (Pritchard forthcoming, 20). Again, in (Pritchard 2010a, 76) we can read: “ knowledge is safe belief that arises out of the reliable cognitive traits that make up one’s cognitive character, such that one’s cognitive success is to a significant degree creditable to one’s cognitive character”. For a defense of the claim that the safety condition is not necessary for virtue reliabilism to account for knowledge see (Palermos forthcoming) 19 That is, the process does not need to be, due to underlying logical or quasi-logical relations, 100% reliable. Notice that memory is supposed to be reliable even though one may misremember.

!

"*!

is to say that it will be objectively reliable in the virtue reliabilist’s sense. What the agent will deem reliable will be that which is objectively reliable, i.e., that which has not been (for the most part) problematic in the past. Furthermore, notice this negative way of deeming processes reliable with which Clark concurs (i.e., that a trustworthy process is one that is not usually subject to critical scrutiny such that it is more-or-less automatically endorsed). What this means is that the agent does not need to have any beliefs about why or whether his belief-forming process is trustworthy; it suffices that it has not repeatedly caught his negative attention in the past. This is in good agreement with the proposed minimalist understanding of subjective justification according to which one does not need to rely on any beliefs but simply on one’s motivation to believe the truth. For example, one will trust one’s vision in appropriate circumstances, just because vision has not been notably problematic in the past (in those circumstances). By being motivated to believe the truth one will thereby employ the belief-forming process that has not in the past (notably) failed to be conducive towards that end, and crucially, one will do so without even thinking about it.
3) “That information contained in the resource should be easily accessible as and when required”.

That is, the agent must be able to employ it as if it was part of his organismic cognitive mechanism. In other words, the resource must be integrated into the agent’s overall cognitive mechanism. So we see that the same features of a process that epistemologists deem important in order for a process to be knowledge-conducive are required by a common-sense functionalist understanding of cognition in order for a process to count as part of one’s mind. This is a promising observation. Notice, however, that even if there were no such close fit between these broad features, we would still be able to show that the two theories are essentially connected. The reason, as we noted before, is that some of the above features (e.g., normality and dispositionality of the process) may be conducive towards the process being knowledge-conducive, but they are neither necessary nor sufficient for that end. Instead, the only requirement is that the process be integrated into the agent’s cognitive character.

#+! ! Now in relation to this, philosophy of cognitive science has recently started shifting its focus away from the above common-sense functionalist criteria for including an external resource into one’s cognitive system. Instead, Chemero (2009), Froese et al. (2013) and I (Palermos 2014) have suggested that the only requirement for an external element to count as a constitutive part of the agent’s cognitive system is that it be non-linearly related to the rest of the agent’s cognitive system. The motivation for this is that, according to dynamical systems theory, these non-linear relations give rise to an overall non-decomposable system that consists of all the contributing parts. And two reasons for postulating the overall system are that these non-linear interactions (1) give rise to new systemic properties that belong only to the overall system and to none of the contributing systems alone (therefore we have to postulate the overall extended system) and (2) prevent us from decomposing the two systems in terms of distinct inputs and outputs from the one subsystem to the other (therefore we cannot but postulate the overall system).20 What is even more interesting to our present purposes, however, is that just as Greco holds that cognitive integration is a matter of interaction and cooperation between cognitive processes, so those non-linear relations that allow us to talk about integration within philosophy of mind and cognitive science emerge only on the basis of cooperative feedback loops between the contributing elements of the overall system. Therefore both in epistemology and philosophy of mind and cognitive science the same criterion (cooperative interaction with the rest of the agent’s cognitive system) is required for a process to be integrated into an agent’s cognitive system and thereby count as knowledge-conducive. This, however, should not really come as a surprise. Given that virtue reliabilism holds that knowledge must be the product of cognitive ability (however that ability may be realized) and that the hypothesis of extended cognition sets out to reveal which processes can count as cognitive abilities (wherever they may be located), this close fit between the two theories seems to be as it should be. The conclusion that follows, then, is that there is no principled theoretical bar disallowing extended belief-forming processes from counting as knowledge-conducive cognitive abilities. Given that virtue reliabilism !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
For a detailed explanation of why the existence of non-linear relations that arise out of the mutual interactions between agents and their artifacts ensures the existence of extended cognitive systems see (Palermos 2014).
20

! makes no specifications as to whether knowledge-conducive cognitive abilities should be located within the agent’s head, then, provided that the

#"!

condition of cognitive integration is met, the epistemic agent may extend his knowledge-conducive cognitive character beyond his organismic cognitive abilities by incorporating epistemic artifacts to it. 3. IS THIS NECESSARY FOR EPISTEMOLOGY AND WHAT ARE THE RAMIFICATIONS? Obviously, the possibility of knowledge-conducive cognitive characters that may nevertheless be extended beyond our organismic cognitive capacities can generate interesting ramifications both for traditional and social epistemology, which may allow us to think about knowledge in new ways. Focusing, however, on the integrated nature of our extendable cognitive characters may not only be important for moving forward, but also necessary for accounting for knowledge as we already think about it. According to Greco (1999, 287), in addition to one’s organismic cognitive abilities of the brain/central nervous system, a person’s cognitive character may also consist of “acquired skills of perception and acquired methods of inquiry including those involving highly specialized training or even advanced technology”. The reason for this move is that we need to account for advanced cases of knowledge where one’s believing the truth is the product of the operation of epistemic artifacts such as telescopes, microscopes, tactile visual substitution systems and so on.21 The problem, however, is that in the traditional conception, cognition takes place strictly within the agent’s head and so artifacts cannot be parts of one’s cognitive character. One way to sidestep this problem for virtue reliabilism, could be to claim that, in such cases, it is merely the agent’s training and skill of using the artifact, as mirrored in the agent’s neural/bodily architecture, that is the most salient factor in the causal explanation of the agent’s cognitive success (i.e., believing the truth). Notice, however, that when an agent employs an epistemic tool, his true belief arises as the product of the interaction between his internal processes and the artifact. According to dynamical systems !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
21

See Bach-y-Rita and Kercel (2003) for a recent review on tactile visual substitution systems.

##! ! theory, then, the cognitive process that allows the agent to detect the truth is not merely ‘aided’ or ‘assisted’ by the artifact but is, instead, constituted by it as it arises out of the ongoing mutual interaction between the agent and the artifact.22 Therefore, in a causal explanation of how the agent acquired his true belief, it will be impossible to disentangle the agent’s training and skill of using the artifact from his actual engagement with it. 23 But even if such decomposition were possible, notice in addition that the part of the process that allows the agent to detect the truth, or in other words to be sensitive to the facts, is the external component. To illustrate this, consider, on one hand, an untrained agent in possession of a properly working artifact. In that case, it is obvious that even though the agent will initially be unable to form any (true or false) beliefs, eventually—provided that he gains sufficient experience such that he can interact with it—not only will he form beliefs, but he will also reliably enjoy cognitive success. On the other hand, think about a well-trained agent, but in possession of a faulty artifact. In this case, despite the agent’s excellent internal skills, it is evident that he would be unable to reach any (non-lucky) true beliefs, no matter how much he tried. It therefore seems that in such cases the most (and maybe the only) significant factor that explains the truth-status of the agent’s belief is the epistemic artifact. In other words, since the agent’s belief is true in virtue of the artifact, the virtue reliabilist must account for it being part of the agent’s cognitive system. Given, however, that cognition is normally supposed to take place within the agent’s head, virtue reliabilists can only account for such cases by wedding their view to the hypothesis of extended cognition. Accordingly, combining the extended cognition hypothesis with virtue reliabilism on the basis of their close fit does not seem to be just an available option for epistemologists, but also necessary for dealing with advanced cases

It should be here noted that not every case of the employment of an artifact is a case of cognitive extension, but only when the agent mutually interacts with it. For an objective criterion of constitution and on what may count as a genuine case of cognitive extension, see (Palermos 2014). 23 Remember that according to virtue reliabilism and the underlying ability intuition on knowledge, knowledge is belief that is true in virtue of cognitive ability, where, according to Greco, “in virtue of” must be understood in causal explanatory terms. Even though several proponents of virtue reliabilism agree on this general causal-explanatory understanding of the view, there is disagreement on whether the relevant cognitive ability should be the “most salient” (Greco 2010) or merely a “significant” (Pritchard 2010b) factor in the causal explanation of how the agent acquired his true belief.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
22

!

#$!

of knowledge where the latter is the product of the employment of epistemic artifacts that the agent mutually interacts with. Apart from the necessity of introducing the extended cognition hypothesis within epistemology, however, this move can also generate interesting ramifications, especially for social epistemology. For instance, it can lead to the further claim that there could be cognitive characters that do not just extend beyond an agent’s organismic capacities, but which are instead distributed amongst several agents along with their epistemic artifacts. The hypothesis of distributed cognition, which has been developed in parallel to the hypothesis of extended cognition (Hutchins 1995 , Theiner et al. 2010, Sutton et al. 2008, Wilson 2005, Heylighen et al. 2007), differs to the latter position only in that this time the cognitive system extends to include epistemic artifacts as well as other agents. And interestingly, most proponents of the view (Sutton et al. 2008, Theiner et al. 2010, Wegner 1985, Tollefsen & Dale 2011) point out again that it is the existence of non-linear cooperative interactions between the contributing members and their artifacts that is the criterion by which we can judge whether we have an integrated distributed cognitive system. Accordingly, there could be knowledge-conducive cognitive characters, which may nevertheless be distributed. This is an interesting possibility, because it can allow us to combine an individualistic approach to knowledge, such as virtue reliabilism, with the hypothesis of distributed cognition in order to account for epistemic group agents: Groups of individuals who exist and gain knowledge in virtue of a shared, common cognitive character that mainly consists of a distributed cognitive ability—a collective cognitive ability that emerges out of the members’ mutual (socio-epistemic) interactions and which is not reducible to the cognitive abilities possessed by the individual members, thereby allowing us to speak of a group agent in itself. This is important, because by recognizing a group of people as a self-standing agent in itself, we can then use an individualistic approach to knowledge to account for knowledge that is collectively produced and which is, thereby, distinctively social. In other words we can make sense of the claim that p is known by S (the group agent), even though it is not known by any individual alone.24
24 For a more detailed explanation of how virtue reliabilism may be applied to epistemic group agents see (Palermos & Pritchard 2013).

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

#%! ! Remarkably, such group agents have already started being studied within cognitive science. Consider for example transactive memory systems (TMSs)—i.e., groups of two or more individuals who collaboratively encode, store and retrieve information. The reason why TMSs are good candidates for distributed cognitive systems—and thereby for epistemic group agents—is that, as Sutton et al. observe (2008), such systems are likely to involve skillful interactive simultaneous coordination of people who can thereby count as a single integrated cognitive system. Therefore, we can use TMSs in order to “conceptualize how people in close relationships may depend on each other for acquiring, remembering, and generating knowledge” (Wegner et al. 1985, 253):
Ordinarily psychologists think of memory as an individual’s store of knowledge, along with the processes whereby that knowledge is constructed, organized, and accessed. […] With transactive memory we are concerned with how knowledge enters the dyad, is organized within it, and is made available for subsequent use by it (ibid., 256).

Apart from incorporating already existing research from cognitive science, however, the combination of virtue reliabilism with the hypotheses of extended and distributed cognition can generate new avenues for research, some of which have for a long time been inaccessible. An interesting example is the intersection between epistemology and the field of history and philosophy of science. These two intimately related fields have so far been at odds—an awkward situation owing to the fact that the former discipline has traditionally being individualistic whereas the latter has for the most part been socially oriented (hardly anyone could deny the social nature of the scientific process, especially after the publication of Kuhn’s The Structure of the Scientific Revolutions, in 1962). The present approach, however, could now provide a useful link between the two fields. Science is primarily performed by individual scientists employing their hardware and software epistemic artifacts or by research teams operating within scientific labs that are uniquely tailored to fit their purposes. Accordingly, the concepts of extended cognitive characters and epistemic group agents could become very handy for a mainstream epistemological analysis of the scientific progress. As Giere and Moffat (2003, 308) note in their discussion of the scientific revolution of the 16th century,

!

#&!
“No ‘new man’ suddenly emerged sometime in the sixteenth century....The idea that a more rational mind...emerged from darkness and chaos is too complicated a hypothesis” [Latour 1986, 1]. We agree completely. Appeals to cognitive architecture and capacities now studied in cognitive sciences are meant to explain how humans with normal human cognitive capacities manage to do modern science. One way, we suggest, is by constructing distributed cognitive systems that can be operated by humans possessing only the limited cognitive capacities they in fact possess.

4. CONCLUSION The topic of cognitive integration is a far-reaching multidisciplinary theme that touches upon a wide range of disparate questions with the potential of bringing several research areas together. Within epistemology, the phenomenon of cognitive integration can reveal how knowers may be subjectively justified despite the absence of positive reasons in favor of their true beliefs, pointing towards a minimalist yet informative epistemic norm: provided that one’s belief-forming process has been integrated into one’s cognitive system/character and that one is motivated to believe what is true, one can be justified in holding the resulting belief merely by lacking any doubts that his way of forming his belief, or that the belief itself, is inappropriate. Within philosophy of mind, cognitive integration can explain how it is possible to extend our cognitive capacities beyond our organisms to the artifacts we employ or other individuals we may interact with. And combining philosophy of mind with epistemology by using the process of cognitive integration as the connecting point provides the necessary means to account for advanced cases of knowledge where the known belief is true in virtue of the operation of epistemic artifacts or even the activity of collaborative groups. Quite likely, however, such an account of knowledge won’t be valuable just within epistemology, philosophy of mind and their intersection. Focusing on the essentially technological and collaborative nature of the scientific process, such an approach to knowledge could finally provide a strong link between the related but so far persistently isolated fields of philosophy of mind, philosophy of science, and epistemology.

#'! ! Acknowledgments I am very thankful to John Greco and an anonymous referee for detailed comments on previous versions of this paper. Research in the area of this paper was carried out as part of the AHRC-funded ‘Extended Knowledge’ research project (AH/J011908/1). References Alter, L., A. & Oppenheimer, D., M. (2007). ‘Overcoming Intuition: Metacognitive Difficulty Activates Analytic Reasoning’. Journal of Experimental Psychology. Vol. 136, No. 4, 569-576. BonJour, L. (1980). “Externalist Theories of Empirical Knowledge,” Midwest Studies in Philosophy, 5: 53–73. —— (1985). The Structure of Empirical Knowledge, Cambridge, MA: Harvard University Press. !! (2002). ‘Internalism and Externalism’, Oxford Handbook of Epistemology, (ed.) P. Moser, 234-64, Oxford: Oxford University Press. Brewer, F. W. & Lambert, B. L. (2001). ‘The Theory Ladeness of Observation and the Theory-Ladeness of the Rest of the Scientific Process’. Philosophy of Science, Vol. 68, No. 3, Supplement: Proceedings of the 2000 Biennial Meeting of the Philosophy of Science Association. Part I: Contributed Papers, pp. S176-S186. Clark, A., & Chalmers, D. (1998). ‘The Extended Mind’. Ananlysis 58, no. 1: 719. —— —— (2008). Supersizing The Mind. Oxford University Press. (2010). ‘Memento’s Revenge: The Extended Mind, Extended’. In the Extended Mind. (2010), Menary (ed.) Cambridge, Massachusetts, MIT press. Chemero, A. (2009). Radical Embodied Cognitive Science. MIT press. Chisholm, R. M. (1977). Theory of Knowledge (2nd ed.), Englewood Cliffs, NJ: Prentice-Hall. Churchland, P. M. (1979). Scientific Realism and the Plasticity of the Mind. Cambridge: Cambridge University Press. —— (1988). ‘Perceptual Plasticity and Theoretical Neutrality: A Reply to

! Jerry Fodor’, Philosophy of Science 55: 167-187. ——

#(!

(1989). A Neurocomputational Perspective. The Nature of Mind and the Structure of Science. Cambridge: The MIT Press.

Estany, A. (2001). ‘The Thesis of Theory-Laden Observation in the Light of Cognitive Psychology’. Philosophy of Science, Vol. 68, No.2 (Jun ., 2001), pp. 203-217. Feyerabend, P. K. (1975). Against Method: Outline of an anarchistic theory of knowledge. Fodor, J. A. (1983). The Modularity of Mind. Cambridge, MA: MIT Press. —— —— (1984). ‘Observation Reconsidered’, Philosophy of Science, 51: 23-43. (1988). ‘A Reply to Churchland’s ‘Perceptual Plasticity and Theoretical Neutrality’’, Philosophy of Science 55: 188-198. Froese, T., Gershenson, C., and Rosenblueth, D., A. (2013). ‘The Dynamically Extended Mind’. http://arxiv.org/abs/1305.1958 Fuller, S. (2012). ‘Social Epistemology: A Quarter-Century Itinerary’, Social Epistemology: A Journal of Knowledge, Culture and Policy 26, 267-83. Giere, R. & Moffat, B. (2003). ‘Distributed Cognition: Where the Cognitive and the Social Merge’. Social Studies of Science. 33/2, pp. 1-10. Goldman, A. (1986). Epistemology and Cognition, Cambridge, MA: Harvard University Press. Greco, J. (1999). ‘Agent Reliabilism’, in Philosophical Perspectives 13: Epistemology (1999). James Tomberlin (ed.), Atascadero, CA: Ridgeview Press, pp. 273-296. —— (2004). ‘Knowledge As Credit For True Belief’, in Intellectual Virtue: Perspectives from Ethics and Epistemology. M. DePaul & L. Zagzebski (eds.), Oxford: Oxford University Press. —— —— (2007) ‘The Nature of Ability and the Purpose of Knowledge’, Philosophical Issues 17, pp. 57- 69. (2010). Achieving Knowledge: A Virtue-Theoretic Account of Epistemic Normativity. Cambridge University Press. Hanson, N. R. (1961). Patterns of Discovery. Cambridge: Cambridge University Press. —— (1969). Perception and Discovery; An Introduction to Scientific Inquiry. San Francisco: Freeman, Cooper.

#)! ! Heylighen, F., Heath, M., Van Overwalle, F. (2007). ‘The Emergence of Distributed Cognition: A Conceptual Framework’. In Proceedings of collective intentionality IV (2004), Volume: IV, Publisher: University of Siena. Hutchins, E. (1995). Cognition in the Wild. Cambridge: MIT Press. Kuhn, T. S. (1962). The Structure of Scientific Revolutions. Chicago: The University of Chicago Press. Oppenheimer, D., M. (2008). ‘The Secret Life of Fluency’. Trends in Cognitive Science. 12(6):237-41. Palermos, O. (2011). ‘Belief-Forming Processes, Extended’, Review of Philosophy and Psychology 2, 741-65. !! (2014). ‘Loops, Constitution, and Cognitive Extension’, Cognitive Systems Research, 27: 25–41. !! (forthcoming). ‘Could Reliability Naturally Imply Safety?’, European Journal of Philosophy, DOI: 10.1111/ejop.12046. Palermos, S. O., & Pritchard, D. (2013). ‘Extended Knowledge and Social Epistemology’, Social Epistemology Review and Reply Collective 2 (8): 105-120. Plantinga, A. (1993a). Warrant and Proper Function. New York: Oxford University Press. —— (1993b). Warrant: The Current Debate, Oxford: Oxford University Press. Justification’, Stanford Encyclopaedia of Philosophy, (ed.) E. Zalta, http://plato.stanford.edu/entries/justep-intext/ Poston, T. (2008). ‘Internalism and Externalism in Epistemology’, Internet Encyclopaedia of Philosophy, (eds.) B. Dowden & J. Fieser, www.iep.utm.edu/int-ext/ Pritchard, D. (forthcoming). ‘Anti-Luck Virtue Epistemology’, Journal of Philosophy. —— —— (2009). Knowledge, London: Palgrave Macmillan. (2010a). ‘Knowledge and Understanding’, in A. Haddock, A. Millar & D. H. —— Pritchard, The Nature and Value of Knowledge: Three Investigations, Oxford: Oxford University Press. (2010b). ‘Cognitive Ability and the Extended Cognition Thesis’. Synthese. Pappas, G. (2005). ‘Internalist versus Externalist Conceptions of Epistemic

!

#*!

Pryor, J. (2001). ‘Highlights of Recent Epistemology’, British Journal for the Philosophy of Science 52, 95-124. Sosa, E. (1988).‘Beyond Skepticism, to the Best of our Knowledge’. Mind, New Series, vol. 97, No.386, pp. 153-188 —— —— (1993). ‘Proper Functionalism and Virtue Epistemology’. Nous, Vol. 27, No. 1, 51-65. (2007). A Virtue Epistemology: Apt Belief and Reflective Knowledge, Oxford: Clarendon Press. Steup, M. (1999). ‘A Defense of Internalism’, The Theory of Knowledge: Classical and Contemporary Readings (3rd Ed.), (ed.) L. Pojman, 310-21, Belmont, CA: Wadsworth. Sutton, J., Barnier, A., Harris, C., Wilson, R. (2008). ‘A conceptual and empirical framework for the social distribution of cognition: The case of memory’. Cognitive Systems Research, Issues 1-2, pp. 33–51. Theiner, G. & Allen, C. & Goldstone, R. (2010). ‘Recognizing Group Cognition’. Cognitive Systems Research, Vol. 11, Issue 4, pp. 378-395. Tollefsen, D. & Dale, R. (2011). “Naturalizing joint action: A Process-based approach”. Philosophical Psychology 25 (3):385-407. Unkelbach, C. & Greifeneder, R. (2013). ‘A general model of fluency effects in judgment and decision making’. In C. Unkelbach & R. Greifeneder (Eds.), The Experience of Thinking (pp. 11-32). New York: Psychology Press. Wegner, M., Giuliano, T., Hertel, P. (1985). ‘Cognitive interdependence in close relationships’. In W. J. Ickes (Ed.), Compatible and incompatible relationships (pp. 253–276). New York: Springer-Verlag. Wilson, R., A. (2005). ‘Collective Memory, Group Minds, and the Extended Mind Thesis’. Cognitive Processing, Vol. 6, Issue 4, pp. 227-236.

! ! !! !

Radically Enactive Cognition in Our Grasp
Daniel D. Hutto University of Hertfordshire

Òthe question is not to follow out a more or less valid theory but to build with whatever materials are at hand. The inevitable must be accepted and turned to advantageÓ

- Napoleon Bonaparte

Abstract

Radically Embodied/Enactive accounts of Cognition, REC, propose to fundamentally shift the way cognitive scientists think about the basic nature of mentality. This paper argues that focusing on the sophisticated but unplanned character of human manual activity enables such accounts to address a standard worry about their scope and reach. A counter proposal for handling such cases by defenders of Conservative Embodied/Enactive account of Cognition, CEC, is examined and found wanting. CEC accounts make appeal to Action Oriented Representations (AORs) to do the work that fans of REC argue is done without representational mediation. It is argued that naturalistically inclined defenders of CEC face a crippling dilemma.

Radically Enactive Cognition in Our Grasp

2

2

Radically Enactive Cognition in Our Grasp

3

1. Reckoning with REC

For those working in the sciences of the mind these are interesting times. Revolution is, yet again, in the air. This time it has come in the form of new wave thinking about the basic nature of mind of the sort associated with radically embodied or enactive approaches to cognition; REC for short. REC approaches are marked out by their uncompromising and thoroughgoing rejection of intellectualism about the basic nature of mentality. As Varela, Thompson and Rosch (1991) saw it the defining characteristic of this movement is its opposition to those theories of mind that Òtake representation as their central notionÓ (p. 172). The most central and important negative claim of REC is its denial that all forms of mental activity depend on the construction of internal models of worldly properties and states of affairs by means of representing its various features on the basis of retrieved information. Not since the ousting of behaviourism with the advent of the most recent cognitive revolution has there been such a root and branch challenge to widely accepted assumptions about the very nature of mentality. In a remarkable reversal of fortune, it is now a live question to what extent, if any, representational and computational theories of the mind Ð those that have dominated for so long Ð ought to play a fundamental role in our explanatory framework for understanding intelligent activity. Defenders of REC approaches argue that representation and computation are neither definitive of, nor provide the basis of, all mentality.

3

Radically Enactive Cognition in Our Grasp

4

From the side-lines, interested onlookers might be forgiven for thinking the revolution is already over; embodied and enactive ways of thinking are already comfortably ensconced, having established deep roots in a number of disciplines. Far from merely being at the gates, the Barbarians are, it seems, now occupying the local cafŽs and wine bars in the heart of the city. Even those who most regret this development are prepared to acknowledge that, de facto, there has been a major sea change. Lamenting the rise of a pragmatist trend in cognitive science, Fodor (2008) acknowledges that REC-style thinking in cognitive science is now the ÔmainstreamÕ. He puts this down to an infectious disease of thought (Ôa bad coldÕ Ð as he puts it p. 10). Others are edgily aware of the spectre of REC approaches Òhaunting the laboratories of cognitive scienceÓ (Goldman and de Vignemont 2009, p. 154). ÔPervasive and unwelcomeÕ is the verdict of these authors: REC may be everywhere but is something to be cured or exorcised, as soon as possible. Despite their growing popularity, which some hope is nothing more than a short-lived trend, REC approaches remain hotly contested. Certainly, it is true that there has yet to be a definitive articulation of the core and unifying assumptions of embodied and enactive approaches to cognition Ð EC approaches Ð radical or otherwise. Indeed, there is some reason to doubt that it will be possible to group together all of the offerings that currently travel under the banner of EC by identifying their commitment to a set of well-defined core theoretical tenets (see Shapiro 2011, p. 3). Nevertheless, if REC approaches, in particular, are to maintain credibility and avoid charges of simply riding the crest of a fashionable wave, at a bare minimum, serious objections from the old guard should be convincingly answered.

4

Radically Enactive Cognition in Our Grasp Some criticisms are easier to deal with than others. One line of argument draws on observations about the proper order and requirements of cognitive explanations. Fodor (2008) hopes to dispatch REC with one fell blow by observing that positing of representationally-based thinking is the minimum requirement for explaining any and all

5

activity that deserves the accolade Ôintelligent respondingÕ Ð an observation predicated on the assumption that we can draw a principled bright line between what is properly cognitive and what is not. Accordingly, he insists that we have no choice but to accept that Òthe ability to think the kind of thoughts that have truth-values is, in the nature of the case, prior to the ability to plan a course of action. The reason is perfectly transparent: Acting on plans (as opposed to, say, merely behaving reflexively or just thrashing about) requires being able to think about the worldÓ (p. 13). In a nutshell, this is FodorÕs master argument for thinking that pragmatist REC-style approaches to the mind must be false: for to think the kinds of thoughts that have truthvalues is to think thoughts with representational content and, presumably, to make plans requires manipulating these representations (and their components) computationally.1 So, in short, if all bona fide intelligent action involves planning, and all bona fide planning involves computing and representation then this is bad news from the frontline for REC rebels. Without a doubt some problems, indeed, perhaps whole classes of problems, are best addressed through advanced careful planning Ð planning of the sort that requires the rulegoverned manipulation of truth-evaluable representations. Sometimes it is not only advisable, but utterly necessary to stand back and assess a situation in a relatively detached manner, drawing explicitly on general background propositional knowledge of

5

Radically Enactive Cognition in Our Grasp situations of a similar type and using that knowledge to decide Ð say, by means of

6

deduction and inference Ð what would be the correct or most effective approach. This can be done before initiating any action or receiving any live feedback from the environment. This is the preferred strategy for dealing with situations, such as defusing bombs, in which a trial and error approach is not advisable. Making use of remote representations also works equally even for more mundane types of tasks Ð those that include, for example, figuring out the best route from the train station to oneÕs hotel in a foreign city where one seeks to do this in advance from the comfort of oneÕs office, long before boarding a plane. Such intelligent pre-planning can be done at a remove, and reliably, if it is possible to exploit and manipulate symbolic representations of the target domain on the assumption that one has the requisite background knowledge and can bring that knowledge to bear. This will work if the domain itself is stable over time since that will ensure that any stored representations remain up to date and accurate. By using representations of a wellbehaved domainÕs features and properties, and having a means of knowing, determinately, what to expect of it if it changes under specific modifications and permutations, it is possible for a problem solver to plan how to act within it without ever having to (or indeed ever having had to) interact with it in a first-hand manner. This is, of course, the ideal end state of high theoretical science. As linguistic beings, humans are representation mongers of this sort and thus regularly adopt this basic strategy to solve problems. Our cultural heritage provides us with a store of represented knowledge Ð in many and various formats Ð that enables us to do so, successfully, under the sorts of conditions just mentioned. But it hardly follows that this

6

Radically Enactive Cognition in Our Grasp type of cognitive engagement is the basis of, required for, or indeed suitable for, all sorts of tasks Ð always and everywhere. Echoing Ryle (1949), No‘ (2009) hits the nail on the head, noting that, Òthe real problem with the intellectualist picture is that it takes rational deliberation to be the most basic kind of cognitive operationÓ (p. 99, emphasis added). Intellectualism of this unadulterated kind Ð of the sort that assumes the existence of strongly detached symbolic representations of target domains Ð has fallen on hard times. It finds only a few hard-core adherents in todayÕs cognitive sciences. Indeed, if anything has promoted the fortunes of REC it has been the dismal failure of this sort of rules and

7

representation approach when it comes to dealing with the most basic forms of intelligent activity. This is the headline-grabbing lesson of recent efforts in robotics and artificial intelligence, which have provided a series of existence proofs against strong representationalism about basic cognition. Pioneering work by Brooks (1991a, 1991b), for example, reveals that intellectualism is a bad starting point when thinking about how to build robots that actually work. There are important lessons to learn by paying attention to architectonic requirements of robots that are able to complete quite basic sorts of tasks, such as navigating rooms while avoiding objects or recognizing simple geometrical forms and shapes. Inverting standard intellectualist thinking, Brooks famously rejected the Sense-Model-Plan-Act approach, and built robots that dynamically and frequently sample features of their local environments in order to directly guide their responses, rather than going through the extra steps of generating and working with descriptions of those environments. These first generation behaviour-based robots, and those that followed after them, succeed precisely because the robotsÕ behaviours are guided by continuous, temporally extended

7

Radically Enactive Cognition in Our Grasp interactions with aspects of their environments rather than working on the basis of represented internal knowledge about those domains, knowledge that would presumably

8

be stored somewhere wholly in the robotsÕ innards. The guiding principle behind BrooksÕ so-called subsumption architectures is that sensing is directly connected with appropriate responding without representational mediation. Crucially, the great success of these artificial agents demonstrates that it is possible for a being to act intelligently without creating and relying on internal representation and models. Very much in line with theoretical worries raised by the frame problem, it may even be that, when it comes to basic cognition, this is the only real possibility. Not just artifice but nature too provides additional support for the same conclusion. Cricket phonotaxis (Webb 1994, 1996) is a vivid example of how successful on-line and successful navigation takes place in the wild, apparently without the need for representations or their manipulation. Female crickets locate mates by attending to the first notes of male songs, frequently adjusting the path of their approach accordingly. They only manage this because the male songs that they attune to have a unique pattern and rhythm Ð one that suits the particular activation profiles of the female interneurons. The capacity of these animals to adjust their behaviour when successfully locating mates requires them to engage in a continuous interactive process of engagement with the environment. In doing so they exploit special features of their non-neural bodies Ð including the unique design of their auditory mechanism Ð as well as special features of the environment Ð the characteristic pattern of the male songs. In this case a beautiful cooperation arises because of the way the cricketÕs body and wider environment features

8

Radically Enactive Cognition in Our Grasp

9

enable successful navigational activity Ð activity that involves nothing more than a series of dynamic and regular embodied interactions. For reasons of space I will not rehearse in full detail precisely how behaviour-based robots or insects make their way in the world. These cases are well known and much discussed (for excellent summaries in greater detail, including other examples see Wheeler 2005, ch. 8 and Shapiro 2011, ch. 5). For the purposes of this essay it suffices to note that when bolstered by the articulation of a supporting theoretical framework, one easily provided by dynamical systems theory, these observations offer a serious and wellknown challenge to the representationalist assumptions of intellectualist cognitive science (Beer 1998, 2000, Thompson 2007, Garz—n 2008, Chemero 2009). In sum, what the foregoing reflections teach us is that there are cases in which bodily and environmental factors play ineliminable and non-trivial parts in making certain types of cognition possible. A familiar intellectualist response to these sorts of examples is to try to cast these wider contributions as playing no more than causal supporting roles that, even if necessary to enable cognition do not constitute or form part of it. For reasons that should be obvious from the foregoing discussion, it is not clear how one might motivate this interpretation and make it stick with respect to the sorts of cases just described. In rejecting representationalism, REC takes at face value what attending to the architectonic details of how these agents work suggests Ð i.e. that the specified bodily and environmental factors are fully equal partners in constituting the embodied, enactive intelligence and cognition of these artificial and natural agents. Accordingly, although for certain practical purposes and interventions it may be necessary to carve off and focus on specific causally contributing factors in isolation, the cognitive activity itself cannot be

9

Radically Enactive Cognition in Our Grasp

10

seen as other than Òa cyclical and dynamic process, with no nonarbitrary start, finish, or discrete stepsÓ (Hurley 2008, p. 12, see also Garz—n 2008, p. 388). Or, put otherwise, when it comes to understanding cognitive acts Òthe agent and the environment are nonlinearly coupled, they, together constitute a nondecomposable systemÓ (Chemero 2009, p. 386). In promoting this sort of line, REC flags up the Ôreal dangerÕ that Òthe explanatory utility of representation talk may evaporate altogetherÓ (Wheeler 2005, p. 200). As Shapiro (2011) notes, the interesting question is whether an anti-representationalist paradigm has real prospects of replacing intellectualist cognitive science altogether. And, as indicated above, he is right to suppose that the two main, complementary Òsources of support for Replacement come from (i) work that treats cognition as emerging from a dynamical system and (ii) studies of autonomous robotsÓ (p. 115). While this is a potentially powerful cocktail, it remains to be seen just how far it might take us. For to make a convincing case for their far-reaching revolutionary ambitions, proponents of REC must take the next step and Òargue that much or most cognition can be built on the same principles that underlie the robotÕs intelligenceÓ (Shapiro 2011, p. 116, emphasis added). Rather than denying that there can be no such thing as non-representational cognition, intellectualists might take heart from this challenge and agree to split the difference, allowing that very basic forms of cognition Ð of the sort exemplified by robot and insect intelligence Ð might be suitable for REC treatment but not the rest. This is to adopt a kind of containment strategy Ð a kind of theoretical kettling or corralling. Intellectualists might be tempted to concede that supporters of the radical left have a point, up to a point Ð

10

Radically Enactive Cognition in Our Grasp allowing that Òrepresentations are not necessary for coordinated responses to an environment with which one is dynamically engagedÓ (Shapiro 2011, p. 153). But this concession would be made in the secure knowledge that it Òwould support only the conclusion that agents do not require representations for certain kind of activities. However, a stronger conclusion, for instance that cognition never requires representational states, does not followÓ (Shapiro 2011, p. 153).

11

REC approaches would be, accordingly, of limited value on the assumption that they wonÕt scale up. Call this the Scope objection. It allows one to accept certain lessons learned from the lab and nature while safe in the knowledge that even if representations are not needed to explain the most basic forms of cognition that this in no way poses an interesting threat to intellectualism since the sorts of cases in question Òrepresent too thin a slice of the full cognitive spectrumÓ (Shapiro 2011, p. 156). This is in line with the oftcited claim that some behaviour is too off-line and representation hungry to be explained without appeal to the manipulation of symbolic representations. In particular, nonrepresentational cognition, which might do for simple robots and animals, isnÕt capable of explaining properly world-engaging, human forms of cognition. But should that assessment prove mistaken Ð if REC approaches were to make substantial in-roads in this latter domain Ð then the boot might just be on the other foot. For it might turn out that representationally hungry tasks only make up a very small portion of mental activity; representationally-based cognition might be just the tip of the cognitive iceberg.

11

Radically Enactive Cognition in Our Grasp

12

2. A Helping Hand

This is where reflection on the special prowess of the human hand comes in handy. It cannot be denied that a great deal of human manual activity is connected with sophisticated forms of cognition. Milner and GoodaleÕs (1995) famous experiments reveal that humans can perform remarkably demanding manual acts, with precision Ð acts requiring the exercise of very fine-grained motor capacities, such as posting items through slots with changing orientations Ð even when they lack the capacity to explicitly report upon or describe visual scenes they are dealing with. Nor, with only rare exceptions, is it credible that humans normally learn how to use their hands in these sorts of ways by means of explicit, representationally mediated instruction, the rules for which only later becoming submerged and tacit. It is not as if children are taught by their caregivers through explicit description how to grasp or reach for items. Far more plausibly, is the hypothesis that we become handy through a prolonged history of interactive encounters Ð through practice and habit. An individualÕs manual know how and skills are best explained entirely by appeal to a history of previous engagements and not by the acquisition of some set of internally stored mental rules and representations. To invoke the favourite poetic motto of enactivists, this looks, essentially, to be a process of Ôlaying down a path in walkingÕ or in this case, ÔhandlingÕ.

12

Radically Enactive Cognition in Our Grasp It is possible that the special manual abilities of humans are sophisticated enough to have provided the platform and spurred on other major cognitive developments. Some

13

very strong claims have been made about the critical importance of the ways in which we use our hands in this regard Ð ways that some believe are responsible for enabling the emergence of other distinctively forms of human cognition, consciousness and culture. For example, Tallis (2003) regards our special brand of manual activity as the ultimate source of our awakening to self-consciousness. He tells us, ÒHerein lies the true genius of the hand: out of fractionated finger movements comes an infinite variety of grips and their combinations. And from this variety in turn comes choice Ð not only in what we do É but in how we do it É [and w]ith choice comes consciousness of actingÓ (p. 175). If Tallis is to be believed, ÒBetween the non-stereotyped prehensions of hominid hand and the stereotyped graspings of the animal paw there is opened a gap which requires, and so creates, the possibility of apprehension to cross itÓ (p. 36). These claims are tempered by the remark that ÒWe may think of the emergence of distinctive capabilities of the human hand as lighting a fuse on a long process that entrained many other parts of the human body and many other faculties as it unfoldedÓ (p. 6, emphasis added). This allows for the possibility that, ÒThe crucially important differences between human and non-human hands do not alone account for the infinitely complex phenomenon, unique in the order of the universe, of human culture. It is not so much the differences Ð which are very important Ð but the ability to make much of the differencesÓ (p. 33).2 Whatever is ultimately concluded about the defensibility of this last set of claims, the point is that if it should turn out that much human manual activity is best explained without appeal to the manipulation of rules or representations then defenders of REC will

13

Radically Enactive Cognition in Our Grasp

14

have made significant progress in addressing the scope challenge. REC approaches will have shown the capacity to advance well beyond dealing with the antics of a behaviourbased robots and insects, having moved deep in the heart of distinctively human cognition. Are there further grounds for thinking that manual activity is best explained in a representation-free way? TallisÕs (2003) philosophically astute and empirically informed examination of the hand provides an excellent starting point for addressing this question. He claims that Òthe hand [is] É an organ of cognitionÓ, and is so Ôin its own rightÕ (p. 28, p. 31). This is not to say that the hand works in isolation from the brain, indeed, Tallis stresses that the hand Ð for him, the tool of tools Ð is the ÒbrainÕs most versatile and intelligent lieutenantÓ (p. 22). Of course, this way of putting things suggests that the hand is, when all goes well, in some way nothing but a faithful subordinate Ð one that works under top-down instruction and guidance from above. This underestimates the bidirectional interplay between manual and brain activity - interplay of the sort that explains why the distinctive manual dexterity of Homo sapiens, which sets us apart even from other primates who also have remarkable abilities in this regard, was likely one of the Ômain driversÕ of the growth of the human brain (p. 22). These ideas can be taken much further if one fully rejects what Tallis calls the standard ploy.

While it is perfectly obvious that voluntary activity must be built up out of involuntary mechanisms, there are profound problems in understanding this. There are particular problems with the standard ploy invoked by movement physiologists: proposing that

14

Radically Enactive Cognition in Our Grasp the automation incorporates ÔcalculationsÕ that the brain (or part of it) ÔdoesÕ, which permit customisation of the programs to the singularities of the individual action (p. 65, emphases added)

15

Invoking the standard ploy amounts to making a hand waving and anthropocentric appeal to representational contents so as to specify and fill out hypothesized motor plans and motor programmes that supply instructive orders from on high, lending intelligence to and directing manual activity. For example, on this view motor plans, intentions and programs are understood as Òpropositional attitudes with contents of the form Ôlet [my] effector E perform motor act M with respect to goal-object GÕÓ (Goldman 2009, p. 238).3 The trouble is that even if we imagine that such representational contents exist, it is difficult to see how they could do the required work. The only chance they could have of specifying what is to be done, and how it is to be done, would be if they go beyond issuing very general and abstract instructions of the sort that Goldman gestures at above. Only very fine-grained instructions would be capable of directing or controlling specific acts of manual activity successfully. This raises a number of questions. How do brains decide which general kind of motor act, M, is the appropriate sort of motor act to use in the situation at hand? This, alone, is no simple business Ð given the incredible variety of possible manual acts.4 And, even if we put that concern aside, proponents of the view that brains can initiate and control manual acts by traditional intellectualist means are left with the problem of explaining how and on what basis brain decides how to execute any given act. A major problem for traditional forms of intellectualism is that the requirements for successfully

15

Radically Enactive Cognition in Our Grasp performing any particular motor act are tied to a unique and changing context. For example, even if everything else remains static the speed, angle of approach and grip aperture need to be altered appropriately at successive stages as one does something as simple as picking up a coffee cup. In a nutshell:

16

a particular challenge É has been to explain how cognition and perception processes regulate complex multi-articular actions in dynamic environments. The problem seeks to ascertain how the many degrees of freedom of the human motor system (roughly speaking the many component parts such as muscles, limb segments, and joints) can be regulated by an internally represented algorithm É and how the motor plan copes with the ongoing interaction between the motor system and energy fluxes surrounding the system, e.g., frictional forces and gravitational forces ... Not even the attempt to distinguish between the motor plan and the motor program has alleviated the problem in the literature (Summers and Anson, 2009) (Araœjo & Davids 2011, p. 12).

Successful manual activity requires bespoke and on the fly customisation. Hence, it is deeply implausible that brain can simply identify what is required for successfully completing a certain type of activity and simply issuing general instructions to be carried out in form of ordering pre-programmed routines to be carried out. The implausibility of this suggestion is underscored by the fact that Òmost of the things we do are unique even though they may have stereotyped componentsÓ (Tallis 2003, p. 67). Not surprisingly, human manual activity Ð despite its unique complexities Ð seems to depend on interactions between the brain, body and environmental interactions which involve

16

Radically Enactive Cognition in Our Grasp essentially the same kinds of dynamic interactive feedback and temporally extended engagements needed to explain the intelligent antics of behaviour-based robots and insects. It must be noted that concomitant with abandoning the standard ploy in favour of a REC based approach comes the admission that what we are dealing in most cases of

17

manual activity are not strictly speaking actions. This will surely be so if we operate with a strict concept of action Ð one that insists on a constitutive connection between actions and intentional states, where the latter are conceived of as requiring the existence of propositional attitudes of some sort. But all that follows from this, as Rowlands (2006) observes, is that Òmost of what we do does not count as actionÓ (p. 97). Respecting the stipulated criterion on what is required for action, many philosophers acknowledge the existence of non-intentional doings, motivated activities and/or deeds. For example, Velleman (2000) recognizes the need to:

define a category of ungoverned activities, distinct from mere happenings, on the one hand, and from autonomous actions, on the other. This category contains the things one does rather than merely undergoes, but that one somehow fails to regulate in the manner that separates autonomous human action from merely motivated activity (p. 4).

On the face of it, the great bulk of animal doings takes the form of sophisticated forms of highly coordinated, motivated activity that falls well short of action if acting requires forming explicit, if non-conscious, intentions and deliberate planning, at any level.

17

Radically Enactive Cognition in Our Grasp Picking up on FodorÕs earlier remark, far from being mere Ôthrashings aboutÕ or Ôreflexive behaviorsÕ, such unplanned engagements appear to be quite skillful, and sometimes even expert, dealings with the world. If REC has the right resources for

18

explaining the wide class of such doings then it has the potential to explain quite a lot of what matters to us when it comes to understanding mind and cognition.

3. The Non-Standard Ploy: Representationalism Rescued?

Despite all that has been said in favour of REC, many will baulk at going so far. There are weaker, and much more conservative and conciliatory ways of taking on board what is best in embodied and enactive ideas without abandoning intellectualism in a wholesale way. For example, intellectualists can happily accept that various facts about embodiment are causally necessary in making certain types of intelligent responding possible and in shaping its character without this concession in any way threatening the idea that cognition is wholly constituted by representational facts or properties. Trivially, it is clearly true that what a creature perceives depends on contingent facts about the nature of its sensory apparatus Ð thus bats, dolphins and rattlesnakes perceive the world differently and perceive different things because they are differently embodied. Moreover, no one denies that what and how we perceive causally depends on what we do Ð thus, it is only by moving my head and eyes in particular ways that certain things become visible and audible. Obviously, these truisms in no way threaten intellectualism. Things can be taken further still without rocking the boat too much. A more daring thesis, one that several authors have lighted upon, is that extended bodily states and

18

Radically Enactive Cognition in Our Grasp

19

processes might Ð at least on occasion Ð serve as representational or information-carrying vehicles. As such, they can play unique computational roles in enabling some forms of cognition (Clark 2008b). Or, in the lingo of Goldman and de Vignemont (2009) perhaps those attracted to embodied and enactive accounts of cognition should be taken as claiming that some mental representations are encoded in essentially bodily formats. These renderings of what enactive and embodied accounts have to offer are conservative with respect to a commitment to representationalism. They are perfectly compatible with asserting that Òthe mind is essentially a thinking or representing thingÓ (Clark 2008, p. 149); or that Òthe manipulation and use of representations is the primary job of the mindÓ (Dretske 1995, p. xiv). Without breaking faith with intellectualism, Conservative Embodied/Enactive Cognition, or CEC, still allows one to recognize Òthe profound contributions that embodiment and embedding makeÓ (Clark 2008b, p. 45). For those who endorse only CEC and not REC the new developments in cognitive science, far from posing a threat to the existing paradigm, can be seen as supplying new tools or Ôwelcome accessoriesÕ of considerable potential value that could augment intellectualist accounts of the mind. CEC-style thinking is best exemplified by a recent bid to save the representationalist baby from the embodied bathwater, by arguing for the existence of action-oriented representations, or AORs. According to Wheeler (2008), who has done more than most to promote this view, an action-oriented representation is one that is:

(i)

action-specific (tailored to a particular behaviour and designed to represent the world in terms of specifications for possible actions);

19

Radically Enactive Cognition in Our Grasp (ii) egocentric (features bearer-relative content as epitomized by spatial maps in an egocentric co-ordinate system); (iii) intrinsically context-dependent (the explicit representation of context is eschewed in favour of situated special-purpose adaptive couplings that

20

implicitly define the context of activity in their basic operating principles) (see also Wheeler 2005, p. 199).

Believing in AORs is consistent with accepting the neural assumption Ð an assumption that pays homage to the intuition that neural states and processes have a special cognitive status. Those attracted to this assumption believe it should be respected because, even though non-neural factors can qualify as representational vehicles, as it turns out, in most cases they do not. As such, the great majority cognitive explanations only ever involve representations that are wholly brain-bound. This is so even in cases in which it is necessary to making appeal to extra-neural but non-representational causal factors in order to explain the particular way that some particular intelligent activity unfolds. By accepting this last caveat, defenders of CEC allow that the full explanation of a given bout of intelligent behaviour need not be strongly instructional in character in the way demanded by the standard ploy. Wheeler (2005) highlights the core features of CEC-style thinking, illustrating the role of AORs by appeal to the architecture of a simple behavior-based robot created by Francheschini, Pichon and Blanes (1992).

20

Radically Enactive Cognition in Our Grasp The robot has a primary visual system made up of a layer of elementary motion detectors (EMDs). Since these components are sensitive only to movement, the

21

primary visual system is blind at rest. What happens, however, is that the EMD layer uses relative motion information, generated by the robotÕs own bodily motion during the previous movement in the sequence to build a temporary snap map of the detected obstacles, constructed using an egocentric coordinate system. Then, in an equally temporary motor map, information concerning the angular bearing of those detected obstacles is fused with information concerning the angular bearing of the light source (supplied by a supplementary visual system) and a directional heading for the next movement is generated (Wheeler 2005, p. 196, first, second, fifth and sixth emphases added).

Wheeler (2005) considers and dismisses a number of possible minimal criteria for being an AOR Ð including appeal to selectionist strategies and decoupleability. After careful review, he settles on the idea that what is necessary and sufficient to distinguish behaviour-based systems that operate with AORs from those that do not is that the former systems exhibit arbitrariness and homuncularity. A system exhibits arbitrariness just when the equivalence class of different inner elements is fixed Òby their capacity, when organized and exploited in the right ways, to carry specific items of information or bodies of information about the worldÓ (p. 218). A system is a homuncular just when (a) it can be compartmentalized into a set of hierarchically organized communicating modules, and (b) each of those modules performs a well-defined sub-task that contributes towards the collective achievement of the overall adaptive solution.

21

Radically Enactive Cognition in Our Grasp For Wheeler, the linchpin holding this account of AORs together is that some cognitive systems are information processing systems. Thus:

22

The connection between our two architectural features becomes clear once one learns that, in a homuncular analysis, the communicating sub-systems are conceptualized as trafficking in the information that the inner vehicles carry. So certain subsystems are interpreted as producing information that is then consumed downstream by other subsystems (p. 218, emphases added).

We can legitimately describe a cognitive system as employing AORs just in case it is a genuine source of adaptive richness and flexibility and it turns out that its subsystems use Òinformation-bearing elements to stand in for worldly states of affairs in their communicative dealingsÓ (Wheeler 2005, p. 219). Satisfaction of the above conditions is all that is required for the existence of weak or minimal representations. In the end, all of the weight in this account is placed on the idea that it suffices for minimal representations to be present in a system, S, if it manipulates and makes use of informational content in well-defined ways. This minimal notion of representation is, no doubt, attractive to cognitive scientists. For anyone in the field it is utterly textbook to be told that information is a kind of basic commodity Ð the raw material of cognition. After all, Òminds are basically processors of information; cognitive devices [for] receiving, storing, retrieving, modifying and transmitting information of various kindsÓ (Branquinho 2001, xii-xiii). There is great latitude for thinking about the processes that enable this. Thus:

22

Radically Enactive Cognition in Our Grasp

23

Mental representations might come in a wide variety of forms; there being no commitment in the claim itself to a specific kind of representation or to a particular sort of representational vehicleÉmental representations might be thought of as images, schemas, symbols, models, icons, sentences, maps and so on (Branquinho 2001, xiv).

Accordingly, representations or representational vehicles Òare items in the mind or brain of a given system that in some sense ÔmirrorÕ, or are mapped onto, other items or sets of items É in the worldÓ (Branquinho 2001, xiv). But what makes something into a vehicle, the essence of representing, is that they bear or possess content. Content is key. Thus:

The whole thrust of cognitive science is that there are sub-personal contents and subpersonal operations that are truly cognitive in the sense that these operations can be properly explained only in terms of these contents (Seager 1999, p. 27, emphasis added).

Dietrich and Markman (2003) define representations as Òany internal state that mediates or plays a mediating role between a systemÕs inputs and outputs in virtue of that stateÕs semantic content. We define semantic content in terms of information causally responsible for the state, and in terms of the use to which that information is putÓ (p. 97,

23

Radically Enactive Cognition in Our Grasp emphasis added). In sum, believing in AORs only requires acceptance of Òthe general idea of inner states that bear contentsÓ (Clark 2002, p. 386). Given this, it might be thought that accepting that at least some cognitive systems

24

employ AORs is a no-brainer. Certainly, it seems that this must be true of human manual activity of the sort described in Section 2 Ð activity of the sort that, if my argument goes through, would enable fans of REC to answer the Scope Objection. After all, even in cases in which that sort of activity is not supported by focused, conscious perception, Òthe motor activity of the hand Ð reaching, gripping and manipulation Ð cannot function in the absence of what is usually called Ôsensory informationÕÓ (Tallis 2003, p. 27). Indeed, Òthe information the hand needs to support its manipulative function is most clearly evident in the first stage É in reaching out prior to grasping, shaping, etc. Here the hand is under primarily visual control: the target is located, the relationship to the body determined, the motion initiated to home in on the target Ð these are all regulated [or, better, assisted] by sight, which measures what needs to be done and the progress of the doingÓ (p. 27). With this in mind, it looks like manual activity is surely dependent on information processing activity of the sort that would qualify as involving AORs Ð hence it is the sort of activity better suited to CEC rather than a REC treatment. If so, any ground gained by supporters of REC in Section 2 would be lost.

4. The Hard Problem of Content

Despite some obvious attractions, the AOR story and the support it lends to CEC over REC, isnÕt beyond question. Indeed, as this concluding section argues anyone who

24

Radically Enactive Cognition in Our Grasp favours CEC must face up to the hard problem of content Ð and I suggest that REC is a small price to pay to allow one to avoid that problem.

25

Before turning to that issue, it is worth highlighting an immediate concern about CEC, and its reliance on AORs. Appeal to AORs seems to secure the fate of minimal representations Ð winning a key metaphysical battle Ð only at the cost of losing a wider explanatory war. For, on the assumption that the AORs need not be decoupled in order to qualify as representations Ð an assumption Wheeler explicitly defends (2005, 2008), and with good reason (see Chemero 2009, ch. 3), defenders of CEC face the charge that Òtalk of representations in coupled systems may be too cheap, or too arbitrary, and thus adds little or nothing to an explanation of how these systems workÓ (Shapiro 2011, p. 147). Chemero (2009) too voices this exact worry, noting that:

the representational description of the system does not add much to our understanding of the system É [thus] despite the fact that one can cook up a representational story once one has the dynamical explanation, the representational gloss does not predict anything about the systemÕs behaviour that could not be predicted by dynamical explanation alone (p. 77).

Although initially cast as a purely explanatory concern it is clear that this issue cannot be kept wholly free of metaphysical considerations. For instance, Chemero goes on to note that Òin terms of the physics of the situation, the ball, the outfielder, and the intervening medium are just one connected thing. In effective tracking, that is, the outfielder, the ball, and the light reflected from the ball form a single coupled system. No

25

Radically Enactive Cognition in Our Grasp

26

explanatory purchase is gained by invoking representation here: in effective tracking, any internal parts of the agent that one might call mental representations are causally coupled with their targetsÓ (Chemero 2009, p. 114). Part of the trouble here is that there does not appear to be any clean cut way to decide, with precision, which systems actually satisfy the relevant conditions for being minimally representational systems. For example, there are diverging opinions about whether WattÕs much discussed centrifugal governor Ð a device originally designed to ensure a constant operating speed in rotative steam engines Ð qualifies as a representational device. This is despite the fact that the relevant parties in the debate are fully agreed about the characteristics of the governorÕs internal design, which are quite elegant and simple. The positions of the deviceÕs spindle arms interact with and modifies the state of a valve which controls the engineÕs speed Ð when the arm is high the valve slows the engine, when the arm is low the valve increases engine speed. In line with the criteria laid out in the previous section, Chemero (2009) concludes that:

It is possible to view the governorÕs arms as [noncomputational] representations É It is the function of particular arm angles to change the state of the valve (the representation consumer) and so adapt it to the need to speed up or slow down. The governor was so designed É to play that role É it is both a map and a controller. It is an action oriented representation, standing for the current need to increase or decrease speed (p. 71, emphases added).5

26

Radically Enactive Cognition in Our Grasp

27

Shapiro (2011) dissents. After careful review of this issue, he concludes that, ÒWattÕs design of the centrifugal governor requires that the angle of the flyball arms carry information about the flywheel speed in order to regulate the speed. Still, the function of the flyball arms is not to carry information, but rather to play a role in regulating the speed of the flywheelÓ (p. 147). The point is that in order to carry out its control function the spindle arms must covary with the relevant changes in engine speed. That they carry information in this sense is an unavoidable, collateral feature of their design that enables them to perform their regulatory work. Apparently, this is not sufficient to qualify as being a true information processor in the relevant sense Ð hence the governor is misdescribed as making use of AORs. Shapiro (2011) goes on to contrast the governor with other kinds of information using devices. For he holds, ÒSome devices surely do include components that have the function to carry information. The thermostat É is such a device. Thermostats contain a bimetal strip because the behavior of this strip carries information about [i.e. covaries with] temperature, and the roomÓ (Shapiro 2011, p. 148). The important difference is that although the governorÕs arms carry information about the flywheel arms, the governor does not use that information in order to perform its control tasks. This is meant to mark the subtle but utterly critical difference between merely complex systems and properly cognitive complex systems. Apparently, the thermostatic systems are designed to use information about the temperature as such in carrying out their work.

27

Radically Enactive Cognition in Our Grasp Put as starkly as this, one might be forgiven for failing to see what changes so

28

dramatically when it comes to the operation of thermostats as opposed to Watt governors. A thermostat regulates the systemÕs temperature, maintaining it at a desired point. Its mechanisms exploit the properties of the bimetallic strip that Ð when all is well Ð responds in reliable ways to temperature changes, bending one way if heated and the opposite way if cooled. The important difference between the two types of systems is not that there are more mechanisms or steps involved in regulating temperature by this means. Rather, the crucial distinction is meant to be that the bimetallic strips in thermostats have the systemic function of indicating specific desired temperatures to other subsystems that use those indications to regulate their behaviour. It is because they function in this special way that devices of this general type are representational Ð they exploit pre-existing indication relations giving them the function to indicate how things stand externally and use those indications in particular ways.6 Following Dretske (1995) if such devices were naturally occurring then they would Òhave a content, something they say or mean, that does not depend on the existence of our purposes and intentions É [They would] have original intentionality, something they represent, say, or mean, that they do not get from usÓ (p. 8, emphasis added). To qualify as representational, an inner state must play a special kind of role in a larger cognitive economy. Crudely, it must, so to speak, have the function of saying or indicating that things stand thus and so, and to be consumed by other systems because it says or indicates in that way. Only then will an internal state or structure meet RamseyÕs (2007) job description challenge.

28

Radically Enactive Cognition in Our Grasp It is plausible that many of the states (or ensembles of states) of systems that enable basic cognition are merely (1) reliably caused by (or nomically depend upon) the

29

occurrence of certain external features, and (2) disposed to produce certain effects (under specific conditions), and (3) have been selected because of their propensities for (1) and (2). Yet states or structures that only possess properties 1-3 fail to meet the job description challenge. They fail to qualify as truly representational mental states having the proper function of saying Ôthings stand thus and soÕ, rather they Ð like the Watt governor Ð only have the proper function of guiding a systemÕs responses with respect to specific kinds of worldly offerings. Exactly, what else is required to be a representation-using system? Wheeler (2005) speaks of the need for communicative transactions between homuncular subsystems. This informational dealing is the basis of true cognition Ð nonetheless, he stresses that this does not imply that the sub-systems Òin any literal sense understand that informationÓ (p. 218). Fair enough, but even if they literally lack understanding, it might be thought that at least such subsystems must be literally trading in informational content Ð using and fusing it Ð even if they donÕt understand what it says. But talk of using and fusing contents, although quite common, cannot be taken literally either. For it is not as if informational content is a kind of commodity that gets moved about and modified in various ways; information is not Òlike a parcel in the mailÓ (Shapiro 2011, p. 35). This being so it seems that bona fide cognitive systems are not special because they literally use and manipulate informational content (not even content that they donÕt

29

Radically Enactive Cognition in Our Grasp

30

understand). They are special, because it is their function to convey informational content without actually manipulating it as such. We are now getting down to brass tacks. For this story to work Ð there must at least be content that these subsystems have the special function to convey Ð there must be something that it is their function to say even if they donÕt understand what they are saying or what is said. But exactly what is informational content? Dretske (1981) speaks about informational content as Òthe what-it-is-we-can-learn from a signal or message in contrast to how-much-we-can-learnÓ (p. 47). He makes clear that he understands a signalÕs informational content to be a kind of propositional content of the de re variety. Propositions or propositional contents have special properties Ð minimally, they are bearers of truth. Assuming that informational contents are propositional is presumably what allows Dretske to hold that when signals carry information to the senses they tell Òus truly about another state of affairsÓ (p. 44). This is, of course, quite compatible with holding that informational content lacks fully-fledged representational properties. Thus one can hold that informational content is supplied by the senses, which is not representational content, and that more is required for informational content to be properly representational. It is at this juncture that defenders of AORs and CEC Ð at least those who subscribe to an explanatory naturalism Ð face a dilemma. Since so much hangs on this it is worth going very slowly over some familiar ground. In the opening passage of DretskeÕs Knowledge and the Flow of Information (1981) we find the locus classicus and foundational statement on how to understand information processing systems in a way

30

Radically Enactive Cognition in Our Grasp that is both required by the CEC story and that expresses a commitment to explanatory naturalism.

31

In the beginning there was information. The word came later ... information (though not meaning) [is] an objective commodity, something whose generation, transmission and reception do not require or in any way presuppose interpretative processes. One is therefore given a framework for understanding how meaning can evolve, how genuine cognitive systems Ð those with the resources for interpreting signals, holding beliefs, acquiring knowledge Ð can develop out of lower-order, purely physical, informationprocessing mechanisms ... Meaning, and the constellation of mental attitudes that exhibit it, are manufactured products. The raw material is information (p. vii, emphases added).

Any explanatory version of naturalism seeks to satisfy what Wheeler (2005) charmingly calls the Muggle constraint: ÒOneÕs explanation of some phenomenon meets the Muggle constraint just when it appeals only to entities, states and processes that are wholly nonmagical in character. In other words, no spooky stuffÓ (p. 5). It is widely supposed that the informational theory of content comfortably meet this constraint. At least, its defenders have attempted to convince us that, when promoting it, there is nothing up their sleeves. This is because, as Jacob (1997) emphasizes:

the relevant notion of information at stake in informational semantics is the notion involved in many areas of scientific investigation as when it is said that a footprint or a

31

Radically Enactive Cognition in Our Grasp

32

fingerprint carries information about the individual whose footprint or fingerprint it is. In this sense, it may also be said that a fossil carries information about a past organism. The number of tree rings in a tree trunk carries information about the age of the tree (p. 45, emphasis added).

This picks out the relevant notion by means of examples. We can call it the notion of information-as-covariance. Although theorists quibble about the strength and scope of the degree of covariance required in order for informational relations to exist, there is consensus that sÕs being F Ôcarries information aboutÕ tÕs being H iff the occurrence of these states of affairs lawfully, or reliably enough, covary. But hereÕs the rub. Anything that deserves to be called content has special properties Ð e.g. truth, reference, implication Ð that make it logically distinct from and irreducible to mere covariance relations holding between states of affairs. While the latter notion is surely scientifically respectable, it isnÕt able to do the required work of explaining content. Put otherwise, if information is nothing but covariance then it is not any kind of content Ð at least not if content of the sort defined in terms of its truth bearing properties. The number of a treeÕs rings can covary with its age; this does not entail that the first state of affairs says or conveys anything true about the second, nor vice versa. The same goes for states that happen to be inside agents and which reliably correspond with external states of affairs Ð these too, in and of themselves, do not ÔsayÕ or ÔmeanÕ anything in virtue of instantiating covariance relations. Quite generally, covariation in and of itself neither suffices for, nor otherwise constitutes content, where content

32

Radically Enactive Cognition in Our Grasp minimally requires the existence of truth bearing properties. Call this the Covariance doesnÕt Constitute/Confer Content (or CCC) principle. The CCC undermines the assumption that covariation is the worldly source of

33

informational content. There is no doubt the idea of information-as-covariance is widely used in sciences; hence, it is not a hostage to fortune for explanatory naturalists. But if the CCC is true, there is a gaping explanatory hole in the official story propounded by those who follow DretskeÕs lead. Anyone peddling such an account is surely violating the Muggle constraint and ought to be brought to the attention of the Ministry of Magic.7 One might opt for the first horn of this dilemma and retain the scientifically respectable notion of information-as-covariance, and thus retain oneÕs naturalistic credentials while relinquishing the idea there is such a thing as informational content. That is the path I recommend, but it requires giving up on CEC since Ð as argued in the previous section Ð the minimal requirement for distinguishing informational processing systems is that they make use of AORs which are defined as content-bearing vehicles. But the distinction between vehicles and contents falls apart, at least at the relevant level, if there are no informational contents to bear. To avoid this one might opt to be impaled on the second horn. This would be to accept that contentful properties exist even if they donÕt reduce to, or cannot be explained in terms of, covariance relations. If contentful properties and covariance properties are logically distinct they might still be systematically related. Hence, it might be hoped that contentful properties can be naturalistically explained by some other means (e.g. by some future physics). Alternatively, they could be posited as explanatory primitives Ð as metaphysical extras that might be externally related to covariance properties. Thus they

33

Radically Enactive Cognition in Our Grasp

34

might have the status that ChalmerÕs (2010) still assigns to qualia Ð they might require us to expand our understanding of the scope of the natural. Contentful properties might pick out properties that Ð like phenomenal properties Ð are irreducible to and exist alongside basic physical properties. If so, the explanatory project of naturalism with respect to them would look quite different Ð it would be to discover the set of fundamental bridging laws that explain how contentful properties relate to basic physical properties. That would be the only way to solve what we might call the hard problem of content. Of course, one might try to avoid both horns by demonstrating the falsity of the CCC by showing how contentful properties Ð e.g. truth bearing properties Ð reduce to covariance properties (Good luck with that!). A more plausible dilemma-avoiding move would be to show that the notion of information that is in play in these accounts is, in fact, meatier than covariance but is nonetheless equally naturalistically respectable. After all, Dretske talks of indication relations not covariance relations, though the two are often confused. Tellingly, in continuing the passage cited above Jacob (1997) remarks that ÒIn all of these cases, it is not unreasonable to assume that the informational relation holds between an indicator and what it indicates (or a source) independently of the presence of an agent with propositional attitudesÓ (p. 45, emphasis added). In making this last point, he stresses that Òthe information or indication relation is going to be a relation between states or factsÓ (p. 49-50). However, following Grice, Dretske is wont to think of indication as natural meaning Ð as in ÔsmokeÕ means ÔfireÕ. But smoke means fire only if it indicates fire to someone. It makes no sense to talk of indication in the absence of a user. Indication is, at least, a three-place relation whereas covariance by contrast, is a two-place relation. To think of

34

Radically Enactive Cognition in Our Grasp

35

indication as the basis for informational semantics therefore is already to tacitly assume that there is more going on than mere covaration between states of affairs.8 This raises questions about how exactly the notion of information-as-indication relates to its scientifically respectable cousin, the notion of information-as-covariance. Moreover, we might wonder if this notion has independent naturalistic credentials of its own. Until these questions are answered promoters of AORs and CEC Ð those that rely on the existence of informational content to distinguish genuine cognitive systems from all others Ð havenÕt really got off the starting blocks with their theorizing.

5. Epilogue: Decoding Information

The ÔcodeÕ metaphor is rife in the cognitive sciences but the cost of taking it seriously is that one must face up to the Hard Problem of Content! In light of the problems with CEC-style stories highlighted above, we have reason to think that on-line sensory signals Ôcarry informationÕ (in one sense) but not that they Ôpass onÕ meaningful or contentful messages Ð contentful information that is used and fused to form inner representations. Unless we assume that pre-existing contents exist to be received through sensory contact the last thread of the analogy between basic cognitive systems and genuinely communicating systems breaks down at a crucial point. In line with REC, there are alternative ways to understand cognitive activity as involving a complex series of systematic Ð but not contentfully mediated Ð interactions between well-tuned mechanisms (see, e.g., Hutto 2011a, 2011b; Hutto and Myin in preparation). RECers press for an understanding of basic mentality as literally constituted

35

Radically Enactive Cognition in Our Grasp by, and to be understood in terms of, concrete patterns of environmentally situated organismic activity, nothing more or less. If they succeed, the above arguments should

36

encourage more cautious CEC types Ð those trying to occupy the mid-left Ð to take a walk on this wild side.9

References

Araœjo, D. and K. Davids (2011). ÒWhat Exactly is Acquired During Skill Acquisition?Ó Journal of Consciousness Studies 18(3-4): 7-23. Branquinho, J. (2001). The Foundations of Cognitive Science. Oxford, Oxford University Press. Brooks, R. (1991a). ÒNew Approaches to Robotics.Ó Science 253: 1227-1232. Brooks, R. (1991b). ÒIntelligence without Representation.Ó Artificial Intelligence 47: 139-159. Chalmers, D. (2010) The Character of Consciousness. Oxford, Oxford University Press. Chemero, A. (2009). Radical Embodied Cognitive Science. Cambridge, MA, MIT Press. Clark, A. (2002). ÒSkills, Spills and the Nature of Mindful Action.Ó Phenomenology and the Cognitive Sciences 1: 385-387. Clark, A. (2008a). Supersizing the Mind: Embodiment, Action, and Cognitive Extension Oxford, Oxford University Press. Clark, A. (2008b). ÒPressing the Flesh: A Tension in the Study of the Embodied, Embedded Mind?Ó Philosophy and Phenomenological Research 76(1): 37-59. Dietrich, E. and A. Markman (2003). ÒDiscrete Thoughts: Why Cognition Must Use Discrete Representations.Ó Mind and Language 18: 95-119.

36

Radically Enactive Cognition in Our Grasp Donald, M. (1999). Preconditions for the Evolution of Protolanguages. The Descent of

37

Mind: Psychological Perspectives on Hominid Evolution. M. C. Corballis and S. E. G. Lea. Oxford, Oxford University Press. Dretske, F. (1981). Knowledge and the Flow of Information. Cambridge, MA, MIT Press. Dretske, F. (1988). Explaining Behaviour: Reasons in a World of Causes. Cambridge, MA, MIT Press. Dretske, F. (1995). Naturalizing the Mind. Cambridge, MA, MIT Press. Fodor, J. A. (2008). LOT 2: The Language of Thought Revisited. Oxford, Oxford University Press. Francheschini, N., J.-M. Pichon, et al. (1992). ÒFrom Insect Vision to Robot Vision.Ó Philosophical Transactions of the Royal Society series B (337): 283-294. Gallagher, S. (2008). ÒAre Minimal Representations Still Representations?Ó International Journal of Philosophical Studies. 16(3): 351-369. Gangopadhyay, N. (2011). ÒThe Extended Mind: Born to Be Wild? A Lesson From Action Understanding.Ó Phenomenology and the Cognitive Sciences 10(3): 377-397. Garz—n, F. C. (2008). ÒTowards a General Theory of Antirepresentationalism.Ó The British Journal for the Philosophy of Science 59 (3): 259-292. Goldman, A. I. (2009). ÒMirroring, Simulating and Mindreading.Ó Mind and Language 24(2): 235-252. Goldman, A. I. and F. de Vignemont (2009). ÒIs Social Cognition Embodied?Ó Trends in Cognitive Sciences 13(4): 154-159. Hurley, S. (2008). ÒThe Shared Circuits Model: How Control, Mirroring and Simulation can Enable Imitation and Mindreading.Ó Behavioral and Brain Sciences 27: 1-58.

37

Radically Enactive Cognition in Our Grasp Hutto, D. D. (2011). Enactivism: Why be Radical? Sehen und Handeln. H. Bredekamp and J. M. Krois. Berlin, Akademie Verlag: 21-44.

38

Hutto, D.D. (2011). ÒPhilosophy of MindÕs New Lease on Life: Autopoietic Enactivism meets TeleosemioticsÓ. Journal of Consciousness Studies 18(5-6): 44-64. Hutto D.D., Myin E. (in preparation). Radicalizing Enactivism. Cambridge, MA: MIT Press Jacob, P. (1997). What Minds Can Do. Cambridge, Cambridge University Press. Makin, G. (2000). The Metaphysics of Meaning. London, Routledge. Milner, D. and M. Goodale (1995). The Visual Brain in Action. New York, Oxford University Press. No‘, A. (2004). Action in Perception. Cambridge, MA, MIT Press. No‘, A. (2009). Out of Our Heads. New York, Hill and Wang. Ramsey, W. M. (2007). Representation Reconsidered. Cambridge, Cambridge University Press. Rowlands, M. (2006). Body Language. Cambridge, MA, MIT Press. Shapiro, L. (2011). Embodied Cognition. London, Routledge. Summers, J. J. and G. A. Anson (2009). ÒCurrent Status of the Motor Programme Revisited.Ó Human Movement Science 28: 566-577. Tallis, R. (2003). The Hand: A Philosophical Inquiry into Human Being. Edinburgh, Edinburgh University Press. Thompson, E. (2007). Mind in Life: Biology, Phenomenology, and the Sciences of the Mind. Cambridge, MA, Harvard University Press.

38

Radically Enactive Cognition in Our Grasp Van Gelder, T. (1995). ÒWhat Might Cognition Be, If Not Computation.Ó Journal of Philosophy 92: 345-381. Varela, F. J., E. Thompson, et al. (1991). The Embodied Mind: Cognitive Science and Human Experience. Cambridge, MA: MIT Press.

39

Velleman, J. D. (2000). The Possibility of Practical Reason. Oxford, Oxford University Press. Webb, B. (1994). Robotic Experiments in Cricket Phototaxis. From Animals to Animats 3: Proceedings of the Third Annual Conference on Simulation of Adaptive Behaviour. D. Ciff, P. Husbands, J. A. Meyer and S. W. Wilson. Cambridge, MA, MIT Press. Webb, B. (1996). ÒA Cricket Robot.Ó Scientific American 275(6): 62-67. Wheeler, M. (2005). Reconstructing the Cognitive World. Cambridge, MA, MIT Press. Wheeler, M. (2008). ÒMinimal Representing: A Response to Gallagher.Ó International Journal of Philosophical Studies. 16(3): 371-376.

1

This intellectualist way of understanding the basic nature of minds taps into a long tradition stretching back at least as far as Plato; it was revived by Descartes in the modern era, and regained ascendency, most recently, through the work of Chomsky during the most recent cognitive revolution. As No‘ (2009) observes:

39

Radically Enactive Cognition in Our Grasp

40

ÒWhat these views have in common Ð and what they have bequeathed to cognitive science Ð is the idea that we are, in our truest nature, thinkers. It is this intellectualist background that shapes the way cognitive scientists think about human beingsÓ (p. 98).

2

There are clear connections that might be forged between this view and DonaldÕs (1999) claim that when it comes to understanding human cognition Òthe most critical element is a capacity for deliberately reviewing self-actions so as to experiment with them É It would be no exaggeration to say that this capacity is uniquely human, and forms the background for the whole of human culture, including languageÓ (p. 142).

3

It is perhaps understandable that in seeking to make sense of this cognitive activity we are naturally inclined to assume the existence of representations that Òinclude not only ÔcommandsÕ and ÔcalculationsÕ, but also Ôif-thenÕ and other logical operations. This shows how it seems impossible to make sense of cerebral control Ð requisition and modification Ð of motor programs, to describe them in such a way that they deliver what is needed while avoiding anthropomorphismsÓ (Tallis 2003, p. 65). The problem is that Òattributing to the brain, or parts of it, or neural circuits, the ability to do things that we, whole human beings, most certainly cannot do, seems unlikely to solve the puzzleÓ (Tallis 2003, p. 65)

40

Radically Enactive Cognition in Our Grasp

41

4

At a pinch, one could give a short list of these, which could include: Ògrasping, seizing, pulling, plucking, picking, pinching, pressing, patting, poking, prodding, fumbling, squeezing, crushing, throttling, punching, rubbing, scratching, groping, stroking, caressing, fingering, drumming, shaping, lifting, flicking, catching, throwing, and much besidesÓ (Tallis 2003, p. 22).

5

Notably, Chemero holds that the centrifugal governor is not a computer even though it can be regarded as a representational device and in the respect he does not break faith with the conclusion of Van GelderÕs original analysis when he first introduced the example into the literature (Van Gelder 1995).

6

Thus ÒIf we suppose that, through selection, an internal indicator acquired a biological function, the function to indicate something about the animalÕs surroundings, then we can say that this internal structure representsÓ (Dretske 1988, p. 94).

7

To make vivid what is at stake it is worth noting that early analytic philosophers were at home with the view that the world is ultimately and literally composed, at least in part, by ÔpropositionsÕ. These were conceived of as bedrock Platonic entities Ð mentionable ÔtermsÕ which, when standing in the right complex relations, constitute judgeable objects of thought. In commenting on RussellÕs version of this idea, Makin underscores the features that parallel many of the properties that Dretske demands of informational content. He stresses that: Òwith

41

Radically Enactive Cognition in Our Grasp

42

propositions, it is crucial to bear in mind that they are not, nor are they abstracted from, symbolic or linguistic or psychological entities É On the contrary, they are conceived as fundamentally independent of both language and mind. Propositions are first and foremost the entities that enter into logical relations of implication, and hence also the primary bearers of truth ... Ôtruth'Õ and 'implication' apply, in their primary senses, to propositions and only derivatively to the sentences expressing themÓ (Makin 2000, p. 11, emphasis added).

8

Others too have noticed this. For example, Ramsey (2007) comments on the peculiar features of the quasi-semantic indication relation as follows: ÒDretske and many authors are somewhat unclear on the nature of this relation. While it is fairly clear what it means to say that state A nomically depends upon state B, it is much less clear how such a claim is supposed to translate into the claim that A is an indicator of B, or how we are to understand expressions like Ôinformation flowÕ and Ôinformation carryingÕÓ (p. 133).

9

I am not alone in trying to persuade those in the CEC brigade to make this shift, see also Gangopadhyay (2011).

42