You are on page 1of 23

2

The Elusive Extended Mind


Extended Information Processing Doesn’t Equal Extended Mind
Fred Adams

1. Introduction
Famously Clark and Chalmers (1998) argue that the mind extends.1 More spe-
cifically, they argue that cognition extends. So memory, belief, desire, reasons,
judgments, calculations, and other cognitive states are of the type that may extend
beyond the boundaries of body and brain, on this view.
Examples of extended cognition would include things such as working long divi-
sion with paper and pencil, rotating Scrabble letters in a tray to see if one can make
a word, rotating a jigsaw puzzle piece to see where it fits, and their most well-known
case, constantly carrying a notebook to store information that plays the functional
role of memory.2
In each case, the extension of cognition is based upon a constant causal cou-
pling of the agent with the part of the environment into which the cognition is said
to extend. Hence, in the notebook case, uncoupled there is no cognition in Otto’s

1
Elsewhere, Clark (2009) denies that consciousness extends. Indeed Aizawa and I were present in
Edinburgh when Clark revealed his reasons for denying that consciousness extends. Our reaction was
that, with minor tweaks and replacements of particulars about cognition, Clark’s reasons mirrored our
own reasons for why cognition does not extend.
2
Some of these examples come from others who accept extended cognition, but I think Clark and
Chalmers (1998) would accept them all. Clark (2008) seems to by the time of his Supersizing the Mind.
Their most famous case is of “Otto and his notebook.” They post the following conditions on the con-
nection between Otto and notebook: the notebook is constantly available and directly available; upon
retrieving information Otto automatically endorses it; and the information in the notebook was con-
sciously endorsed in the past and is there because of the endorsement. These conditions have come to
be known as the “trust and glue” conditions. The last condition in subsequent articles seems sometimes
to be dropped. I cannot find where Clark and Chalmers (1998) specify whether these are necessary or
sufficient conditions (or both).

21
22 Extensions and Alterations

notebook. If Otto’s cognition extends into the notebook, it does so only because it
is constantly causally coupled with Otto and with “trust and glue” playing the same
informational input-output role of saving, storing, and retrieving information as is
played by the memory storage areas of the brain in normal cases. In their example,
Otto is losing this normal memory capacity and is using the notebook to take over
some of this cognitive memory function.3

2. Previous Responses to the Claim


Elsewhere I’ve agreed that, in principle, it is possible that cognition extends but
that none of the examples of actual cases given to date suffice as instances of ex-
tended cognition (Adams and Aizawa 2001, 2008a, 2008b, 2009, 2010a, 2010b;
Aizawa and Adams 2005). Over the years, via many interactions in person and in
print, I have come to appreciate the tenacity with which Clark maintains this view.
Our interactions have always been friendly and congenial, despite a profound dif-
ference of views. Neither of us seems to have budged from our initial positions on
this matter, despite significant attention to the detail of argument presented on each
side. I think in some cases there are important misunderstandings that have arisen,
and I will begin there.

3. Misunderstandings
Over the years, in exchanges between Clark and Chalmers and Adams and Aizawa,
several disagreements have aired.4 However, I think there are also some significant
misunderstandings that I would like now to address. Clark (2010, 82) too thinks
that in our exchanges there have been “mutual failures of communication: crossed
wires concealing a couple of real, but rather more subterranean, disagreements.” But
unfortunately, I think Clark’s attempt to address these may have made matters worse.
At one point Clark (2010, 83) says, “This talk of an object’s being or failing to be
‘cognitive’ seems to me almost unintelligible, when applied to some putative part

3
They even say the notebook contains some of Otto’s dispositional beliefs.
4
Andy Clark sent me a draft of “The Extended Mind” paper prior to its publication. From the be-
ginning, Ken Aizawa and I have been skeptical of actual cases of extended mind, though we agree to the
physical possibility (Adams and Aizawa 2001, 2008a, 2008b). In addition to what we’ve said in print,
I remember an email exchange I had with Clark and Chalmers while I was at a conference in Italy in
2005. I suggested that a significant difference between Inga (who finds her way to MOMA the normal
way) and Otto (who only gets to MOMA with his notebook) is that Otto has to use perception to ac-
cess the information in his notebook. In their 1998 paper, Clark and Chalmers seemed to dismiss this
worry. But I noticed that Chalmers in his foreword to Clark (2008) now seems to think it is a more
relevant difference between Inga and Otto.
The Elus ive E x tended Mind 23

of a cognitive agent or of a cognitive system.” This is in response to our denial that


cognitive processing extends into Otto’s notebook.
What is Clark’s plaint? Is it that there is no, or we do not need a, mark of the cog-
nitive? Sometimes I think that is what he is saying. I will address below why, if that is
what he is saying, he is giving up the thesis that cognition extends. Alternatively, is
he saying that calling a thing cognitive is odd? I agree that calling a notebook “cog-
nitive” sounds odd. I think it is false that cognition extends to Otto’s notebook, but
still it sounds odd to call a notebook cognitive. But remember, the question is not
whether the notebook (by itself) is cognitive. No, the question is whether, when
coupled to Otto, cognition extends into the notebook. That had better not be odd
or “unintelligible,” for that is Clark and Chalmers’s (1998) thesis. If that is unintelli-
gible, then so is the “extended mind” thesis.
Clark is correct to point out that Adams and Aizawa make the following claim.
From the fact that Otto is causally (and perhaps informationally) coupled to his
notebook, it does not follow that Otto’s cognitive processing extends into the
notebook. At least, it doesn’t follow without explaining clearly what a cogni-
tive process is and how it is extending into Otto’s notebook (which Clark and
Chalmers never explain). Hence Adams and Aizawa (2008a) labeled the move
“the coupling-constitution fallacy” and demonstrated ample textual evidence of
its commission.5
Now Clark is also clearly suggesting that it is “unintelligible” to apply “cognitive”
to some part of a cognitive system. This is beyond belief. Surely the human body is
a cognitive system, but not every part of the human body is engaged in cognitive
processing, in cognizing, in cognition. If Clark and Chalmers claim otherwise, this
would surely be a reductio of their view. And if it does not make sense to talk about
where cognition occurs, then how could it be intelligible to propose that “cognition
extends”? But is that not precisely the thesis of their “Extended Mind?” Surely, it is.
Clark (2010, 84) maintains that he and Chalmers do not commit the coupling-
constitution fallacy because they include conditions for incorporating the coupled
entity into the cognitive system. But, of course, that won’t do. Otto’s circulatory
system is coupled to his brain and respiratory systems. Hence Otto as a whole is a
large cognitive system. But, though clearly part of Otto, neither circulation nor res-
piration is cognition. Neither of them is a topic for a chapter on cognition. No, the
reason Clark and Chalmers commit the coupling-constitution fallacy is that they
never tell us what counts as cognition, as cognitive processing. Clearly not every-
thing in the brain itself is engaged in cognitive processing. And just as clearly not
everything meeting Clark and Chalmers’s conditions for “coupling” to a cognitive
system satisfies the conditions constituting cognitive processing.

5
I think the central mistake Clark and Chalmers make is to think that any information-driven be-
havioral exchange helping a coupled system solve a cognitive task via “trust and glue” constitutes cog-
nition. The problem is that it does not (Adams 2010, 2014, 2016).
24 Extensions and Alterations

I now turn to a mysterious misunderstanding. Clark (2010, 84)  suggests that


according to Adams and Aizawa, “some objects or processes, in virtue of their own
nature are . . . candidate parts for (inclusion in a cognitive process).” Why does Clark
say this? Because at one point Adams and Aizawa consider what makes a process a
cognitive process and answer “the nature of X, of course?” But Clark seems to mis-
take our “X” for a biological essence or a physical kind of essence, where Adams and
Aizawa referred to a cognitive kind of essence. That is, we maintain that there is a
kind of processing that counts as cognitive processing. What kind is that? The cog-
nitive kind—whatever that turns out to be. We gesture at some properties that may
well be included (Adams and Aizawa 2008a). And since then I have tried to say a bit
more about the nature of X (cognition) (Adams and Garrison 2013, 2014). But in
no case have either of us ever maintained that cognition must be biologically based
or like human cognition. Since we have been very clear about this, it is even more
surprising that Clark misunderstands.6

4. Mark of the Cognitive


Mark Rowlands once opened an APA session at the Central Division saying that
there are two things wrong with the label “extended cognition.” First, what counts as
extension? And second, what counts as cognition?7 For the thesis of extended cog-
nition to be meaningful and taken seriously in science, both of these concepts need
clarification. I’m mainly interested in the latter. If cognition genuinely extends, then
there is something for cognitive science to investigate: cognition. It must be a kind of
thing, perhaps even a natural kind. In any case, it must have features that are able to
be investigated by a science. Unfortunately, Clark and Chalmers never offer features
of processes of the mind in virtue of which they are cognitive processes. Indeed
Clark seems to go out of his way not to do so. This may be because he thinks it is not
necessary or because he thinks there is no such mark of the cognitive. Actually, nei-
ther option is good for his thesis that the mind extends, as I’ll now explain.
One troubling feature of their 1998 paper is that they never say what makes
something a cognitive process. Instead they appeal to what has come to be known
as the “parity principle.” This principle says that if a process done outside the head
would be counted as a cognitive process were it done inside the head, then it is a

6
It is true that Adams and Aizawa think cognitive processing involves nonderived content. This is
what Searle has called “intrinsic content.” In Searle’s case, this may require particular kinds of causal
properties mainly had by biochemical kinds. But we were very careful not to identify with Searle’s
point of view and even changed the name to “nonderived” content and gave ample examples of de-
rived versus nonderived content to avoid just the kind of mistake Clark seems to be making—although
I think he must know better. See Clark 2010, 87!
7
This is likely where our agreement ends (Adams and Garrison 2013, 2014).
The Elus ive E x tended Mind 25

cognitive process even when done outside the head. Location doesn’t matter. A pro-
cess does not become noncognitive based solely on where it takes place. I agree! If a
cognitive process now clearly brain-bound were somehow able to bridge the space
between body, brain, and world, then that process would not cease to be cognitive.
But what is frustrating about their 1998 appeal to the parity principle is that they
never say what makes a brain-bound process cognitive. Surely, it is not merely that
it is in the head.
Susan Hurley (2010) once said that the brain is no “magical membrane,”
meaning that taking place within the brain is not what makes something cognitive,
nor is this location required for being cognitive. Again, I agree! Unlike real estate,
location does not matter. What makes something a cognitive process is not its lo-
cation. So while I agree, the flip side of my agreement is that Clark and Chalmers
should have at least addressed the matter of what makes something cognitive when
in the head if they were going to appeal to the parity principle to say that the same
process could be cognitive if it spans body, brain, and world. Then the parity prin-
ciple would have some bite. Because then we would know what makes something a
cognitive process and would be disinclined to think that location matters. However,
Clark and Chalmers (1998) never even address the issue of what cognition is. If
one doesn’t know what it is for a brain-bound process to be a cognitive process,
it doesn’t much help to be told that one not brain-bound could also be a cogni-
tive process if it is of a type that would be cognitive if brain-bound. While true,
the parity principle itself is completely unhelpful in determining what counts as
cognition.
We may be able to speculate on why they did not address the matter. First, it is
hard. It is hard to know what to say makes a process a cognitive process. Second,
there is no clearly agreed answer. More than once I’ve heard neuroscientists and
philosophers say that what makes something cognitive is that it is in the brain. But
barring that, cognitive scientists are likely to simply give one a list of processes that
count as cognitive processes—but without telling us what it is in virtue of which
they are on the list. And third, Clark and Chalmers may be suspicious of there even
being a mark of the cognitive. The first two I understand. It is hard. (I’ve tried.) And
there is no consensus on what the mark is. (I’ve searched.) But if they really believe
that there in fact is no mark, then I think that’s bad for them. Seriously bad.

5. Rejection of the Need of a Mark of the Cognitive


Hurley once told me that everyone in psychology knows that vision, audition,
memory, reasoning, and other such objects of study are cognitive processes.8 So

8
I think she reasserts this in Hurley 2010.
26 Extensions and Alterations

why do we need a mark of the cognitive? What is more, it is unlikely that there is any
sort of agreement among scientists about what makes something cognitive.
However, agreement on which processes we count as cognitive is not the same
thing as agreement upon what it is about them that makes them cognitive. As I say,
more than once I’ve asked really smart scientists and philosophers what makes
something cognitive only to be told that something is cognitive because “it occurs
in the brain.” This tells me that agreement on the list and agreement on the mark are
not the same thing.
Clark (2010) too at one point seems to want to reject the need for a mark.
He says:

The alternative, to paraphrase Dennett, is that cognition is as cognition


does. That is to say, we should individuate the cognitive by its characteristic
effects, not by its characteristic causes. The notion of a cognitive process,
if that is correct, is best unpacked as the notion of a process that supports
certain kinds of behavior (actual and counterfactual). This is the notion
that allows Otto's notebook, in virtue of its gross functional poise to count
as the gross supervenience base for Otto's cognizing, and, more specifi-
cally, for some of his dispositional believings. Why ask for more? (93)

In an important footnote (#4) Clark denies that consciousness can be characterized


solely in terms of its effects. I  think the same is true of cognition. The mistake
Clark is making here, in my view, is treating cognition as would a behaviorist, not
a cognitivist. One cannot help but notice that what Clark says about cognition is
quite empty and somewhat circular because we are not told what counts as Otto’s
“cognizings.” Cognition is what cognition does. But what is it? What does it do?
Which effects are the cognitive ones? The ones produced by cognitive processes, of
course. So this quote very much represents both Clark’s rejection of a mark of the
cognitive and rejection of the need of one.
Clark and Dennett both know well the problems that led to the demise of be-
haviorism. So it is all the more surprising to see a remark such as this by them about
cognition. One of my staple examples in classrooms is to lay my finger beside my
nose and ask the class “What did I do?” I then give them the following choices:

1. Scratched my nose.
2. Did this just to make an example in the class.
3. Signaled a mafia hit.

Which it was, of course, depends upon the reason why I did it. To say cognition is
as cognition does is right if understood in this way. But to go on and say it is only
the effects, not the causes (i.e., reasons), that are relevant could not be further from
the truth.
The Elus ive E x tended Mind 27

In another quote Clark (2010, 96—97) says, “Just so, there is surely no value in
pursuing the question, asked of Otto’s notebook, ‘is it cognitive?’ Instead we must
attend (and Chalmers and I did attend in the published essays) to the role of the
notebook in the larger organization of which biological Otto is a part.”
But if there is no mark of the cognitive, then when told cognition extends, what
have we learned? It is not clear what claim has been made. “Cognition” must mean
something. There must be conditions that fix the meaning/reference of the term.
And if there is no mark of the cognitive, then the truth conditions for the claim
that cognition extends are indeterminate. And if indeterminate, I  fail to see why
anyone in cognitive science should be interested. For then there would be no laws
or explanations that hold in virtue of possessing a scientifically investigable kind. So
there must be something about cognition in virtue of which it is scientifically inter-
esting, if it is in virtue of it that certain types of explanations hold. This won’t be the
case if the meaning of “cognition” is scientifically indeterminate.

6. Supervenience Is Too Weak


So what is the bottom line in our disagreement over extension of cognition? It
seems to me that for Clark (2008), a cognitive agent incorporates information from
the environment via a perception-action loop. Then that information is poised for use
in solving cognitive tasks, and that information use amounts to cognitive extension.
He sometimes says that the supervenience base for solving cognitive tasks is ex-
tended beyond boundaries of body and brain.
There may be conditions of “trust and glue” or others that are added for the causal
coupling of the cognitive agent with the portion of the environment supplying the
information that is to be used, but I’m not sure this adds to the notion of “looping”
that Clark sometimes references. Still, the main ingredient that Clark appeals to
is the poise and use of the information provided by the coupled system, via the
perception-action information loop into the environment.
Okay. So what’s wrong with that? Isn’t that what it is for cognition (mind) to
extend? No.9 Why not? It is true that cognitive processing exploits information
extracted from the environment and poised for use in solving cognitive tasks. But
cognition involves more than that. It involves mental structures that rise from the
level of information only to the level of meaning (Adams and Aizawa 1994).
Let’s start simply. Suppose I tie a string around my finger to remember to stop
and get milk on the way home. This has plenty of “trust and glue.” (I take my finger
everywhere and trust it implicitly.) This has information poised to help me solve the
cognitive task of getting milk on my way home. This broadens the supervenience

9
Remember that Clark does not think this is sufficient for consciousness to extend! I don’t think it
is sufficient for cognition to extend either.
28 Extensions and Alterations

base of my solving the task. So does this show that my mind extends to my finger (to
the string on my finger)? Hardly!
Now Otto’s notebook is no different in principle. It just has more strings (more
information in it that Otto can use to solve cognitive tasks). But if we raise the fol-
lowing question about the finger and string—is it part of my mind, or is it just a cog-
nitive aid (prompt), reminding me about the milk—I maintain that it is the latter.
When it comes to Otto and his notebook, I think it is just more of the same. Does
the information Otto extracts from his notebook (via its poise and extraction) con-
stitute cognitive processing, or is it just informational and causal support for Otto’s
cognition? I maintain it is the latter. Clark maintains that it is the former.10
Why the disagreement? For one, because I maintain that Clark needs to tell us
when use of poised, stored information amounts to cognitive processing and when
it does not. He seems to want to deny that there is a difference. I want to insist on
there being a difference. How do I do that? For one, by suggesting the nonderived
content requirement (Adams and Aizawa 1994, 2001, 2008a, 2008b). For another,
by saying a bit more about what makes something a cognitive system (Adams and
Garrison 2013, 2014). And finally, by suggesting instances where Clark’s conditions
should be met but where one would deny extension—as in my finger and string ex-
ample. But let’s revisit Otto.
Suppose there are conjoined twins, Otto and Notto. (So “trust and glue” is satis-
fied in spades.) Notto now plays the exact role of Otto’s notebook. Otto’s memory
is failing. So instead of a notebook, Otto tells Notto everything that he otherwise
would put in the notebook. Then, with the information within Notto poised and
ready to aid Otto in getting to MOMA (the Museum of Modern Art), Otto shows
up on time, the same as Inga (who gets there by normal means). Now what? Does
Otto’s mind extend into Notto’s? How could it be otherwise for Clark? This seems
to me a clear reductio of the view. Notto supplies information that reminds Otto
of what he needs to do and how to get to MOMA. But Otto’s cognition does not
extend into Notto’s mind. The information that Notto shares with Otto is poised
to be used by Otto to remind him of what to do and where to go, but this informa-
tion from Notto is mere causal support (information input) for Otto’s cognitive use.
And while the processing of information within Notto is surely cognitive processing
of information, it is Notto’s cognitive processing of information, not Otto’s. Once
Notto shares the stored information with Otto and Otto begins using it, then and
only then does its use constitute Otto’s cognitive processing.

10
Several years ago I invited Susan Goldin-Meadow to come to my university and talk about her
work. While here I showed her some passages in Clark (2008) that talk about her work on gesture
and the role it plays in cognition. I asked her if she thought her work showed that gestures constituted
cognitive processes or were just causal support for cognition. She assured me that she thought they
were causal support alone. She also said she was not sure these two positions could be empirically
teased apart.
The Elus ive E x tended Mind 29

So I think the difference between Clark’s view and mine is that I think there is more
to cognition than acquisition and storage of information that may be poised to help
one solve a cognitive task. I maintain that this can happen below the level of cognitive
processing of information. Sensing is not thinking (not cognizing), but information re-
ceived by sensory detectors is surely poised and ready to be used by the mind in thinking
and in solving cognitive tasks. But sensing is not thinking, and it is not cognizing.
And mere supervenience is also too weak a relation to constitute cognition.
Supervenience is a dependence relation. So the mere fact that Otto’s ability to solve
cognitive tasks supervenes on his brain and his notebook (as coupled in the appro-
priate ways Clark requires) does not show that the cognition extends into the note-
book. It doesn’t show that the information processing that supervenes on the notebook
constitutes cognitive processing. As I say, supervenience is a dependency. But Otto’s
cognitive system supervenes upon his circulatory system and his respiratory system
too. Remove these and Otto will solve no cognitive tasks. Nonetheless cognition is nei-
ther respiration nor circulation. And cognition is not the motor process of putting in-
formation into his notebook nor the perceptual process of extracting information from
the notebook. Cognition is what Otto does with that information once extracted: the
process of mental manipulation of meaningful symbols with nonderived content (just
as Adams and Aizawa have always maintained).
A kind of nonepistemic sensing (or seeing) is not believing. (It is not thinking.)
And motor movements are not actions. They become actions only when done for
reasons. Having reasons for movements involves cognition, but not the movements
themselves. This is why I disagree with Clark and Dennett, who seem to say (in the
quote earlier) that cognition is not about the causes of behavior but about its effects.
Cognition is about the causes, the reasons for behavior.
So in the end, I maintain that cognitive processing is more than picking up or
storing information that is poised for future use. It is different because it involves
the informational inputs rising to the level of being meaningful symbols (with
nonderived content). Clark is happy to rest with the view that cognition happens
below the level of the construction of such meaningful symbols. But this would
open the floodgates to a whole host of systems:  plants, bacteria, iPhones. These
are information processors with informational loops that extend into the environ-
ment. They exploit that information to meet their needs (or needs of users). And
as a result, there have been many scientists and philosophers who have maintained
that plants and bacteria too are cognitive systems. They say this because they have
lowered the bar on what counts as cognitive processing. This is another reason why
I resist Clark’s lowering of the bar (Adams 2018a, 2018b). I think Clark holds that
information-driven behavioral exchange in a cognitive system (when appropriately
coupled to a cognitive agent) constitutes cognitive processing.11 I do not.

11
Of course, he never actually says that, but this is what I’m taking away from most of what he says
about “looping” and “coupling” in Clark 2008.
30 Extensions and Alterations

7. Conclusion
I have addressed what I think are the main reasons why Clark believes cognition
extends. I’ve also addressed some of the misunderstandings that have appeared in
print over the years. I readdressed the need for a mark of the cognitive and why
Clark’s resistance to offer one threatens to undermine his thesis of extended mind.
And finally, I’ve suggested that the likely reason why we disagree is that he has an
“information”-only (plus “coupling”) requirement on cognition and cognition ex-
tension, where I maintain that this gives only causal support for cognition, not con-
stitution of cognitive processing.12 Will one of us move the other? I sort of doubt it
after all these years, but I still enjoy the attempts.13

References
Adams, F. 2010. “Why We Still Need a Mark of the Mental/Cognitive.” Cognitive Systems Research
11:324–331.
Adams, F. 2014. “What Is a Cognitive Process?” Foundations of Science 19(2):133–135.
Adams, F. 2016. “Information and Cognition.” In The Routledge Handbook of Philosophy of
Information, edited by L. Floridi, 345–356. Routledge.
Adams, F. 2018a. “Cognition wars.” Studies in History and Philosophy of Science Part A, 68, 20–30.
Adams, F. 2018b. “Extended Knowledge?” In Extended Epistemology, edited by A. Clark, J. Kallestrup,
O. Palermos, D. Pritchard, and J. A. Carter, 79–89. Oxford University Press.
Adams, F., and K. Aizawa. 1994. “Fodorian Semantics.” In Mental Representations, edited by S. Stich
and T. Warfield, 223–242. Basil Blackwell.
Adams, F., and K. Aizawa. 2001. “The Bounds of Cognition.” Philosophical Psychology 14: 43–64.
Adams, F., and K. Aizawa. 2008a. The Bounds of Cognition. Blackwell/Wiley.
Adams, F., and K. Aizawa. 2008b. “Why the Mind Is Still in the Head:  Challenges to Active
Externalism.” In Cambridge Handbook of Situated Cognition, edited by P. Robbins and M.
Aydede, 78–95. Cambridge University Press.
Adams, F., and K. Aizawa. 2009. “Embodied Cognition and the Extended Mind.” In Routledge
Companion to the Philosophy of Psychology, edited by P. Calvo and J. Symons, 193–213.
Routledge.
Adams, F., and K. Aizawa. 2010a. “Defending the Bounds of Cognition.” In The Extended Mind, ed-
ited by R. Menary, 67–80. MIT Press.
Adams, F., and K. Aizawa. 2010b. “The Value of Cognitivism in Thinking about Extended Cognition.”
Phenomenology and the Cognitive Sciences 9(4): 579–603.
Adams, F., and R. Garrison. 2013. “The Mark of the Cognitive.” Minds & Machines 23:339–352.

12
I’d like to thank Charlotte Shreve and John A. Barker for very helpful conversations while I was
putting this together. Naturally I also must thank Ken Aizawa for our many years of collaboration on
this topic.
13
I’m very pleased to have been included in this book honoring the work of Andy Clark.
I  have enjoyed the good-natured aspect our disagreements. We have met in person many times
and corresponded about these matters, and that interaction has always been affable. Despite our
disagreements, I respect Clark very much and look forward to our continuing friendship. I would also
like to thank the editors for very helpful editorial advice.
The Elus ive E x tended Mind 31

Adams, F., and R. Garrison. 2014. “Mark of the Cognitive: Reply to Elpidorou.” Minds & Machines
24:213–216.
Aizawa, K., and F. Adams. 2005. “Defending Non-Derived Content.” Philosophical Psychology
18:661–669.
Clark, A. 2008. Supersizing the Mind. Oxford University Press.
Clark, A. 2009. “Spreading the Joy? Why the Machinery of Consciousness Is (Probably) Still in the
Head.” Mind 118:963–993.
Clark, A. 2010. “Coupling, Constitution, and the Cognitive Kind: A Reply to Adams and Aizawa.” In
The Extended Mind, edited by R. Menary, 81–100. MIT Press/Bradford.
Clark, A., and D. Chalmers. 1998. “The Extended Mind.” Analysis 58:7–19.
Hurley, S. 2010. “The Varieties of Externalism.” In The Extended Mind, edited by R. Menary, 101–
155. MIT Press/Bradford.
3

Clark on Language, Cognition,


and Extended Cognition
Kenneth Aizawa

There is a fairly well-known, roughly Fodorian/Chomskyan (F/C) theory of the re-


lationship between language and thought. According to this theory, understanding
an utterance in a natural language is a matter of pairing that utterance with a sen-
tence in a language of thought. This language of thought is often referred to as
“Mentalese.” Clark (2006a, 2006b) describes this as a translation model of language.1
In place of this view, Clark proposes a complementarian model of language. As a first
pass, the model understands “public language as a complementary resource that
works with the more basic machinery [of the brain] without installing any funda-
mentally new styles of representation or processing within that machinery” (Clark
2006b, 293).
In this chapter I have three principal goals. The first is to offer and invite clari-
fication regarding the two models of language and cognition, partly by reference
to a common distinction between cognition and behavior. I will not go so far as to
offer any theory, account, or definition of either cognition or behavior, but simply
presuppose that there is such a distinction. In addition, I will provide reasons to
think that the distinction is, at least at times, common ground between Clark and
Fodor. Importantly, I will invite some clarification about why we should take the
two models to be alternatives.
Second, while I take the complementarian model to be broadly correct, I will
briefly provide some reason to doubt one of Clark’s more specific proposals re-
garding the role of language in cognition. Clark proposes that public uses of natural
language are enablers of certain forms of metacognition. According to Clark (1998,
177), the use of natural language—what one might call “verbal behavior”—enables
“self-evaluation, self-criticism and finely honed remedial responses.” A challenge to

1
As we shall see below, this name is not entirely unproblematic.

32
Clark on Lang uage, Cog ni tion , and E x tended Cog niti on 33

this view, however, appears on the horizon in the form of languageless adults. There
are individuals who neither speak nor sign a natural language, but who nevertheless
appear to have at least some apparently normal metacognitive abilities. Whatever
enables their “second-order cognitive dynamics” is evidently not their use of nat-
ural language. And in the case to be reviewed, it appears that whatever enables their
“second-order cognitive dynamics” cannot even be a more minimal practice of
using symbols.
Third, I will review the case the complementarity model is supposed to make
for the hypothesis of extended cognition. Here I will remind the reader of some
of the more familiar types of arguments for extended cognition before turning to
Clark’s appeal to the complementarity view of language. Clark’s argument here has
the same weaknesses as other complementarity arguments, but here the conclusion
will be developed somewhat differently by drawing attention to the early distinction
between cognition and behavior.2

1. The Translation Model of Language


To begin to clarify the issues, we should try to settle on some common terminology.
We should avoid fruitless debates arising from differences in terminology. As a first
step—one that will prove to be crucial—we might recall the common distinction
between cognition and behavior/performance.3 This distinction has been part and
parcel of cognitive science since the cognitive revolution. Noam Chomsky and B. F.
Skinner, who disagreed about any number of issues regarding psychology, at least
agreed that there is a difference between cognition and behavior. Chomsky and
Skinner agreed that cognition was thought to be among the factors internal to an
organism that influences the organism’s behavior; where they disagreed was in their
estimates of the scientific value of postulating such endogenous factors. Chomsky
took them to be essential to correct scientific explanations, where Skinner thought
they were misguided.
Another component of the F/C approach involves embracing a distinction
between perception and cognition.4 The locus classicus for this view is perhaps
Fodor (1983). The distinction takes seriously the sense that there is a difference

2
See Adams and Aizawa (2008, 143–147) for an earlier discussion of complementarity arguments
for extended cognition.
3
In what follows, “behavior” and “performance” will be used interchangeably.
4
As Rowlands (2010), notes “cognition” can be used broadly or narrowly. In a broad usage, “cog-
nition” covers so-called central processing of thinking and reasoning as well as peripheral perceptual
processes, such as hearing, seeing, and parsing of natural language. In a narrow usage, “cognition” refers
only to the central processes, leaving aside the peripheral perceptual processes. Here “cognition” is
used narrowly.
34 Extensions and Alterations

between merely thinking about, say, a tree and perceiving a tree. It is a distinction
that Descartes embraced in terms of “imagination” and “reason.” It is a distinction
Kant embraced in terms of “intuition” and “understanding.” Simplifying a bit for
present purposes, Fodor and Chomsky hypothesize that there is a single central
system for cognition but additional peripheral components dedicated to multiple
sense modalities and natural language processing.5 According to the F/C model,
it is a special purpose language module that pairs utterances with thoughts, under-
stood as sentences in a language of thought.
To see how the model works in rough outline, consider Jack saying to Jill, “Please
pass the salt.” Assuming that Jill is a normal speaker of English, Jill understands the
utterance when her language module correctly pairs the utterance with the appro-
priate thought. Once Jill understands that Jack wishes for her to pass the salt, she
presumably scans the dinner table for the salt shaker, reaches for it, and hands it to
Jack. Jill’s behavior in this case might be the detection of the utterance and the sub-
sequent passing of the salt to Jack. Jill’s cognitive processing might be the interaction
of her thought that Jack wants her to pass the salt, the (perceived) location of the
salt, and her values with respect to Jack’s wishes that result in her decision to grant
Jack’s request. Jill’s behavior is a function, in part, of her cognitive and perceptual
processes, but also a function of the condition of her muscles. If she had recently
exercised her biceps to exhaustion or had been lying on her arms so that they had
fallen asleep, her behavior or performance in reaching for the salt may have been dif-
ferent. She might have reached for the salt shaker but missed it. Or she might have
reached for the salt shaker but have been unable to grasp it. So, on the F/C model,
cognitive, perceptual, and muscular processes are among the endogenous processes
that influence behavior.
Another feature of the F/C view is that one must distinguish the computational
mechanisms of natural language thought to be realized in the brain, on the one hand,
from speech and from texts that are to be found, to a first approximation, in the body
and environment, on the other.6 The computational mechanisms are, according
to the F/C theory, among the endogenous causes of the production of speech
behavior.7 These mechanisms employ a syntactically and semantically combinatorial
language. Other endogenous causes include, for example, the musculature of the

5
Cf. Fodor 1983, 47.
6
Clark (1998, 162), adopts this terminology but does not use it all the time.
7
Clark rejects much of the F/C conception of the computational mechanisms for language in favor
of a connectionist view. I am assuming, however, that, as a matter of logic, it is not part and parcel
of the complementarity view of language. I am assuming that one can take “public language” to be a
complementary tool without thereby rejecting the F/C conception of the internal apparatus for lan-
guage processing. The qualification “a matter of logic” is meant to acknowledge the fact that Clark
(2000a, 2000b) seems to embrace both the complementarity view (though not by that name) and the
connectionist view of the mind/brain.
Clark on Lang uage, Cog ni tion , and E x tended Cog niti on 35

face. Utterances are, perhaps, the patterns of sound in the environment.8 Written
language, or orthography, is yet another thing. There are, after all, human languages
that have no associated writing system. It is speech, utterances, and orthography
that are the “externals tools” that are supposed to scaffold cognition. It is these that
are supposed to be akin to sextants, compasses, maps, and slide rules.9

2. The Complementarity Model of Language


There are advocates of embodied, enactive, and extended cognition who appar-
ently wish to reject the common distinction between cognition and behavior and,
perhaps, the putative endogenous mechanisms of perception and cognition.10 At
times, Clark seems to be among them. So, at one point, Clark (2000a, 309) writes
the following: “For on my account, our practices of moral talk and exchange, and
our collective efforts to create the kinds of abstract, shared conceptions (of ‘charity,’
‘rights,’ ‘equality,’ ‘opportunity,’ etc.) that such discussions require are part of what
constitutes our practices as genuinely moral in the first place.” Depending on how
one reads this, it might well amount to a truism. Maybe there are some beings whose
moral practices are not constituted by discourse, but such beings would not have
our moral practices. This at least threatens to be a truism, since practices are a kind of
behavior. But at another point in the same paper, after referring to moral discourse,
Clark claims, “In the case at hand, such representations do not simply oil the wheels
of moral debate, they actively constitute the thinking as moral.” Then a bit later, he
adds, “The picture is one in which the moral realm comes into view, and moral cog-
nition is partially constituted, only by the joint action of neural resources we share
with other animals and the distinctively human infrastructure of linguaform moral
debate and reason” (311.) Now Clark seems to be making the much more contro-
versial claim that part of our moral cognition—our moral thinking—is constituted
by sentences in natural language discourse. This looks to be a special case of ex-
tended cognition, hence something that is decidedly more controversial.11
While the discussion in Clark (2000a, 2000b) seems to run indiscriminately
over the putative distinction between cognition and behavior, Clark (1998, 2006a,
2006b) seems to embrace the cognition/behavior distinction. (Clarification on

8
For present purposes, we may set aside signed natural languages.
9
Note that on the F/C view, both natural language and linguistic/cognitive processing in the brain
involve a syntactically and semantically combinatorial language of thought, whereas for Clark, it is only
natural language that has this structure. For Clark information processing in the brain is done in a non-
linguaform manner.
10
What advocates of embodied, enactive, and extended cognition think about the putative modu-
larity of mind is largely conjectural, since they rarely if ever touch on the topic. For further discussion
of the cognition/behavior distinction, see Shapiro 2013; Aizawa 2014, 2015, 2017.
11
See Adams and Aizawa (2001, 2008) and Rupert (2009) for much further discussion.
36 Extensions and Alterations

this point hereby invited.) Clark (1998, 163) is perhaps the least explicit of these
three papers on this topic, but we do find “the idea of language as a computational
transformer which allows pattern-completing brains to tackle otherwise intrac-
table classes of cognitive problems.” Insofar as “pattern-completing” is a stand-in
for cognition, and “tackling a problem” is a kind of behavior or performance, then
Clark (1998) appears to be committed to the cognition/behavior distinction. Clark
(2006a, 370) proposes that certain work in embodied cognitive science “highlights
the transformative effects of bodily form, bodily activity and material environ-
mental scaffolding on mind and cognition.” One might think that “bodily activity”
is a phrase for describing behavior and that bodily activity/behavior transforms
(i.e., [causally] effects some changes in) a distinct category of thing, namely, mental
and cognitive processes. Read this way, what Clark is suggesting seems to be correct
and consistent with the F/C model. Behavior causally influences perception and
cognition. Clark (2006a) describes a simulation from Clowes and Morse (2005)
that includes a simple recurrent neural network (SRNN). The model is supposed
to simulate a task in which the network uses, or does not use, “commands” to guide
its behavior in moving objects left, right, up, or down. Some variants of the model
use the “command” only once, but others “reactivate” the command internally.12
Clowes and Morse describe their models in terms that seem to respect the familiar
distinction between cognition and behavior: “We examine experiments designed to
explore the way language directed toward self, which has previously been borrowed
from an inter-agent communication, can come to play a role in intra-agent cogni-
tion. Our purpose is to try and elucidate the way that language becomes involved in
behavioral tasks and how it can sculpt cognitive development” (101). Clark (2006b,
302) also describes the models in ways that are friendly to the cognition/behavior
distinction:  “Clowes and Morse found that under the control condition  .  .  .  the
agents take longer to learn to succeed at any of the tasks, and seem unable to learn
to succeed at all four. This is because improvements in one task seemed to always
result in impairment to performance on one or more of the others.” These texts are
not maximally explicit when it comes to the putative cognition/behavior distinc-
tion, but they at least suggest that Clark might embrace it. But does he? Some clarity
here would be helpful.
Notice that by accepting that there is a difference between cognition and behav-
ior Clark need not go so far as to give a definition, account, or theory of cognition
or of behavior. Nor does he need to provide a “mark of the cognitive.” He can, there-
fore, in principle, offer some clarification, while avoiding becoming embroiled in at

12
It is unclear that anything particularly representational or linguistic is behind the performance
of the network. The “commands” are just activated nodes. There is no obvious need for these activated
nodes to correspond to representations, much less anything like complicated grammatical structures
found in natural language.
Clark on Lang uage, Cog ni tion , and E x tended Cog niti on 37

least some of the methodological handwringing that has beset the concerns about
the mark of the cognitive.13
Accepting a cognition/behavior distinction seems to be consonant with some of
Clark’s discussion of the complementarity model, but it would also seem to be in-
dependently philosophically important. Suppose we follow those who identify cog-
nition and behavior. There are few philosophers or cognitive scientists who would
doubt that cognition, understood as behavior, is extended. Clearly, behavioral
processes are extended into the body and one’s artifacts. The process of hammering,
for example, extends into the arm, the hammer, and perhaps the nail. Maybe you
don’t need a nail to hammer, but how could one hammer without a hammer and a
hand and arm to wield it? So, if one wants to advance an interesting version of the
hypothesis that cognition is extended, one should probably not mean that cogni-
tion is behavior and that behavior is extended.
So, at this point, it appears that, at least at times, Clark is amenable to there being
a distinction between cognition and behavior and that there is some prima facie
reason for him to avoid the identification of cognition and behavior. Nevertheless
were Clark to begin down this path, there seem to be risks for some of his other
concerns. So suppose that cognition is not the same thing as behavior. From time
to time, Clark refers to a human “pattern-completing brain” (cf. Clark 1998, 163;
2006b, 302). But this invites the following line of thought. Perhaps cognition is pat-
tern completion. But, then, it looks like Clark’s view is that cognition is brain-bound
insofar as pattern-completing is brain-bound. So cognition is not extended. (And
Clark appears never to have claimed that pattern completing extends beyond the
brain.) But if cognition is not pattern matching, why is it not? Here some clarifica-
tion would be helpful.14
As a second point of clarification, it would be helpful to have some fuller ac-
counting of why Clark takes the complementarity model to be an alternative to the
translational model. Notice that Clark (2006b, 292) suggests, “The alternative to
the translation picture, that I wish to pursue here, makes the role of public language
more like that of the spade.” Later he adds in more detail:

On a pure translation view, it is hard to see how our linguistic encounters


can do anything more than inculcate a kind of useful shorthand for ideas
whose very thinkability requires only the more fundamental tokenings
(in mentalese or neuralese) with which they have come to be associated.
The alternative on offer is a “hybrid model” according to which some of
the cognitive benefits that language brings depend on the complementary

13
Cf. Aizawa 2015, 2017.
14
Notice that a parallel argument might be based on the supposition that predictive processing is
brain-bound. Cf. Clark 2015.
38 Extensions and Alterations

action of actual material symbols (and image-like inner encodings of such


symbols) and more biologically basic modes of internal representation.
(304)15

On its face, however, it is unclear in what sense that complementarity model is


an alternative to the translational model.16 In principle, both could be true. One
might think that the reason utterances and texts can be useful in the performance of
tasks is that they typically trigger thoughts in the brain that are relevant to a task at
hand and that thoughts that are relevant to the task at hand influence performance.
Indeed, thinking about a task often, though not necessarily universally, improves
performance on the task. So it would be helpful for Clark to articulate in a bit more
detail what he means by the complementarity model being an alternative to the
translational model.

3. The Complementarity View and Metacognition


Turn now to the complementarian view of language, according to which speech and
writing are like other tools. Clark (1998, 162) proposes, “Public language . . . is a
species of external artifact whose current adaptive value is partially constituted by
its role in re-shaping the kinds of computational space that our biological brains
must negotiate in order to solve certain types of problems, or to carry out certain
complex projects.” Clark (2006a, 271) similarly claims speech and texts “transform
the tasks that confront an intelligent agent.” This seems right. Speech and texts are
like other tools in enhancing our performance. We can perform tasks with speech
and texts that we cannot perform without them. We could probably not teach much
of philosophy without speech and/or texts. Just to be clear, the foregoing reads
Clark’s proposal as saying that speech, writing, and other tools change behavior/
performance, not that they change cognition. A principle of charity seems to sup-
port this reading, but so do Clark’s comments that are meant to distance his pro-
posal from another by Daniel Dennett.17 Clark (1998, 6) tells us that his view “does
not depict experience with language as a source of profound inner re-programming
(pace Dennett).”
Clark (1998) conjectures more than that speech and texts are tools. He also
offers the more specific proposal that they are tools that enable us to think about
our thinking:  “Perhaps it is public language which is responsible for a complex
of rather distinctive features of human thought viz., our ability to display second
order cognitive dynamics. By second order cognitive dynamics I  mean a cluster of

15
Consult, as well, the abstract to Clark 2006b.
16
One possibility is broached in footnote 7 above.
17
See Dennett 1991. See also Clark’s (1998) discussion of Vygotsky’s work.
Clark on Lang uage, Cog ni tion , and E x tended Cog niti on 39

powerful capacities involving self-evaluation, self-criticism and finely honed reme-


dial responses” (Clark, 1998, 11).
One source of evidence that might bear on this hypothesis is research on
languageless adults. These are individuals who are born deaf, so they do not acquire
spoken language. Moreover they often come from poor communities where they
are not exposed to any form of sign language, hence do not acquire a sign language.
Susan Schaller (2012) provides a case study of one such individual, Ildefonso. Of
some of her early interactions with Ildefonso she writes, “I tried once more to ex-
plain without language that language existed, to explain without names that eve-
rything had a name. I failed, and his face showed that he knew he had let me down.
We were only inches apart, but we might as well have been from different planets;
it seemed impossible to meet” (Schaller 2012, 26, italics added). Schaller believed
that, without language or even symbols (see below), Ildefonso understood that he
was failing his teacher. He thought his thinking was failing to figure out what was
going on. Consider, next, the dramatic moment when Ildefonso gets it, that there
are symbols for things:

He broke through. He understood. He had forded the same river Helen


Keller did at the water pump when she suddenly connected the water
rushing over her hand with the word spelled into it. Yes, w-a-t-e-r and c-
a-t mean something. And the cat-meaning in one head can join the cat-
meaning in another’s head just by tossing out a cat.
Ildefonso’s face opened in excitement as he slowly pondered this revela-
tion. His head turned to his left and very gradually back to his right. Slowly
at first, then hungrily, he took in everything as though he had never seen
anything before: the door, the bulletin board, the chairs, tables, students,
the clock, the green blackboard, and me.
He slapped both hands flat on the table and looked up at me, demanding
a response. “Table,” I signed. He slapped his book. “Book,” I replied. My
face was wet with tears, but I obediently followed his pointing fingers and
hands, signing: “door,” “clock,” “chair.” But as suddenly as he had asked for
names, he turned pale, collapsed, and wept. Folding his arms like a cradle
on the table, he lay down his head. My fingers were white as I clutched the
metal rim of the table, which squeaked under his grief more loudly than his
sobbing. (44–45)

For the longest time, Ildefonso understood that he was not getting what he was
supposed to be getting, but then, in an instant, he realized that the sign Schaller
had been producing meant “cat.” Moreover, at that instant, he knew that he knew.
Evidently it was not that Ildefonso had command over sentences that expressed
thoughts that enabled him to evaluate his past and current thoughts. Instead, with
the mastery of one sign, he had figured out that there are signs. At that point he had
40 Extensions and Alterations

yet to grasp the more complicated signs that are sentences. So there are reasons to
doubt Clark’s hypothesis that sentences of natural language produce second-order
cognitive dynamics. Instead these observations appear to be consistent with the F/
C idea that there is a domain-specific language-acquisition device that handles only
language. Ildefonso’s language module did not develop properly, but his metacogni-
tive science continued to develop more or less normally.
Exactly what happened with Ildefonso is subject to interpretation, as are any
data, especially behavioral data that are meant to reveal underlying cognitive (or
metacognitive) capacities. So, for example, in comments on an earlier draft of this
paper, Matteo Colombo proposed, “A more minimal reading of that would be that
he was just showing awareness of having disappointed someone (easy enough to do
without metacognitively representing this) and showing understanding that things
have names and being excited by this—also easy enough to do without representing
this as a metacognitive state (e.g. ‘I know that I know’). That is, things might look
metacognitive without actually involving metacognitive representations.” This
reading would, indeed, attribute less metacognitive ability to Ildefonso, but is it as
plausible an interpretation? On Colombo’s reading, Ildefonso is aware that he is
disappointing Schaller, but he would have no awareness of why he is disappointing
Schaller. It would be mysterious to Ildefonso why Schaller was reacting as she was.
Why would he even think she is disappointed? On Schaller’s interpretation (favored
above), Ildefonso is aware that he is disappointing, because he is not getting what
she is trying to teach him. Moreover his getting the connection between words
and things explains why he is so happy. He might be glad to know that things have
names. That’s something good to know, like knowing where the bathrooms are, or
such other facts. But why does Ildefonso have such a deeply emotional response on
learning that things have names? Perhaps because he now sees what Schaller had
been struggling to show him. He realizes that he did not understand what she was
trying to teach him, but now he sees that he does.18

4. The Complementarity Model


and Extended Cognition
Finally, turn to what might be the most momentous proposed consequence of the
complementarity model, the support it is supposed to give to the hypothesis of ex-
tended cognition. Clarity is served by noting that there are multiple “families” of
arguments for extended cognition. So, for example, there are “parity” or “cognitive
equivalence” arguments that attempt to show that processes that span the brain, body,

18
A review of the literature on the metacognitive abilities of nonhuman animals (who I take not to
possess natural language) and languageless adults might reveal further relevant evidence.
Clark on Lang uage, Cog ni tion , and E x tended Cog niti on 41

and world are in some relevant sense on a par with cognitive processes in the brain.
These arguments are often based on the well-worn “Inga-Otto” thought experiment or
on three modes of playing the video game Tetris. What is characteristic of the parity
arguments is the treatment of two or more cases as importantly alike, whereas the mul-
tiple versions of the argument vary in what they take to be the important similarities.
Perhaps the cases are treated alike by folk psychology, by scientific psychology, or by
some “generalized” psychological taxonomy.19 A second sort of argument relies on
some sort of “coupling” between cognitive processes in the brain, on the one hand,
and bodily and environmental processes, on the other. The coupling could be a matter
of reciprocal causal connections between cognitive processes in the brain, on one side,
and bodily and environmental processes, on the other. Alternatively, it could be some
variant on the idea of conditions of “trust and glue.”
Clark (1998, 181) offers a version of a causal coupling argument:

Certain external (to the biological unit) props and aids may play such a
deep role in determining the shape and flow of our thoughts as to invite
depiction as part and parcel of the very mechanism of human reason. This
depiction is most plausible in the case of the external props of written text
and spoken words. For interactions with these external media are ubiqui-
tous (in educated modern cultures), reliable and developmentally basic.
Our biologic brains, after learning, expect the presence of text and speech
as much as they expect to encounter weight, force, friction and gravity.
Language for us is a constant, and as such can be safely relied upon as
the backdrop against which on-line processes of neural computation take
shape and develop.

This looks to be among the earliest versions of the causation-constitution fallacy.


Texts and speech play such an important causal role in the shape and flow of our
thoughts that we should think that these texts and speech constitute parts of our
thought. The fallacy lies in supposing that the existence of a causal connection, even
one that is reliably available, is sufficient for constitution.20 This is a well-worn topic
that perhaps needs no further attention here.
But consider Clark’s treatment of Edwin Hutchins’s theory of ship navigation in
light of our earlier distinction between cognition and behavior, a distinction that
Clark at least sometimes appears to accept and that would seem to be required for a
nontrivial version of the hypothesis that cognition extends. Clark (1998, 181) writes:

The overall cognitive competencies which we identify as mind and in-


tellect may thus be more like ship navigation than capacities of the bare

19
For more discussion, see Adams and Aizawa 2008, 133–143.
20
For more discussion, see Adams and Aizawa 2008.
42 Extensions and Alterations

biological brain. Ship navigation (see Hutchins 1995) is a global emergent


from the well-orchestrated adaptation of an extended complex system
(comprising individuals, instruments, and practices). Much of what we
uncritically identify as our mental capacities may likewise, I suspect, turn
out to be properties of the wider, extended systems of which human brains
are just one (important) part.

But cognitivists generally take navigating a ship to be something like a group be-
havior. It is a behavior that a group undertakes together. But if that is right, then
how can that be squared with the distinction between cognition and behavior?
Here it looks as though Clark is proposing that we take cognitive competences to be
behaviors. Ship navigation does seem to be like something one might call a global
emergent of individuals, instruments, and practices. Behaviors are often emergent.
Jack’s asking Jill to pass the salt and her complying are behaviors that are the product
of the individuals, instrument (salt shaker), and practices (such as eating food with
salt and having the person nearest the shaker pass it). So it looks again as though
Clark is proposing that we treat behavior as cognition. Or take some of the later
discussion,

it is our capacity to create and operate upon external representations that


allows us to use manipulations of the physical environment as integral
parts of so many of our problem-solving routines. In thus reaching out to
the world we blunt the dividing line between the intelligent system and
the world. We create wider computational webs whose understanding and
analysis may require us to apply the tools and concepts of cognitive science
to larger, hybrid entities comprising brains, bodies and a variety of external
structures, traces and processes. (Clark 1998, 182)

If we take the wider computational webs to be behavior, then we would be at a point


where we could apply the tools and concepts of cognitive science to the kinds of
hybrid entities that Clark proposes. But is Clark proposing that we treat behavior
as cognition?

5. Conclusion
Clark (1998, 2006a, 2006b) offers any number of surprising conjectures regarding
the role of language in cognition. My principal goal here has been to invite him to
clarify and elaborate some of these conjectures. Does Clark accept a distinction be-
tween cognition and behavior? Why or why not? In what sense is the complemen-
tarity model an alternative to the translation model? What are the predictions of the
complementarity model for the metacognitive abilities of languageless adults? Does
Clark on Lang uage, Cog ni tion , and E x tended Cog niti on 43

the hypothesis of extended cognition found in Clark (1998) amount to the hypoth-
esis of what mainstream cognitive scientists would describe as the hypothesis of
extended behavior? Why or why not?21

References
Adams, F., and K. Aizawa. 2001. “The Bounds of Cognition.” Philosophical Psychology 14(1): 43–64.
Adams, F., and K. Aizawa. 2008. The Bounds of Cognition. Malden, MA: Wiley-Blackwell.
Aizawa, K. 2014. “The Enactivist Revolution.” AVANT. Pismo Awangardy Filozoficzno-Naukowej
(2): 19–42.
Aizawa, K. 2015. “What Is This Cognition That Is Supposed to Be Embodied?” Philosophical
Psychology 28(6): 755–775.
Aizawa, K. 2017. “Cognition and behavior.” Synthese 194(11): 4269–4288.
Carruthers, P., and J. Boucher. 1998. Language and Thought: Interdisciplinary Themes. Cambridge,
UK: Cambridge University Press.
Clark, A. 1998. “Magic Words:  How Language Augments Human Computation.” In Language
and Thought:  Interdisciplinary Themes, edited by P. Carruthers and J. Boucher, 162–183.
Cambridge, UK: Cambridge University Press.
Clark, A. 2000a. “Making Moral Space:  A Reply to Churchland.” Canadian Journal of Philosophy
30(supl): 307–312.
Clark, A. 2000b. “Word and Action:  Reconciling Rules and Know-how in Moral Cognition.”
Canadian Journal of Philosophy 30(sup1): 267–289.
Clark, A. 2006a. “Language, Embodiment, and the Cognitive Niche.” Trends in Cognitive Sciences
10(8): 370–374.
Clark, A. 2006b. “Material Symbols.” Philosophical Psychology 19(3): 291–307.
Clark, A. 2015. Surfing Uncertainty:  Prediction, Action, and the Embodied Mind. Oxford:  Oxford
University Press.
Clowes, R. W., and A. F. Morse. 2005. “Scaffolding Cognition with Words.” In Proceedings of 5th
International Workshop on Epigenetic Robotics:  Modeling Cognitive Development in Robotic
Systems, edited by L. Berthouze, F. Kaplan, H. Kozima, Y. Yano, J. Konczak, G. Metta, J.
Nadel, G. Sandini, G. Stojanov, and C. Balkenius, 101–105. Lund, Sweden: Lund University
Cognitive Studies.
Dennett, D. 1991. Consciousness Explained. New York: Little, Brown.
Fodor, J. 1983. The Modularity of Mind: An Essay on Faculty Psychology. Cambridge, MA: MIT Press.
Hutchins, E. 1995. Cognition in the Wild. Cambridge, MA: MIT Press.
Rowlands, M. 2010. The New Science of the Mind: From Extended Mind to Embodied Phenomenology.
Cambridge, MA: MIT Press.
Schaller, S. 2012. A Man without Words. Berkeley: University of California Press.
Shapiro, L. 2013. “Dynamics and Cognition.” Minds and Machines 23(3): 353–375. doi: 10.1007/
s11023-012-9290-2.

21
Many thanks to Matteo Colombo and Liz Irvine for helpful comments on an earlier draft of this
paper. Unfortunately I could not pursue all of their promising leads.

You might also like