You are on page 1of 10

1132691

research-article2022
PST0010.1177/27538699221132691Possibility Studies & SocietyGallagher

POSSIBILITY
STUDIES
Inaugural Issues: Possibility Studies, Past, Present and Future - Enacting Chance & SOCIETY
Possibility Studies & Society

Surprise! Why enactivism and 1­–10


© The Author(s) 2022

predictive processing are parting ways: Article reuse guidelines:


sagepub.com/journals-permissions

The case of improvisation DOI: 10.1177/27538699221132691


https://doi.org/10.1177/27538699221132691
journals.sagepub.com/home/pst

Shaun Gallagher
University of Memphis, USA
University of Wollongong, Australia

Abstract
Can we explain how the various factors of knowledge, skill, habit, environmental constraints and affordances
interact or integrate in improvisational performance? In attempting to explain how this integration takes
place, I’ll consider two possible approaches: predictive processing (PP) and enactivism. I’ll argue that PP,
which, on a neuroscientific view, conceives of the mind as set up to avoid surprise, will not be able to
explain improvisation if it remains true to its own principles. In contrast, I’ll argue, enactivism, as a form of
embodied cognition that takes the explanatory unit to be the brain-body environment, can offer a better
explanation of improvisation. I’ll also argue that the notion of habit is central to this account.

Keywords
Enactive cognition, habit, improvisation, performance, predictive processing

In the case of jazz and dance improvisation there is performance. The further question, however, is
a skill-informed selective uncertainty, that is, you precisely how this integration takes place. In this
need to know a lot about your craft, about different paper I’ll consider two possible approaches to
traditions within your genre, and have skilled answering this question: predictive processing (PP)
know-how, but you don’t know what you are going and enactivism. I’ll argue that PP, which conceives
to do or where the music or movement will take of the mind as set up to avoid surprise, will not be
you. In improvisation one aims for surprise (self- able to explain improvisation if it remains true to
surprise), using techniques that allow you to its own principles. In contrast, I’ll argue, enactiv-
achieve novelty (such as the breaking of habits or ism can offer a good explanation of improvisation.
the avoidance of usual affordances). But also, part
of what allows for the possibility of surprise is
Predictive processing
chance, the source of which may be the environ-
ment or other performers. (a rough and simple guide)
Can we explain how these various factors— Predictive processing continues a long tradition,
knowledge, skill, habit, environmental factors, and going back to Helmholtz (1867), of understanding
affordances—interact? I’ve argued that the model perception as an inferential process. It models this
of a meshed architecture (Christensen et al., 2016;
Gallagher 2021) provides a good starting point. It
Corresponding author:
provides a way to map out the cognitive, motor- Shaun Gallagher, Department of Philosophy, University of
control, environmental, social, cultural, and nor- Memphis, Clement Hall 327, Memphis, TN 38152, USA.
mative factors that need to be integrated in Email: s.gallagher@memphis.edu
2 Possibility Studies & Society 00(0)

process in terms of prediction error minimization fundamental way open or porous to the world, or on
(PEM), which begins with the assumption that the which it is in some strong sense embodied, extended
brain has no access to the external world. By means or enactive. Instead, PEM reveals the mind to be
of perceptual inference, neuronal processes repre- inferentially secluded from the world, it seems to be
more neurocentrically skull-bound than embodied or
sent that world by forming probabilistic hypothe-
extended, and action itself is more an inferential
ses about the world based on an internal Bayesian
process on sensory input than an enactive coupling
(statistical), hierarchical, generative model that it with the environment. (Hohwy, 2016, p. 259)
constructs informed by prior knowledge (priors).
The prediction structure can be quite complex, Although Andy Clark endorses a more liberal
involving a nested cascade of precision-weighted embodied perspective, he also maintains the idea
predictive processes in the brain. If the predictions that the brain is in some sense isolated from the
don’t match sensory input, the brain revises its world.
model and corrects its inferences in order to mini-
mize prediction errors (conflicts between sensory [The brain] must discover information about the
input and the internal model). likely causes of impinging signals without any form
of direct access to their source. . .. [A]ll that it
The human brain, PP here suggests, commands a rich, “knows,” in any direct sense, are the ways its own
integrated model of the worldly sources of sensory states (e.g., spike trains) flow and alter. In that
inputs, and uses that long-term model to generate (restricted) sense, all the system has direct access to
on-the-spot predictions about the probable shape and is its own states. The world itself is thus off-limits.  .  .
character of current inputs. The rich, integrated (Clark, 2013, p. 183)
(generative) model takes a highly distributed form,
spread across multiple neural areas that may The brain’s task is to take “patterns of neural acti-
communicate in complex context-varying manners. vation and, on that basis alone, infer properties of
(Clark, 2018, p. 522).
the stimulus” (Clark, 2016). Still, Clark allows for
a more extended concept of the cognitive system,
The aim is to reduce surprisal, that is, the unex-
namely by emphasizing active inference. In con-
pected variation from the general statistical pat-
trast to perceptual inference, active inference
terns that are expected to hold in a situation.
involves holding its world-model steady, but
Surprisal is a subpersonal version of surprise.
engaging in bodily movement or action in a way
The idea that the organism (or brain) has no
that changes the world, thereby changing sensory
direct access to the world is usually explained in
input to gain a match with the model. This is an
terms of a mathematical formalism called a Markov
alternative way of reducing prediction error.
blanket—which defines the boundary of the cogni-
tive system. Where do you draw that boundary? [Active] “inference,” as it functions in the [PP] story,
According to one well-known neurocentric is not compelled to deliver internal states that bear
account, the boundary does not extend beyond the richly reconstructive contents. It is not there to
brain (Hohwy, 2013); on another account (Clark, construct an inner realm able to stand in for the full
2016), the Markov blanket may extend to include richness of the external world. Instead, it may deliver
the body and parts of the environment. Accordingly, efficient, low-cost strategies whose unfolding and
one point of debate within the PP approach con- success depend delicately and continuously upon the
cerns whether the boundary should be understood structure and ongoing contributions of the external
as cutting the cognitive system off from the world, realm itself as exploited by various forms of action
and intervention. (Clark, 2016, p. 191).
or as allowing for a coupling or interaction of sys-
tem and world?
According to Jakob Hohwy, there is a strict Active inference is often framed in terms of the
closure. “free energy principle,” which is an alternative and
more general way of describing PEM. I will return
PEM should make us resist conceptions of [a mind- to this point, but I note here that this moves PP
world] relation on which the mind is in some closer to the enactive starting point, which involves
Gallagher 3

a self-generating (autopoietic) system that couples features of improvisation. As such, if PP is to help


with the environment for purposes of sense our understanding of how improvisation functions,
making. it will need to include an account of novelty. But
here’s the problem.
Can PP explain improvisation?
1. According to PP, an organism’s main aim is
Clearly, prediction is important for improvisa- to maintain its organization (i.e., stay alive),
tion—to improvise and deal with uncertainty, a and the means by which an organism real-
performer must project ahead in time to anticipate izes this aim is through prediction-error
a variety of things. Performers need to anticipate minimization, which is equivalent to mini-
under pressing time constraints, what comes next, mizing surprise/surprisal as much as
including the actions of others with whom they are possible.
dancing or playing music. It seems clear that PP 2. Novelty is something that by definition
could be an approach that helps to explain improvi- increases surprise, moving outside the
sation. For example, according to Clark (2016): standard statistical regularities faced by an
“[t]o deal rapidly and fluently with an uncertain organism.
and noisy world, brains like ours have become 3. Accordingly, an organism will do its best to
masters of prediction – surfing the waves’ noisy stay away from novelty as much as
and ambiguous sensory stimulation by, in effect, possible.
trying to stay just ahead of them” (p. xiv).
Schaefer (2015) offers just such a PP account of Philipp Schwartenbek and colleagues phrase the
musical processing. As an individual responds to question in the right way.
music, she applies a mental model (prior knowl-
edge), “an internal representation of a percept or If our main objective is to minimize surprise over the
action, as it is built up through experience and sta- states and outcomes we encounter, how can this
tistical learning” (2014), in order to predict the explain complex human behavior such as novelty
incoming flows of sensory information. Michael seeking, exploration, and, furthermore, higher level
and Wolf (2015) apply this account to the example aspirations such as art, music, poetry, or humor?
of coordination required for ending a song during (Schwartenbeck et al., 2013, p. np)
an improvised performance. How do several musi-
cians decide when and how to end a song without Improvisation seems a counterexample to the prin-
agreeing on it in advance? “In the context of music, ciple that organisms avoid surprise: people actively
two jazz performers might both prefer, having just seek out and enjoy novelty. They often find it
finished what appeared to be the final verse of a pleasurable. One possible response to this objec-
jazz standard, to start up again for a surprising tion is that surprise (at the personal level) is not the
additional repetition of the chorus – but only if the same as surprisal (at the subpersonal level) (Clark,
other does so as well” (2014). They suggest that the 2018). Indeed, surprise may actually come about
musicians’ priors include a set of stock endings that because of a tension between the system’s predic-
are common across a performance community; tions meant to limit surprisal, and the surprising
these operate as constraints on prediction, and this, event. Moreover, we might actually predict affec-
together with embodied and gestural cues can help tive surprise in specific environments.
navigate the end of a performance, and limit sur- According to Schwartenbeck et al. (2013), to the
prise (for the performers). extent that our behavior involves active inference, “the
Ryan (2020), however, points out that this can- agent’s generative model includes hidden (future)
not explain novelty in improvisation (also see states and actions that the agent might perform (and
Froese & Ikegami, 2013). Novelty and surprise are their consequences). This implies that the agent has to
not coextensive concepts, yet both are important represent itself in future states performing specific
4 Possibility Studies & Society 00(0)

actions. In other words, it necessarily implies a model affectivity. This response again, I think, pushes PP
with a sense of agency”. Schwartenbeck et al., apply toward a more enactive-embodied view to the
this explanation to economic decision making. The extent that it involves affectivity. Barrett and Bar
representing of itself in future states means that the help to clarify a more embodied PP, which, they
agent infers a “policy” (an action selection rule that argue, is consistent with views on active inference.
entails a sequence of actions) that minimizes surprise Their affective prediction hypothesis suggests that
about future outcomes. Such policies are defined by “responses signaling an object’s salience, relevance
fixed goal states. or value do not occur as a separate step after the
object is identified. Instead, affective responses
These goal priors are fixed and do not depend on support vision from the very moment that visual
sensory input: they represent a belief concerning stimulation begins” (2009). Along with the earliest
states the agent will end up in. Desired goal states visual processing, a train of muscular and hormo-
will be accorded a high probability (log-likelihood) nal changes occur throughout the body, generating
of being encountered, resulting in a low surprise
“interoceptive sensations” from organs, muscles,
when this state is indeed visited. Undesirable states,
by contrast, will be assigned with a low prior
and joints associated with prior experience, which
probability and therefore become highly integrates with current exteroceptive sensory infor-
surprising. . .. In the framework of active inference, mation. The organism as a whole (or at least in spe-
therefore, agents do not try to maximize reward but cific aspects that include its morphology, its
minimize surprise (about future states). affectivity, its skill level) is included as an impor-
tant part of the priors. Friston (2013) suggests, in
Improvisational performance, however, as we’ll this regard, that rather than having a model, the
see below, sometimes involves a sense of distrib- organism is the generative model. “We must here
uted agency, and does not necessarily involve a understand ‘model’ in the most inclusive sense, as
definite fixed goal state. If this account of active combining interpretive dispositions, morphology,
inference can deal with economic decision making and neural architecture, and as implying a highly
where alternative actions are clearly delineated, it’s tuned ‘fit’ between the active, embodied organism
not clear that it can capture the type of uncertainty and the embedded environment” (Friston et al.,
involved in improvisation. 2012, p. 6). In perception, then, bodily affective
Referencing Schwartenbeck et al., Clark (2018) changes are integrated with sensorimotor process-
addresses the problem directly. Indeed, he agrees ing so that before we fully recognize an object, an
that this kind of solution does not reach the type of affordance, or another person for what it or he or
improvisational behavior we are concerned with. she is, our bodies are already configured into over-
all peripheral and autonomic patterns shaped by
These prediction-error-minimizing agents exhibit at prior associations, responding and contributing to
most a modest and instrumentally-motivated shaping subsequent actions.
tendency towards play, exploration, and the search This means that perception or action, is more
for novel experiences. These prediction error mini­ than a nested cascade of predictions in the brain.
mizing agents remain locked, it seems, into an
Indeed, it’s not clear how an affective, pleasurable
information-theoretic journey whose guiding prin­
enjoyment of surprise or novelty fits with the nar-
ciple is in some way unacceptably conservative. It is
a journey which, if successful, will be marked only row conception of PEM. Even if there may be pre-
by the attainment of expected goals and meta-goals. diction errors involved if our action or expression
(p. 528). doesn’t go as planned, it’s not clear that there are
error criteria for pleasure or enjoyment. More gen-
He suggests that information-theoretic imperatives erally, the improvisational process poses a chal-
don’t sufficiently address the problem. lenge for a PP representationalist account since in
In this respect Clark cites the work of Barrett the process of improvisation we are not trying to
and Bar (2009), and the important role of model the world as it is.
Gallagher 5

Although Clark (2018) tries to keep affectivity 3. Cognitive processes acquire meaning in
purely within the brain (in terms of “interoceptive part by their role in the context of action,
predictive coding”—an interchange between corti- rather than through a representational map-
cal and sub-cortical processes), ultimately, he ping or replicated internal model of the
admits, “As your bodily states alter, the salience of world
various worldly opportunities alters too.” In the 4. Cognitive systems are ecologically embed-
end he appeals to cultural constraints, “ecologi- ded, intersubjective, and socially situated
cally unique, self-engineered contexts of culture, (vs. classic cognitive science, which tends
technology, and linguaform exchange”—all of this, to be internalistic)
arguably, outside the brain and not entirely within 5. Dynamical systems theory can model the
our control. Clark considers this idea that such cul- dynamical coupling and coordination
tural constraints may actually constrain and limit across brain–body–environment
possibilities for improvisation and novelty. 6. More complex cognitive functions are
grounded in sensorimotor coordination,
The skilled pianist has learnt to reduce prediction error affective, and autonomic aspects of the full
with respect to complex melodies and motor repertoires, body.
and the skilled mathematician with respect to properties 7. Complex cognitive functions, such as
and relations among numbers, theorems, and other reflective thinking or deliberation, are exer-
constructs. But the musical and mathematical traditions
cises of skillful know-how.
within which they operate reflect the operation of
cultural forces such as practices of writing, reflecting,
disseminating, and peer review. (Clark, 2018) How does an enactive approach explain improvisa-
tion? In contrast to PP, which specifies an impor-
Clark argues that it could go either way—environment tant role for top-down predictions within the
and cultural practices may provide limits, or may ena- individual system, enactivism takes improvisation
ble creativity. The question remains, how does it work to involve distributed control across brain, body,
when it enables creativity? And then, almost as if on and physical and social environment, something
cue: “These powerful effects are further explored in closer to active inference, but where the individual
work by ‘enactivists’ sympathetic to PP—for exam- is one dynamically coupled component and not in
ple, Rietveld and Kiverstein (2014), Bruineberg et al. full control.
(2018), Gallagher et al. (2013)” (Clark, 2018). That As Michael Kimmel puts it:
seems exactly right; the focus on embodiment, affect,
and environmental coupling, actually pushes us (away The phenomenological consequence of distribu­
from PP) toward more enactive accounts. tedness is ambiguous “ownership of the movement.”
Occasionally, dancers would report a, what we call,
“wasn’t me” moment: they sense that neither of the
Enactivism and dynamical dancers has initiated the movement. A dancer
processes explicitly described the experience as follows: “there
are ways and times that we aren’t separate agencies.
Enactivism is an embodied (action-oriented, anti- When we come together and the fact of togetherness
representationalist) approach to cognition. Gall­ starts to drive what happens.” (Kimmel et al., 2018,
agher (2017) has summarized enactive approaches p. 24).
in seven principles.
This same distributed control or distributed agency
1. Cognition is not simply a brain event. It where control or agency is an “emergent product of
emerges from processes distributed across agent environment coupling” (Kronsted, 2021) can
brain–body–environment. be found in some forms of improvisation (see also
2. The world (meaning, intentionality) is not Høffding, 2019; Schiavio et al., 2022; also van der
pre-given or predefined, but is structured Schyff et al., 2018; Walton et al., 2018). Factors in
by perception and action the improvisational setting involve not only the
6 Possibility Studies & Society 00(0)

performer as an agent with specific intentions, but solicitation of an affordance, or acts on the
a meshed architecture of physical, social, cultural, affordance in a nonhabitual manner. What fills this
normative factors that operate in a dynamical, non- gap is not a surety of what happens next, but a let-
linear fashion, in which an agent’s coupling or ting happen of whatever happens, a trust in the pro-
attunement with the environment is modulated cess. How the improvisation unfolds is a product of
both by sensorimotor schemes and a broad range of the whole system, which includes the performer’s
environmental features. This may also involve agentive intentions, rather than any one individual
habitual behaviors, although habit should not be component; the system changes its internal states
understood to mean automatic response. Instead of and, at the same time, by means of action, changes
blind automatic repetition, habit is an open and the environment in ongoing dynamical feedback
adaptive way in which an autonomous agent learns loops (contra any strict Markov blanket).1
to cope with familiar or unfamiliar situations.
Merleau-Ponty (2012), for example, argues that a
habit is developed when the body “acquires the PP and/or enaction?
power of responding with a certain type of solution Isn’t it possible to integrate predictive processing
to a certain form of situation” (p. 143). Habit accounts and enactivism, as some have suggested?
involves intelligent agentive response, where intel- (Gallagher & Allen, 2018; Kiverstein, 2020; Parr
ligence is built into the movement. John Dewey et al., 2020; Ramstead et al., 2021; Wiese & Friston,
likewise distinguishes between intelligent and rou- 2021). Ramstead et al., addressing some of the ter-
tine habit. minological and conceptual differences, suggest
that proponents of PP don’t really mean inference
Repetition [i.e., automaticity] is in no sense the
essence of habit. . .. The essence of habit is an
when they say “inference,” or representation when
acquired predisposition to ways or modes of they say “representation.” This may be what Cicero
response. . .. Habit means special sensitiveness or calls “alterations of terminology rather than of sub-
accessibility to certain classes of stimuli, standing stance.” More substantial conceptual alterations
predilections and aversions, rather than bare may also be in order, however. Gallagher and Allen
recurrence of specific acts. (Dewey, 1922, p. 42). (2018) have argued that one has to give up neuro-
centric internalism and the concepts of inference,
If habit were strictly automatic there would be no internal model, and representation, and go with
possibility of doing something different. Habit enactivist processes of affordance, attunement, and
involves a heedful attitude; and only because it resonance. One might then ask, however, what is
does can one break habit in the improvisational set- left of PP?
ting. Both habitual behavior and the breaking of Constant et al. (2020) propose a truce that would
habit involve agentive intention. Improvisation can introduce a division of labor. “[There] is a subtle,
happen when a performer resists the usual yet crucial point, which becomes apparent when
affordance, or breaks her habitual move, gives up considering the probability distributions involved
control (although in a controlled fashion, so to in various inference processes in the brain” (p. np).
speak), and then lets other factors take over. This Here they make the distinction between (1) the
breaking of habit (declining the usual affordances), brain predicting expected sensory neural activa-
which happens in the immediacy of performance, tions matching the probability of the generative
comes without a prediction of what happens next, model/belief—a purely internal process, and (2)
except the most general prediction or experienced the prediction made about the (hidden) cause of
anticipation that something will happen (Kronsted such activations. “Getting it right” (truth or success
& Gallagher, 2021). The phenomenology of agency conditions) in (1) “is about getting it right with
attests to the idea that the improvisation is not com- respect to one’s own beliefs; e.g., successfully
pletely under the performer’s control (Høffding, exploring the state space of one’s own model of the
2019). There is an experienced opening, or “gap” world. . .. This means that under active inference,
(McDowall, 2018), when the agent resists the there are two layers of success involved, one
Gallagher 7

defined over the model, and one defined over the of entropy, that is, systemic death. To maintain
agent-world coupling.” homeostasis, and avoid entropy (which in thermo-
According to Constant et al., the first is repre- dynamic terms means too much free energy in the
sentational and fits well with PEM; the second is system), then you should reduce free energy (which
dynamical and could be modeled on active infer- means reducing prediction errors and surprisal).2
ence or enactive processes. Specifically, they To survive, an organism needs to avoid surprise.
argue, the second (non-representational) process is The claim is then that FEP aligns well with the con-
sufficient for deontic actions—automatic behavior, cept of autopoiesis, the emergent or self-organizing
which for them includes habitual actions (such as persistence of an organism in virtue of its own
stopping at stop signs). dynamical structure. On the enactivist view,
autopoiesis characterizes the continuous processes
[Such actions] do not have success conditions qua of life and mind.
brain processes, but rather have success conditions The strategy here, to put it simply, is that if FEP
qua agent-world coupling processes. They are simple and autopoiesis can be understood to be consistent,
observation-action loops; not rich and reconstructive then there is a good basis for integrating PP and
policy selection loops.... inferences about states of enactivism. This is the view of some enactive-
the world – that admit a representationalist
inclined authors who think we can mesh the basic
interpretation – are now replaced by direct action,
without any intervening inference or representation
FEP version of predictive active inference models
of the consequences of action. with enactivist principles (e.g., Bruineberg et al.,
2018). They emphasize the embodied action that
the organism uses to control its own viability
In the case of deontic actions what does the work is conditions.
“a kind of perceptually maintained motor-informa-
tional grip on the world: a low-cost perception- So within the free-energy framework, it is action that
action routine that retrieves the right information does the work of actually minimizing surprisal.
just-in-time for use, and that is not in the business Actions change an organism’s relation to the
of building up a rich inner simulacrum” (Clark, environment, thereby changing the sensory states of
2016, p. 11). the organism. . .. (Bruineberg et al., 2018)
Although Constant et al. (2020) pitch this as a
kind of détente, it’s not clear how it moves beyond Action is not something happening in the brain,
older peace negotiations that involved splitting the and it’s not just providing new sensory input for
difference between higher-order representation- neural processing; it’s what the whole organism
hungry cognition (albeit now framed in PP terms) does in its interactions with the environment, or
and more basic perceptual-motor processes (e.g., under a different description, what a person does in
Clark & Toribio, 1994), or how enactive, heedful the world, and this changes the world as much as it
or intelligent habit fits this picture. changes the brain. On this view, the priors that
Another attempt at integrating PP and enactive inform action are not assumptions or beliefs that
approaches is made by stepping back from some of inform inferences; they’re embodied skills, pat-
these details into a more basic and abstract frame- terns of action-readiness, and affective dispositions
work. For Karl Friston, the important thing is get- that mesh with an affordance space.
ting the mathematics (in the statistical models) In contrast to these various attempts to integrate
right, rather than philosophical interpretations that PP and enactivist approaches, Di Paolo et al. (2022)
might range from internalist to embodied and enac- identify some deeper (technical and theoretical)
tive versions. Furthermore, for Friston, predictive reasons why PP is not compatible with enaction. To
models generalize to life itself explained in terms put this in its simplest terms, the technical part of
of the free energy principle (FEP). According to the argument involves contrasting autopoietic sys-
FEP, biological systems are defined by the ten- tems with FEP systems. In autopoietic systems
dency to resist the second law of thermodynamics; organization, as an invariable, is distinguished
to do otherwise would entail an unbounded increase from structure, as a changeable feature—a
8 Possibility Studies & Society 00(0)

distinction not maintained in the FEP-Markov understood in terms of heedful, agentive, intelli-
blanket discussion. So there is a confusion when gent behavior, allows for the possibility of break-
FEP systems are equated with autopoietic ones. ing habit in the improvisational setting.
There is a similar confusion between autopoietic On the enactive view, improvisation is thus not
operational closure (according to which the system equivalent to a deontic action modeled on an FEP
nonetheless remains open to the environment) and nonequilibrium steady-state.
the formalism of the Markov blanket (which sug-
gests statistical independence from the environ- [Thus], phenomena such as developmental spurts in
ment and structural boundaries). Perhaps more skill level signal changes in dynamical configurations
relevant to the issue of improvisation, as I’ll try to (novel constraints, emergent parameters, changing
make clear, there is also a contrast between the variable sets). The variability entailed in changing
dynamical configurations has been postulated as the
non-equilibrium steady-state features of FEP sys-
origin of motor creativity (Orth et al., 2017), the very
tems, and the historicity of (the importance of the idea of which is rendered problematic on the
system’s history for) the enactive system. Di Paolo assumption of non-equilibrium steady states [as in
et al. (2022) make this last point more generally in FEP]. (Di Paolo et al., 2022)
terms of different conceptions of brain dynamics
(hierarchical comparators in PP, vs. dynamical Motor creativity—essential for improvisation in
“history-dependence even in the most basic neuro- jazz and dance—reflects the history and skill of the
scientific scenario of stimulus processing”). The agent. Performers, based on their well-trained
point about historicity, has implications for under- skills, and well-formed habits (which involve a
standing development, plasticity, skill acquisition, heedful flexibility rather than automaticity or
habit formation, and the ability to break habits— repetitiveness) are able to move beyond controlled
which brings us back to our topic—improvisation engagement to the point of not-knowing (embrac-
in performance. ing a kind of uncertainty or surprise) about what
precisely will happen—letting the system (brain-
body-environment) move in unpredictable, surpris-
Improvisation ing ways—without a prediction of what happens
One way to cash out in pragmatic terms, the theo- next. What happens next is that brain-body-envi-
retical incommensurability that Di Paulo et al. ronment couple in a novel way—they join forces to
point to, is to consider the example of improvisa- enact something unpredictable—they create corti-
tion. The contrast between PP/FEP models and cal patterns, and behaviors, and new affordances
enactive approaches becomes very clear. Impro­ that are unique to each event.
visation is the opposite of deontic action, as char- Enactivism explains the possibility of breaking
acterized by Constant et al. (2020). Recall that habits (declining usual affordances, resisting predic-
deontic actions are said to be automatic, habitual tions) across the various timescales of performance,
actions, “simple observation-action loops; not rich generating higher degrees of uncertainty about what
and reconstructive policy selection loops. . .. with- will happen. Such processes may differ from one
out any intervening inference or representation of genre to another, and from one situation to another
the consequences of action.” They are good exam- (jazz vs. dance improvisation (see e.g., Kronsted &
ples of nonequilibrium steady-states. As Constant, Gallagher, 2021; Ryan, 2020; Ryan & Gallagher,
Clark and Friston put it: “the back story to active 2020) or solo performance versus group performance
inference shows that this kind of dynamical behav- (see e.g., Schiavio et al., 2022; Van der Schyff &
ior is a necessary aspect of any self-organization to Krueger, 2019). If the deeper principles of FEP fail to
nonequilibrium steady-state in any random dynam- account for such processes, the assumptions made by
ical system that possesses a Markov blanket.”3 The PEM also seem unable to provide an account for how
conception of habit, as a kind of automatic, repeti- the system pursues novelty, creativity.
tive action, in this context is antithetical to enactive For both theoretical and pragmatic reasons,
conceptions of habit (see above). Again, habit, then, PP and enactive views offer different
Gallagher 9

explanations. We can cash out this difference by of clarity in the logical structure of the reasoning
looking specifically at how these accounts apply to leading to FEP” (Colombo & Wright, 2018, p. 2).
improvisation. To the extent that PP, following the 3. Or as Friston puts it: “Any ergodic random dynami-
FEP, emphasizes steady-state equilibriums, the cal system that possesses a Markov blanket will
appear to actively maintain its structural and
automaticity implied in deontic actions, and the
dynamical integrity” (Friston, 2013, p. 2)
minimization of uncertainty (novelty, surprise), it
is difficult to see how it can capture the dynamical
creative configurations (involving novel con- References
straints, emergent parameters, changing variable
Barrett, L. F., & Bar, M. (2009). See it with feeling:
sets) which, according to enactivist accounts, are Affective predictions during object perception.
characteristic of skilled performance and Philosophical Transactions of the Royal Society B:
improvisation. Biological Sciences, 364(1521), 1325–1334.
Bruineberg, J., Kiverstein, J., & Rietveld, E. (2018). The
Declaration of conflicting interests anticipating brain is not a scientist: The free-energy
The author declared no potential conflicts of interest principle from an ecological-enactive perspective.
with respect to the research, authorship, and/or publica- Synthese, 195(6), 2417–2444.
tion of this article. Christensen, W., Sutton, J., & McIlwain, D. J. (2016).
Cognition in skilled action: Meshed control and the
varieties of skill experience. Mind & Language,
Funding
31(1), 37–66.
The author disclosed receipt of the following financial Clark, A. (2013). Whatever next? Predictive brains, situ-
support for the research, authorship, and/or publication ated agents, and the future of cognitive science.
of this article: Support for this research was provided by Behavioral and Brain Sciences, 36(3), 181–204.
the Australian Research Council (ARC) project, minds Clark, A. (2016). Surfing uncertainty: Prediction, action,
in skilled performance, grant number: DP170102987; and the embodied mind. Oxford University Press.
and the Moss Chair of Excellence in Philosophy at the Clark, A. (2018). A nice surprise? Predictive processing
University of Memphis. and the active pursuit of novelty. Phenomenology
and the Cognitive Sciences, 17(3), 521–534.
ORCID iD Clark, A., & Toribio, J. (1994). Doing without represent-
ing? Synthese, 101(3), 401–431.
Shaun Gallagher https://orcid.org/0000-0002-
Colombo, M., & Wright, C. (2018). First principles in
3147-9929
the life sciences: The free-energy principle, organi-
cism, and mechanism. Synthese, 198, 3463–3488.
Notes Constant, A., Clark, A., & Friston, K. J. (2021).
1. To be sure, even for Hohwy, the Markov blanket is Representation wars: Enacting an armistice through
not always a strict boundary, and he seems to grant active inference. Frontiers in Psychology, 11,
to the enactive analysis the possibility of organism- 598733.
environment coupling, although he continues to Dewey, J. (1922). Human nature and conduct. Henry
insist on a type of boundary-based decoupling. See Holt & Co.
Hohwy (2016, p. 277). Di Paolo, E., Thompson, E., & Beer, R. (2022). Laying
2. Mann and Pain (2022) note that the FEP is some- down a forking path: Tensions between enaction
what obscure in its meaning. The term “energy” in and the free energy principle. Philosophy and the
FEP “is not energy in the standard sense. Rather, Mind Sciences, 3, 3. https://doi.org/10.33735/
it is an information-theoretic term. . ..” They quote phimisci.2022.9187
Colombo and Wright (2018) who suggest: “FEP’s Friston, K. (2013). Life as we know it. Journal of The
epistemic status remains opaque, along with its Royal Society Interface, 10(86), 20130475. https://
exact role in biological and neuroscientific theo- doi.org/10.1098/rsif.2013.0475
rizing. Conspiring against its accessibility are the Friston, K., Adams, R. A., Perrinet, L., & Breakspear,
varying formalisms and formulations of FEP, the M. (2012). Perceptions as hypotheses: Saccades
changing scope of application, reliance on unde- as experiments. Frontiers in Psychology, 3, 151.
fined terms and stipulative definitions, and the lack https://doi.org/10.3389/fpsyg.2012.00151
10 Possibility Studies & Society 00(0)

Froese, T., & Ikegami, T. (2013). The brain is not an Psychology, 8, 1903. https://doi.org/10.3389/
isolated “black box,” nor is its goal to become one. fpsyg.2017.01903
Behavioral and Brain Sciences, 36(3), 213–214. Parr, T., Da Costa, L., & Friston, K. (2020). Markov
Gallagher, S. (2021). Performance/Art. Mimesis Inter­ blankets, information geometry and stochastic ther-
national Edizioni. modynamics. Philosophical Transactions of the
Gallagher, S. (2017). Enactivist interventions: Rethinking Royal Society A, 378(2164), 20190159.
the mind. Oxford University Press. Ramstead, M. J. D., Kirchhoff, M. D., Constant, A., & Friston,
Gallagher, S., & Allen, M. (2018). Active inference, K. J. (2021). Multiscale integration: Beyond internalism
enactivism and the hermeneutics of social cogni- and externalism. Synthese, 198(Suppl 1), 41–70.
tion. Synthese, 195(6), 2627–2648. https://doi. Rietveld, E., & Kiverstein, J. (2014). A rich landscape
org/10.1007/s11229-016-1269-8 of affordances. Ecological Psychology, 26(4), 325–
Gallagher, S., Hutto, D. D., Slaby, J., & Cole, J. (2013). 352.
The brain as part of an enactive system. Behavioral Ryan, K. (2020). Embodied cognition and jazz improvi-
and Brain Sciences, 36(4), 421–422. sation (Doctoral dissertation, The University of
Helmholtz, H. (1867). Treatise on physiological optics Memphis).
(Vol. III, 3rd ed.). Dover. Ryan, K. J., Jr., & Gallagher, S. (2020). Between eco-
Hohwy, J. (2013). The predictive mind. Oxford University logical psychology and enactivism: Is there reso-
Press. nance? Frontiers in Psychology, 11, 1147. https://
Hohwy, J. (2016). The self-evidencing brain. Noûs, doi.org/10.3389/fpsyg.2020.01147
50(2), 259–285. Schaefer, R. S. (2015). Mental representations in musi-
Høffding, S. (2019). A phenomenology of musical cal processing and their role in action-perception
absorption. Palgrave-Macmillan. loops. Empirical Musicology Review, 9(3/4), 161.
Kimmel, M., Hristova, D., & Kussmaul, K. (2018). Schiavio, A., Ryan, K., Moran, N., van der Schyff, D.,
Sources of embodied creativity: Interactivity and & Gallagher, S. (2022). By myself but not alone:
ideation in contact improvisation. Behavioral Agency, creativity and extended musical historic-
Sciences, 8(6), 52. ity. Journal of the Royal Musical Society. https://
Kiverstein, J. (2020). Free energy and the self: An doi.org/10.1017/rma.2022.22
ecological–enactive interpretation. Topoi, 39(3), Schwartenbeck, P., Fitzgerald, T., Dolan, R. J., & Friston,
559–574. K. (2013). Exploration, novelty, surprise, and free
Kronsted, C. (2021). An enactivist model of improvisa- energy minimization. Frontiers in Psychology, 4,
tional dance (Doctoral dissertation, The University 710. https://doi.org/10.3389/fpsyg.2013.00710
of Memphis). Van der Schyff, D., & Krueger, J. (2019). Musical
Kronsted, C., & Gallagher, S. (2021). Dances and empathy, from simulation to 4E interaction. In A.
affordances: The relationship between dance training F. Corrêa (Ed.), Music, sound & mind. ABCM.
and conceptual problem-solving. Journal of Aesthetic https://www.researchgate.net/profile/Dylan-Van-
Education, 55(1), 35–55. Der-Schyff/publication/326998297_Musical_
Mann, S. F., & Pain, R. (2022). Teleosemantics and the Empathy_From_Simulation_to_4E_Interaction/
free energy principle. Biology & Philosophy, 37, links/5b71f82992851ca65057e17c/Musical-
34. https://doi.org/10.1007/s10539-022-09868-9 Empathy-From-Simulation-to-4E-Interaction.pdf
McDowall, L. (2018). Exploring uncertainties of lan- van der Schyff, D., Schiavio, A., Walton, A., Velardo,
guage in dance improvisation. In V. Midgelow V., & Chemero, A. (2018). Musical creativity and
(Ed.), The Oxford handbook of improvisation in the embodied mind: Exploring the possibilities of
dance. Oxford University Press, pp. 185–205. 4E cognition and dynamical systems theory. Music
Merleau-Ponty, M. (2012). Phenomenology of percep- & Science, 1, 2059204318792319.
tion. Routledge. Walton, A. E., Washburn, A., Langland-Hassan, P.,
Michael, J., & Wolf, T. (2015). Why apply a hierarchical Chemero, A., Kloos, H., & Richardson, M. J. (2018).
predictive processing framework to music perception Creating time: Social collaboration in music improv-
and performance? Empirical Musicology Review, isation. Topics in Cognitive Science, 10(1), 95–119.
9(3/4), 177. Wiese, W., & Friston, K. J. (2021). Examining the con-
Orth, D., Kamp, J. van der, Memmert, D., & Savelsbergh, tinuity between life and mind: Is there a continuity
G. J. P. (2017). Creative motor actions as emerg- between autopoietic intentionality and representa-
ing from movement variability. Frontiers in tionality? Philosophies, 6(1), 18–18.

You might also like