Phenomenology and the Cognitive Sciences

, Volume 9, Issue 4, pp 579–603

The value of cognitivism in thinking about extended cognition


  • Frederick Adams
    • Department of Linguistics and Cognitive ScienceUniversity of Delaware
    • Department of PhilosophyCentenary College of Louisiana

DOI: 10.1007/s11097-010-9184-9

Cite this article as:
Adams, F. & Aizawa, K. Phenom Cogn Sci (2010) 9: 579. doi:10.1007/s11097-010-9184-9


This paper will defend the cognitivist view of cognition against recent challenges from Andy Clark and Richard Menary. It will also indicate the important theoretical role that cognitivism plays in understanding some of the core issues surrounding the hypothesis of extended cognition.


CognitivismExtended cognitionCoupling-constitution fallacyNon-derived contentMental representation

It is mere prejudice to suppose that cognitive processes occur within the brain. There is simply no principled, non-question-begging, reason to think that cognition only occurs within the confines of the brain or central nervous system. Such is a common charge in the extended cognition literature.1 But, in truth, a principled basis for thinking that today, cognitive processes typically occur only within the brain, or central nervous system, is in plain view. It lies in the familiar cognitivist view that cognition involves certain sorts of manipulations of non-derived representations. One might plausibly say that cognitive processes typically take place in the brain, because that is where representational manipulations of the sort that constitute cognition typically happen to be. This is no kind of logical or conceptual truth. It is merely a contingent feature of the way the actual world happens to be organized in the early twenty-first century.2

A central theme of Adams and Aizawa (2001, 2008) presses the foregoing point. The advocates of extended cognition need to address the ability of cognitivism to demarcate the regions of the world that contain cognitive processes versus those that do not. Indeed, advocates of extended cognition need to come to grips with some of the other virtues of cognitivism. Cognitivism does more than provide a non-question-begging reason to think that the mind is in the brain or central nervous system. For one thing, cognitivism helps us avoid the most common fallacy in the extended cognition literature, namely, the “coupling-constitution fallacy.”3 Too easily, the advocates of extended cognition succumb to the belief that if a cognitive agent causally interacts with some object in the external world in some “important” way—if that agent is coupled to an object—then that agent’s cognitive processing is constituted by processes extending into that object. Yet, causal interactions do not always change the types of processes involved. Take a burning match to a piece of paper and the combustion process will extend into the paper, but take the same match to a steel wrecking ball and it will not extend into the ball. What will extend and what will not depends on what is coupled. With the cognitivist theory of cognition in hand, one can ask of any specific case whether coupling the cognitive process to the bodily or environmental process will lead to the cognitive process extending. Will the extended portion of the process involve the requisite types of manipulations of representations? There are probably other theories of cognitive processes that could do as much as cognitivism does to help avoid the coupling-constitution fallacy, but cognitivism still deserves credit for doing this.

A second virtue of cognitivism for the extended cognition debate is that it helps us define a field of scientific investigation. It offers us a relatively well-defined subject for a field of cognitive science to study. It is the study of some, but only some, of the ways in which some organisms transform information, ways it might be possible to recreate in computers or robots. By contrast, the advocates of extended cognition have not reached consensus on what they might pursue as an extended cognitive science. If one thought that extended cognitive processes involved certain sorts of transformations of certain sorts of mental representations, one could be a cognitivist about larger regions of space–time than has been traditional.4 Some advocates of extended cognition, however, have suggested that cognitive processes are dynamical systems processes.5 Of course, presumably not just any dynamical systems processes will do, since the pendulum in a grandfather clock is a dynamical system, but presumably not a cognitive system. Maybe when the dynamical systems approach is fleshed out, it will turn out that only a dynamical system that also meets the requirements of cognitivism would work. Or maybe some form of ecological psychology or phenomenology will do. Or maybe we do not even need a science.6

Despite these virtues that cognitivism enjoys, not to mention the ways in which it supports and makes sense of various empirical findings, Richard Menary (2006) and Andy Clark (2005, 2007, 2008, 2010) take exception to the case we have made on behalf of cognitivism. They resist Adams and Aizawa’s appeal to the hypothesis that cognition involves non-derived representations.7 They persist in succumbing to the coupling-constitution fallacy. Finally, they have not provided a plausible successor to the idea that cognitive science is about some of the ways in which some organisms manipulate some representations. They do not have a plausible account of what cognitive science is a science of. This paper will elaborate on these issues. “Non-derived content” will defend the hypothesis that cognition involves non-derived content; “The coupling-constitution fallacy” will show how both Clark and Menary fail to avoid the coupling-constitution fallacy, and, finally, “A principled internalist view of the mind and how to make a brain-tool science” will turn to the idea that a “brain-tool science,” even construed as a science of what complements the brain, is a motley undisciplined subject for a would-be science. A word of caution before we begin: this discussion is not for the faint of heart, as it is the nth move in a protracted debate. It will presuppose some familiarity with a debate in progress.

Non-derived content

In keeping with the principle that the best defense is a good offense, Menary and Clark defend extended cognition by challenging one component of our version of cognitivism, the idea that it involves intrinsic, or non-derived, representations.8 So, suppose we concede that there is no such thing as intrinsic or non-derived representation. This, of course, leaves untouched the idea that part of what distinguishes cognitive processes from non-cognitive processes is the way in which cognitive processes transform or manipulate mental representations. Cognitive processing is not just any old type of symbol manipulation or information processing.9 Like most cognitivists, we hypothesize that cognitive processing involves specific sorts of information processing not generally present in the familiar interactions humans have with tools in their environments. The differences in the ways digital video recorders and human visual systems process the information in light separates consumer electronics from vision science. This is part of the reason that cognitivism gives us some means of drawing a principled distinction between what goes on inside the brain and what goes on in the brain, body, and environment.

Clark and Menary’s challenge to non-derived content also leaves untouched the idea that cognition involves representations, even if they are not non-derived representations and even if the derived/non-derived distinction were to prove untenable or ill-founded. The weaker hypothesis that cognition involves representations is sufficient to challenge at least some of the purported instances of extended cognitive processing. One instance is Clark and Chalmers’ discussion of the play of Tetris. In this game, blocks descend from the top of a computer screen and are to be fit into a rising wall of blocks. Aside from differences in the way in which information is processed, there are different roles for representations between playing this game by using mental rotation and by pressing a game controller button. In mental rotation, one is manipulating representations of the falling zoids. In pressing the game button, one is manipulating the falling zoids themselves. The zoids on the screen are not representations.10 A similar point applies to the idea of Robert Gibbs (2001) that a windsurfer’s interactions with board and sail create extended intentions. The fact that nothing in the board or the sail is a representation gives us one reason to think that it does not involve extended intentions or extended cognition. So, even the weakened theory of the mark of the cognitive is not entirely ineffectual in resisting some purported cases of extended cognition.

Now that we see the scope of Clark’s and Menary’s challenge, we can turn to the hypothesis of intrinsic content per se. Here, two preliminary points will aid us in meeting Clark’s and Menary’s objections. The first thing to observe about the derived/non-derived distinction is that it concerns the conditions in virtue of which an object bears a particular content. A thought might bear the content that the cat is hungry in virtue of satisfying some conditions on non-derived content, whereas a particular inscription on a piece of paper might bear that same content by satisfying some other conditions on derived content. To put the matter another way, there are two questions one might ask of a representation. The first is what content that representation bears; the second is what conditions make it the case that it bears that content.

The second thing to observe is that the content of one representation can be another representation. There can be representations of representations. “+” represents the addition function, where “the plus sign” represents the symbol “+.” A stop sign has the content of the imperative “stop!,” but the expression “stop sign” is a representation of stop signs, which bear the imperative content. And the process can iterate. One might take a photograph of the expression “the plus sign,” which would be a representation of the expression “the plus sign,” which would be a representation of “+,” which represents a mathematical function. Keeping these two observations clearly in mind takes us a long way toward steering clear of Clark’s and Menary’s objections to intrinsic content.

Clark on intrinsic representation

The principal challenge of Clark (2005) to intrinsic representations proceeds in two stages.11 In essence, the first stage begins with someone, say Gary, thinking about the quasi-syllogism,

All men are mortal

Socrates is a man

Therefore, Socrates is mortal.

Gary thinks about this syllogism using the familiar apparatus of Venn diagrams drawn with pencil and paper as shown in Fig. 1. In checking for the validity of this syllogism, Gary relies on numerous conventions, such as that circles represent sets, that one circle’s being inside of another means that all the members of the first set are members of the second, and that an “X” inside of a circle represents an individual being a member of the set. Now, however, suppose that, instead of using pencil and paper to draw out the circles and “X,” Gary relies upon mental imagery which we can depict in Fig. 2 as a kind of fuzzy, low-resolution representation of Fig. 1. As Clark fully appreciates, this situation is no challenge to our view. The items in Fig. 1 can have their content in virtue of social conventions, hence be derived representations, where the items in Fig. 2 can have their content in virtue of satisfying conditions on non-derived representations.12
Fig. 1

A pencil and paper drawing of Gary’s Venn diagrams
Fig. 2

Gary’s mental image of Venn diagrams

The trap for our view is then supposed to be tripped in the second stage of Clark’s argument. In this stage, we are invited to consider yet another thought experiment. Now, instead of a normal human being like Gary, we are invited to consider a Martian whose mental representations ex hypothesi do not satisfy the conditions of any true theory of non-derived representations. Instead, the Martian’s mental images are, say, mere bit maps of the sort created by digital scanners. So, where representations in Gary’s version of Fig. 2 have the content they do in virtue of conditions on non-derived content, the representations in the Martian’s version of Fig. 2 have the content they do in virtue of conditions on derived content. But, in this situation, Clark believes, we should unhesitatingly accept the view that this Martian does the same cognitive processing as does Gary.

The tension here is supposed to be between the theory that cognition involves non-derived representations and,…, well,…what? Clark is not explicit about what he is pitting against our theory, but it looks like mere pre-theoretical intuitions about what we would call cognition. We do not put a lot of weight on thought experiments that use behavior to determine whether some alien apparatus is alive or not, so we also do not put a lot of weight on thought experiments that use behavior to determine whether some alien apparatus is thinking or not. For scientific purposes, we ought to rely on deep theoretical similarities and differences, rather than on superficial pedestrian responses. Only a logical behaviorist would maintain that behavioral similarity to bodily movements produced by genuine cognitive processes is sufficient for attributing genuine cognitive processes. Since Clark is no behaviorist, he cannot take behavioral similarity of the Martian and Gary to be decisive. If, however, behavior is only evidence for cognition, then whether there is genuine cognition in the Martian depends on how the behavior (similar to Gary’s) is being produced. Unless Clark gives us a genuine (and good) reason to believe the internal mechanisms in the Martian are cognitive, Clark’s efforts are not all that threatening.

Menary’s follow-up on intrinsic representations

Menary’s handling of the hypothesis of intrinsic content begins with Clark’s example of Venn diagrams but develops in different ways. We will consider each of these in turn. To be fair with Menary, let us quote his first objection in toto, then see how it goes wrong.

But one problem here is that the processes that apply to an image of a Euler circle are not the same as the processes that apply to the Euler circle in virtue of its conventionally determined meaning. Grant the internalist the point that the image of a Euler circle gets in my head because of some causal linkage with external Euler circles. The inferences that I make that involve Euler circles depend upon the conventions governing the properties and uses of Euler circles, something that the image cannot provide. So there are also mental representations of the conventions governing the properties and uses of Euler circles, which get in my head because of some causal linkage (asymmetric dependence say) with the outside world. But it is not the causal linkage that determines the content of these representations: the content of each of these representations is the convention. So, unless there are mental representations with conventional content, there can be no cognitive processing of Euler circles. (Menary 2006, p. 336).

The first sentence seems to be driving at the following. There are certain cognitive processes that will apply to the mental image shown in Fig. 2 when it is treated merely as an image with circles and writing and there are other cognitive processes that apply to the mental image shown in Fig. 2 when it is treated as a mental image for evaluating quasi-syllogisms. In particular, as Menary proposes in the third sentence, the cognitive processes in the second case presumably involve mental representations of certain conventions, such as that one circle being inside another means that one set is a subset of another. Further, Menary supposes that, while the image of the Euler circles shown in Fig. 2 may get their meaning in virtue of some causal theory of mental content (the second sentence), the mental representation of the conventions do not get their content from the mental image in Fig. 2 (per the last part of the third sentence). So, in the fourth sentence, we come to the conclusion that there are mental representations of the conventions, such as that one circle being inside another means that one set is a subset of the other, and that these mental representations get their content via some causal theory of reference. So far, so good. What remains is hard to interpret. The problem apparently finally arises in the second to last sentence. The mental representations of the conventions cannot get their content by causal linkages. Why? Because the content of the representations is the convention. So, the problem is not with the mental representation illustrated in Fig. 2, but with the mental representations of the conventions, such as what it means for one circle to be contained within another. From this, Menary gets his final point that unless there are mental representations with conventional content, there can be no cognitive processing.

So, let us go back to the claim that the mental representations of the conventions cannot get their content by causal linkages. First, we must emphasize that, for purposes of engaging the hypothesis of extended cognition, we are officially agnostic about what particular conditions might confer non-derived content on mental representations. One might give a causal theory of mental content, a picture theory, or some form of functional role semantics. Second, even when we have entertained causal theories of mental content, we have never entertained the idea that all mental representations get their content in virtue of some causal theory of mental content. Though this view is often attacked, it is not clear that any philosophers hold this view.13 Third, it is unclear that Menary has a compelling reason for this claim about the mental representations of conventions. Why does the fact that the content of these representations is a convention show that the causal theory does not apply? After all, it just is the causal view that a syntactic item “X” can mean X and be caused by X. So, why can we not have a syntactic item “X” that (1) means, say, circle containment means set theoretic containment and (2) is caused to tokened by circle containment meaning set theoretic containment? Menary owes us an argument here.

Perhaps, the idea is that a convention is not the kind of thing to which a human being, or another cognitive agent, could be causally linked, much in the way that an abstract object, such as a number, is not the sort of thing to which a cognitive agent could be causally linked. But, even if the causal theorist concedes that it is (conceptually or metaphysically or nomologically) impossible for cognitive agents to be causally linked to conventions, the causal theorist still has options. Many causal theorists believe that mental representations occur in a language of thought, a system of syntactically and semantically combinatorial representations in many respects like a first-order language. In such a language of thought, there will be semantically atomic and semantically molecular representations. The causal theorist can, therefore, first, maintain that mental representations of conventions are complex or molecular mental representations that get their meanings by way of the meanings of certain atomic mental representations and the way in which those mental representations are composed. Then, the atomic mental representations get their meaning by way of causal connections. So, a mental representation M1 could have the content “‘i’ before ‘e’ except after ‘c’” in virtue of the meanings of “i,” before, “e,” etc. and the way in which those components are put together. So, even if we were defending a causal theory of mental content in our “intrinsic content” hypothesis (which we were not), we would not necessarily have a problem. Menary apparently underestimates the resources of a causal theory of mental content.

But, suppose that Menary can come up with a good reason to think that the mental representations of the conventions cannot be generated by a causal theory. It still does not follow that the contents of these mental representations are generated by conventions. Return to the final sentence quoted above, namely, “unless there are mental representations with conventional content, there can be no cognitive processing of Euler circles.” We need to be clear here. The phrase “mental representations with conventional content” is ambiguous. On the one hand, it can mean mental representations whose content is determined by convention. We maintain that there are no such things. On the other hand, it can mean mental representations whose content is that of a convention, such as that circle containment means set theoretic containment. We maintain that there are plenty of these. When one typically thinks “‘i’ before ‘e’ except after ‘c,’” one is likely tokening a mental representation whose content is a convention, but not a mental representation whose content is conventionally determined. Presumably Menary means to conclude that unless there are mental representations with content determined by conventions, there can be no cognitive processing of Euler circles (in syllogistic reasoning). But, this conclusion presupposes that if the mental representations of the conventions do not get their content by causal connections, then they must get them by way of conventions. But there are other options. Even if the mental representations of conventions do not get their content from causal connections, it does not follow that they much get their content by conventions. So much, then, for Menary’s first objection to the intrinsic content idea.

Menary mixes his exposition of the foregoing argument with another. He writes,

But one problem here is that the processes that apply to an image of an Euler circle are not the same as the processes that apply to the Euler circle in virtue of its conventionally determined meaning. … internal processes don’t apply to the image of an Euler circle in the same way that we directly manipulate the external Euler circle. Therefore, we cannot carry out the same operations on Euler circles in the head that we can by directly manipulating them if we are guided by A&A’s stipulation. … By A&A’s reasoning, cognition that involves representations with intrinsic content would be quite limited; but clearly we aren’t so limited. This is because we have developed external representations schemes and methods for manipulating them (ibid., pp. 336–7).

In the text leading up to this, Menary evidently supposes that he needs to argue that on our view, the intracranial processes that manipulate mental images of Euler circles are different from the extracranial processes that manipulate Euler circles on paper. But, that is really not necessary, since much of our argumentation against “pencil and paper memories” has been directed toward pointing to the many ways in which internal cognitive processes differ from external physical, chemical, and biological processes. We have been at pains to note how Otto’s jottings in his notebook involve different sorts of information processing than do Inga’s “jottings” in her brain. We have been at pains to note how normal mental rotation differs from physical rotation of zoids on the Tetris screen. We have been at pains to note how mental arithmetic differs from arithmetic with pencil and paper.

Menary thinks these differences in processing are a problem for us because it would mean that human cognitive processing would be quite limited. But, it is, in fact, our view that humans have more or less fixed cognitive capacities. That seems to us pretty evident. Indeed, it is because our cognitive capacities are so limited that we have recourse to tools to overcome our limitations. It is the simple fact that doing things with tools so frequently enables us to complete tasks so much more quickly, reliably, or easily that we develop them and rely on them. Why, for example, do we so often rely on pencil and paper to do arithmetic problems? Because we have a relatively limited short term memory capacity, but by writing the numbers in columns, marking numbers to carry, and so forth, we free ourselves from our dependence on our highly fallible short-term memory. And why do we still more often rely on inexpensive pocket calculators, rather than pencil and paper? Because using a calculator is generally faster and more reliable than using pencil and paper. With a calculator, one does not have to write the problem down nor correctly calculate the elementary products in a large multiplication problem. We see nothing embarrassing in this at all. We are cognitively limited, and the use of non-cognitive tools helps us overcome these limitations.14 So, we see nothing wrong with drawing a contrast between intracranial addition and paper and pencil addition. Nor do we see that this commits us to making incorrect assessments of human cognitive limitations.

Menary’s next objection is that we have an impoverished view of mental content. He writes,

the mental image of an aardvark is not the sole constituent of my concept of an aardvark. In fact most of the content of my concept of an aardvark will have been fixed by the conventional methods that A&A find anathema. Concepts go beyond what asymmetric dependence or function of indication can offer (ibid.)

Several things are going on here. One is that Menary seems to think that Adams and Aizawa think that all concepts are mental images. That would explain his claim that the mental image of an aardvark is not the sole constituent of one’s concept of an aardvark. But happily, we grant that not all concepts are mental images. We can also admit that non-derived representations indirectly help us acquire concepts. So, for example, the English word “aardvark” would be useful in talking about aardvarks and telling people about some of their features. And talking about aardvarks is probably a good way to get the concept of an aardvark. And, indeed, this utility of derived representations is plausibly construed as part of the reason we have non-derived representations for things in the first place. Note that non-derived representations enable us to acquire concepts, but this is not the same thing as saying that non-derived representations enable us to develop concepts by being directly imported into the brain, as if that makes any sense. We do not think that images can be somehow directly inserted or implanted into one’s cognitive economy. And, we find here again that Menary thinks we are committed to some specific causal theory of mental content. But, again, for the space of our challenge to the hypothesis of extended cognition, we are not specifically committed to any particular theory of how non-derived content might arise. So, for present purposes it does not matter whether Jerry Fodor’s asymmetric causal dependency theory does not work or whether Fred Dretske’s function of indication theory does not work.15

Finally, we arrive at that purported dilemma we mentioned earlier:

Either A&A’s intrinsic mental representations will be too rich—too similar to conventional representations—such that the intrinsic-conventional distinction becomes vacuous or they will be too meager, in which case they won’t be of much use in completing cognitive tasks. Either way the intrinsic-derived distinction looks unhelpful (ibid).16

One might wonder why intrinsic mental representations must be “too rich” or “too meager,” rather than just right. Menary does not spell this out. But, set this aside and accept the premise merely for the sake of argument. As we see things, neither case is really harmful. Let the set of all non-derived mental representations and the set of all derived representations have exactly the same contents. This would make mental representations “maximally rich.” Why would this make the distinction between the two types vacuous? Indeed, one can readily see a reason why it would be a good thing to have all the non-derived contents expressible in a derived fashion as well. Suppose that Gary has a vast set of thoughts in a system of non-derived mental representations squirreled away in his brain, one of them being that the icy water of the lake looks inviting. Further suppose he wants to communicate this last fact to us. It would be convenient for Gary to have a set of publicly available derived expressions, such as those of a natural language that would enable him to communicate his thoughts. And if he wanted to share his thoughts for posterity, it would be convenient to have a system of writing that could express his thoughts for the ages.17 So, as far as we can tell, having a “maximally rich” system of derived representations would seem to have its usefulness, since it would be all the better for communicating Gary’s vast set of thoughts. And, even something less than maximally rich might still be pretty good as well (this is essentially the point we made above in response to Menary’s aardvark example. We do not see a problem here. In this case, we can see what is meant by the derived/non-derived distinction and why it is important). So, the first horn of the purported dilemma does not appear to constitute a problem for us.

Let there be a vast gulf between the contents expressible using non-derived representations in the mind and the contents expressible using derived representations in the physical, chemical, and biological world. If that were the case—if the system of mental representations were “too meager,” one could understand the struggle to develop other forms of communication and intellectual inspiration, such as perhaps some of those found in mathematics, music, and art. So, as in the first horn of Menary’s dilemma, the derived/non-derived distinction is unthreatened. As best we can tell, whatever relations there might be between the contents of our hypothetical non-derived representations and of derived representations, the distinction offers us at least the beginnings of an attempt to understand some features of the human condition.

The coupling-constitution fallacy

Move now from the rather focused issue of non-derived content to the most pervasive type of argument for the hypothesis of extended cognition, the coupling-constitution arguments. In these arguments, one uses certain sorts of observations about a cognitive agent’s causal interactions with the external, non-biological world—a cognitive agent’s “coupling” with the world—to make the case that part of the external, non-biological world realizes an agent’s cognitive processes. Clark and Chalmers (1998) present the argument quite succinctly: “we will argue that beliefs can be constituted partly by features of the environment, when those features play the right sort of role in driving cognitive processes. If so, the mind extends into the world” (Clark and Chalmers 1998, p. 12). Clark himself gives a lengthier version of the argument in this passage,

Confronted, at last, with the shiny finished product the good materialist may find herself congratulating her brain on its good work. But this is misleading. It is misleading not simply because (as usual) most of the ideas were not our own anyway, but because the structure, form and flow of the final product often depends heavily on the complex ways the brain cooperates with, and leans upon, various special features of the media and technologies with which it continually interacts.... The brain’s role is crucial and special. But it is not the whole story. In fact, the true (fast and frugal!) power and beauty of the brain’s role is that it acts as a mediating factor in a variety of complex and iterated processes which continually loop between brain, body and technological environment. And it is this larger system which solves the problem.... The intelligent process just is the spatially and temporally extended one which zig-zags between brain, body, and world (Clark 2001, p. 132).

Here is Menary setting out the argument quite succinctly,

The real disagreement between internalists [like Adams and Aizawa] and integrationists [like Menary] is whether the manipulation of external vehicles constitutes a cognitive process. Integrationists think that they do, typically for reasons to do with the close coordination and causal interplay between internal and external processes (Menary 2006, p. 331).

So, both Clark and Menary apparently endorse this kind of argument, at least at times.

The fallacy in this kind of argument is easy to see. This simply lies in the fact that, in general, it does not follow from the fact that a process X is coupled or causally connected to a cognitive process Y that X itself, or the entire X–Y process, is cognitive. Stating the problem a bit more generally, the fact that a process X is coupled to a process of type Y does not show that that X is, in fact, also a (part of a) Y process or that the entire X–Y process is of type Y. In an airplane, the combustion of fuel can be coupled to the rotation of a propeller, but under normal circumstances the combustion is limited to the engine. The same point can be made with a biological process. In the rods of the human eye, the isomerization of rhodopsin is coupled by a biochemical pathway to the release of the neurotransmitter glutamate, but the isomerization is a change in the structure of the photopigment molecule. The isomerization is limited to the rhodopsin molecule and does not extend throughout the entire rod or into the cell membrane. What the coupling-constitution arguments overlook is the fact that, in general, the power of linking components together does not always come by linking together components that do the same thing. It is by combining components that perform different processes that one typically gets something useful.18

On presenting the coupling-constitution fallacy, one sometimes encounters the charge that the coupling-constitution fallacy involves the imposition of a (dubious, armchair) metaphysical distinction on the proponents of extended cognition and that the distinction between coupling and constitution should be rejected by advocates of extended cognition. Rather than relying on antiquated metaphysical principles, we should let the best science of the mind inform our views of the bounds of cognition19 The principal problem with this reply is that a coupling-constitution distinction appears to be central to the hypothesis that cognitive processes are realized by the brain, body, and environment. Rejecting or doing away with the distinction is to abandon the most common means of articulating the extended cognition hypothesis. In the passage cited above, for example, Menary frames the issue between internalists and integrations as to whether the mind is constituted by processes involving environmental tools or, as we would say, the mind merely causally interacts with the environment by way of tools. Clark (2007), for his part, accepts a framing of the issue of extended cognition put forth in Rupert (2004). According to Rupert, the hypothesis of embedded cognition (HEMC) asserts that

Cognitive processes depend very heavily, in hitherto unexpected ways, on organismically external props and devices and on the structure of the external environment in which cognition takes place,

where the hypothesis of extended cognition (HEC) maintains that

Human cognitive processing literally extends into the environment surrounding the organism, and human cognitive states literally comprise—as wholes do their proper parts—elements in that environment.

A coupling-constitution distinction is implicit in this dichotomy. Where HEMC maintains that cognitive processes causally depend on external tools and feature of the environment, HEC maintains that cognitive processes constitutively depend on external tools and features of the environment. If Clark abandons the distinction between causation and constitution, hence between HEMC and HEC, then what is left for Clark to defend?

So, Clark and Menary, like at least many proponents of extended cognition, cannot readily abandon a distinction between metaphysical relations such as causation and constitution. They can, however, in principle, give up on arguments that begin with causal relations and infer a constitutive or realization relation. They could instead try to make a case for extended cognition based on some form of cognitive equivalence argument. They could try to provide real world cases in which the processes that take place inside of brains are cognitively equivalent to processes that propagate through brain, body, and environment, hence that we should recognize these cases are instances of extended cognition. Here is where cognitivism could help. Clark and Menary could try to argue that processes that take place inside the brain manipulate non-derived mental representations in a particular sort of way and that brain–body–environment processes do as well. But, Clark and Menary seem to reject cognitivism. So in order to invoke cognitive equivalence arguments, they would have to invoke some other plausible theory of when two processes are cognitively equivalent. But, neither Clark nor Menary nor any other advocates of extended cognition seems to have a theory of what cognition is that does justice to what cognitive psychologists have been investigating for decades. These sorts of considerations suggest that, all told, it is too much trouble for the advocates of extended cognition simply to abandon the coupling-constitution arguments. Instead, they may wish to persist, as do Clark and Menary, in trying to find a way to make these sorts of arguments compelling.

Clark’s hippoworld argument

Having already replied to much of Clark’s rejoinder in defense of coupling-constitution arguments,20 we here limit ourselves to one argument that we have yet to address. This argument might be thought to vindicate coupling-constitution arguments. This is Clark’s “Hippo-world” thought experiment, which goes something like this.21 For idiosyncratic historical reasons (perhaps the enduring influence of the philosophical “picture” of one Hippo-Descartes), all neuroscientific attention is focused on the hippocampus. Several decades of research using single cell recording techniques, brain lesions in animals, and so forth has lead to important and replicable findings about the processes that occur within the hippocampus. Scientists on Hippo-world are nothing if not inventive and one day some turn their scientific attention to the rest of the brain, where they too begin to make some significant progress. They discover new neural circuits and processes that are linked to processes within the hippocampus. And among these new scientists, there are some who are so bold as to conjecture that cognitive processes occur within the whole of the brain. Inspired by this new science, some philosophers, Hippo-Clark and Hippo-Chalmers, among countless others, openly disavow the remnants of the traditional Hippo-Cartesian view of the mind and declare that “Cognitive processes ain’t (all) in the hippocampus!”

Yet, there are philosophers who resist the enthusiastic philosophical interpretations of the studies of the entire brain. Some philosophers, such as Hippo-Adams and Hippo-Aizawa, conjecture that, in its studies of the hippocampus, science has discovered the scientific essence of cognition itself. More boldly, these philosophers conjecture that what occurs within the hippocampus is cognitive, where what occurs within the other regions of the brain is of a distinct non-cognitive character. These philosophers maintain that genuinely cognitive processes, processes bearing the mark of the cognitive, take place within the hippocampus, where other supportive non-cognitive processes take place within the other regions of the brain.

In reflecting upon Hippo-world, Clark wants his reader to have the impression that the critics of whole brain cognition of the Hippo-A-Team variety have an overly narrow vision of what cognition is. Like Adams and Aizawa of earthly fame, Hippo-Adams and Hippo-Aizawa display an overly restricted conception of cognition. In Hippo-World, there are extra-hippocampal processes that complement those of the hippocampus and that are integrated with hippocampal processes to augment the power and scope of human intelligence. Similarly, on earth, there are bodily and environmental processes that complement those of the brain and that are integrated with brain processes to augment the power and scope of human intelligence. Hippo-Adams and Hippo-Aizawa should accept extra-hippocampal brain processes as genuinely cognitive processes, just as Adams and Aizawa should accept environmental processes as genuinely cognitive.

Unfortunately for Clark, such prima facie plausibility as the Hippo-World argument might enjoy evaporates on more careful examination. The Hippo-World account relies upon a kind of philosophical misdirection. In describing the case, Clark provides the usual extended cognition sorts of observations of the causal roles played by the surrounding structures and the interplay between them. He observes the coupling of the hippocampus and the extra-hippocampal brain and the way in which the whole can do more than the isolated parts. And in this case, one is inclined to think that the whole of the brain is indeed a cognitive processor. One, therefore, might suppose that coupling and integration can indeed lead to the extension of cognition. One gets the distinct impression that the whole brain is cognitive in virtue of the coupling of the extra-hippocampal regions to the hippocampal regions. The misdirection, however, becomes apparent when we note that, it is perfectly legitimate to suppose that the whole of the brain is indeed a cognitive processor, but not in virtue of the fact that it is coupled or integrated with the hippocampus. What makes the whole of the brain a cognitive processor is the fact that the whole bears the mark of the cognitive?22 As cognitivists would point out, the whole of the brain realizes processes that transform or manipulate non-derived representations. This has nothing to do with the coupling considerations brought forth by many advocates of the hypothesis of extended cognition.

But, there is also a second way in which one might interpret Hippo-Adams and Hippo-Aizawa more sympathetically. Although it is Clark’s thought experiment and he is free to develop it as he wishes, let us suppose that the Hippo-world brain, just like the real world brain, is a cognitive processor. If so, then by our lights, there will be non-derived representations in the hippocampus and in the extra-hippocampal regions of the brain. Further, in both worlds the brain will carry out transformations or manipulations of these non-derived representations. So, there would be this level of similarity. Nevertheless, there would still be some point to Hippo-Adams and Hippo-Aizawa drawing a theoretical difference between intra-hippocampal processes and extra-hippocampal brain processes. What Hippo-Adams and Hippo-Aizawa might be trying to bring to the attention of the scientific and philosophical community are the ways in which, as a matter of contingent empirical fact, hippocampal processes differs from extra-hippocampal brain processes, not to mention environmental processes. The hippocampus carries out, let us say, spatial mapping of the environment, but not visual or auditory or linguistic processing. Part of what this means is that there are different kinds of information processing going on in the hippocampal and the extra-hippocampal brain regions, not to mention in the external environment.23 Neither Hippo-Adams and Hippo-Aizawa nor Adams and Aizawa are satisfied to use “cognition” as a label for just any old information processing on non-derived representations. Hippo-Adams and Hippo-Aizawa share with Adams and Aizawa a concern over the indiscriminate lumping together of distinguishable types of processes. Hippo-Adams and Hippo-Aizawa wish to distinguish hippocampus processes from more general “cognitive processes,” where Adams and Aizawa wish to distinguish cognitive processes from more general causal processes. There is sense to what the Hippo-A-Team is saying, just as there is sense to what the Earthly-A-Team is saying. So, it seems that the discussion of Hippo-world does not avoid the tendency to fall prey to the coupling-constitution fallacy.

Menary on coupling-constitution arguments

Menary’s reply to the coupling-constitution fallacy has two parts. The first is that we are begging the question against the hypothesis of extended cognition; the second is that we misunderstand how the argument is supposed to work. To meet Menary’s first challenge, we need only tweak the details of the way in which we state the coupling-constitution fallacy. To meet the second, we argue that he either undermines the force of his own use of coupling arguments or he begs the question against the advocate of intracranial cognition. Because Menary’s discussion of coupling and constitution is short and to the point, we can quote him at length.

A&A [Adams and Aizawa] (2001) claim that the causal coupling of X to Y does not make X a part of Y. The alleged fallacy assumes something like the following picture: an external object/process X is causally coupled to a cognitive agent Y. The Otto example fits this picture: a notebook coupled to a discrete cognitive agent, whereby the notebook becomes part of the memory system of that agent because it is coupled to that agent. Cognitive integrationists [those who support Menary’s version of the extended cognition] should resist this picture. It is a residual form of internalism, because it assumes a discrete, already formed cognitive agent. And this is precisely the picture we are arguing against. If we accept the picture of a cognitive agent as implementing a discrete cognitive system, before they ever encounter an external vehicle, then we will have accepted the very picture of cognition we set out to reject. This does not fit with the aim of cognitive integration, which is to show how internal and external vehicles and processes are integrated in the completion of cognitive tasks (such as remembering the location of MOMA). (Menary 2006, p. 333)

The suggestion appears to be that we should never think of a lone human being as a discrete cognitive system. Humans are, so this line goes, always cognitive systems integrated into a network of interacting components. Humans in their mere biological being are never cognitive systems. Put more boldly, perhaps, insofar as humans are cognitive beings, they are essentially users of external vehicles.

Suppose that, just for the sake of argument, it is true that, insofar as humans are cognitive agents, they are never entirely bereft of external vehicles that they manipulate. That is, suppose that every human cognitive agent always engages some external vehicle or other in her cognitive processing. Even this concession is not adequate to circumvent the coupling-constitution fallacy. We can simply reformulate the problem to incorporate Menary’s idea. So, suppose that, simply for the sake of argument, Otto’s biological mass never in itself suffices to form a cognitive system. Otto’s cognitive being is always enmeshed in a network of tools. Still, think of “young Otto” before the onset of Alzheimer’s disease. Young Otto was embedded in one network of tools. Presumably this network of tools will not include the notebook that will one day, say, 30 years later, be manufactured in some factory and subsequently purchased by “Old Otto” who has come to suffer from Alzheimer’s Disease. That is, assume that one’s cognition does not extend into currently nonexistent tools that one will use in the future. Now consider “Old Otto” following the onset of Alzheimer’s, but prior to the purchase of the notebook. Still, the notebook lying on a store shelf never seen by Old Otto is not part of Old Otto’s cognitive apparatus. How, then, does the notebook become part of Old Otto’s cognitive apparatus on a coupling argument? One might suspect it begins with Otto’s coming to regularly use the apparatus. It begins when Old Otto begins to manipulate his notebook. But, it is right here that the coupling-constitution fallacy is committed. It is committed when one makes the move to include new cognitive processing mechanisms, such as the notebook. So, even Menary’s strong hypothesis that cognitive agents are never without their cognitive processes extending into tools is not enough to avoid the coupling-constitution fallacy.

We should emphasize the importance of the foregoing reply to Menary. The reply lets the advocate of extended cognition dictate a case in which she can assume that a cognitive process extends from an individual’s brain into that individual’s body and a set of environmental tools at time t0. Now suppose that the individual acquires a new tool at t1, a tool that the individual uses a lot, one with which the individual causally interacts in some very intense way. The point of the coupling-constitution fallacy is that causally coupling a process X to a cognitive process Y is not sufficient to make the X process into a cognitive process nor does it make the entire X-Y process into a cognitive process. So, it is not the fact that area V1 is connected to area V2 (or whatever) that makes area V1 into a cognitive processor. It is the types of information processing that V1 does on non-derived representations that makes it a cognitive processor.

Consider now Menary’s second defense of coupling arguments

For the cognitive integrationist the picture is like this: my manipulation of the notebook and my brain together constitute a process of remembering. In cases like these, the process of remembering cannot be described exclusively in terms of biological memory or solely in terms of the manipulation of external representations, because it is a hybrid process. Schematically: X is the manipulation of the notebook reciprocally coupled to Y—the brain processes—which together constitute Z, the process of remembering. Once we have this picture, it is easy to see that A&A have distorted the aim of cognitive integration. The aim is not to show that artifacts get to be part of cognition just because they are causally coupled to a pre-existing agent, but to explain why X and Y are so coordinated that they together function as Z, which causes further behavior (Menary 2006, pp. 333–334).

One thing Menary may be saying is that he does not want to make an inference from coupling to constitution. He does not want to use the fact that Otto manipulates his notebook in certain ways as evidence for the hypothesis that Otto’s use of his notebook constitutes an extension of cognitive processing. He simply wants to stipulate, or define, or hypothesize that Otto’s manipulation of his notebook constitutes an extension of cognitive processing. This reading seems to be the import of the part about “The aim is not to show that artifacts get to be part of cognition just because they are causally coupled to a pre-existing agent.” If that, however, is what Menary is about, then that is not a defense of what other extended cognition theorists have said. Menary is not coming up with a way to vindicate the coupling-constitution arguments; he is instead offering a statement of extended cognition. And, by Adams and Aizawa lights, giving up the coupling-constitution considerations as an argument for extended cognition is entirely in order. The arguments are, after all, fallacious.

Menary is, of course, free to abandon an argument, but he is also abandoning the argument that others have given. So, Alva Noë writes, “According to active externalism, the environment can drive and so partially constitute cognitive processes” (Noë 2004, p. 221). Menary is abandoning Noë’s argument. More strikingly, Menary is abandoning his earlier argument: “The real disagreement between internalists [like Adams & Aizawa] and integrationists [like Menary] is whether the manipulation of external vehicles constitutes a cognitive process. Integrationists think that they do, typically for reasons to do with the close coordination and causal interplay between internal and external processes” (Menary 2006, p 331). The second sentence seems to mean that considerations of causal coupling do provide reasons for thinking that cognition sometimes extends.

Set aside now the question of abandonment. This is not the only problem with Menary’s reply. In the last sentence of the passage cited above, Menary claims that what we are supposed to do is explain why X and Y are so coordinated that they together function as Z, which causes further behavior. In this sentence, there is an ambiguity about “functions as.” Sometimes we say that a screwdriver functions as a hammer, as when we use the handle of the screwdriver to tap in a nail. Sometimes, we say that a fork functions as a knife, as when one cuts pizza with a fork. In these cases, something functions as Z, even though it is not Z. A screwdriver functions as a hammer, even though it is not a hammer. A fork functions as a knife, even though it is not a knife. So, if Menary’s idea is that we have to explain why the combination of Otto and his notebook function as the process of remembering, even though it is not remembering, then that is no problem for the internalist. After all, the internalist view is that when Otto manipulates his notebook, he is not remembering where the MOMA is. The notebook enables Otto, along with his remaining cognitive faculties, to compensate for the fact that he does not remember.

But, maybe this “functions as” is not meant in this way. Perhaps that is simply an uncharitable reading. Suppose, instead, that what Menary has in mind is that what we are supposed to do is explain why the manipulation of the notebook (X) and the brain processes (Y) are so coordinated that they together constitute the process of remembering (Z) which causes further behavior. But wait! The internalist is not going to accept the obligation to explain this, because the internalist rejects the idea that the manipulation of the notebook and the processes are so coordinated as to constitute the process of remembering. According to the internalist, the only remembering that is going on is whatever happens to be in Otto’s head. Demanding that the internalist explain why Otto’s use of his notebook constitutes remembering is a question-begging demand.

A principled internalist view of the mind and how to make a brain-tool science

The introduction to this paper noted some virtues of cognitivism. We can now return to one of these virtues; the ability to delimit a proper subject for a science of cognition. To do this, recall some of Clark and Chalmers’s comments on Inga and Otto. They assert that

For in relevant respects the cases are entirely analogous: the notebook plays for Otto the same role that memory plays for Inga. The information in the notebook functions just like the information constituting an ordinary non-occurrent belief; it just happens that this information lies beyond the skin (Clark and Chalmers 1998, p. 13)

Certainly, insofar as beliefs and desires are characterized by their explanatory roles, Otto’s and Inga’s cases seem to be on a par: the essential causal dynamics of the two cases mirror each other precisely. (ibid).

In all important respects, Otto’s case is similar to a standard case of (non-occurrent) belief. The differences between Otto’s case and Inga’s case are striking, but they are superficial. … To provide substantial resistance, an opponent has to show that Otto’s and Inga’s cases differ in some important and relevant respect. But in what respect are the cases different? (ibid., pp. 14–15).

To provide the “substantial resistance” Clark and Chalmers were after, Adams and Aizawa (2001), and Rupert (2004), drew attention to various features of normal human memory that psychologists had discovered over the last century and more. Plausible differences uncovered by scientific research in cognitive psychology would seem to count as important and relevant respects in which Inga and Otto differed. This is where cognitivism helps us draw a principled distinction between the intracranial and the transcranial.

Among the replies Clark and Menary adopt is to make a virtue out of necessity.24 Rather than supposing that functional similarities make the case for extended cognition, they want to claim that functional dissimilarities and complementarity of function make the case.25 Here is what they have to say in their own words. Wilson and Clark, write,

no part of the arguments for extended cognition turn[s] on, or otherwise require[s], the similarity of the inner and outer contributions. This point also deflects a related concern that Adams and Aizawa express. They say that since the causal arrangements whereby external stuff contributes to action seem very different from those in place when internal stuff does so, there can be no unified science of the extended mind. Thus, they note (Adams and Aizawa 2001, p.61), that biological memory systems display a number of effects (such as recency effects, priming etc.) that are not currently features of external modes of storage, such as Otto’s notebook. True enough. Such differences, however, in no way compromise the case for extended cognition. For that case depends not on fine-grained functional identity but upon the deep complementarity of inner and outer contributions whose joint effect (e.g., effective remembering) seems apt for the solution of a cognitive task, intuitively identified (Wilson and Clark 2009, pp. 24–25).

The first sentence of this passage seems to be flatly contradicted by the passages from Clark and Chalmers (1998). But, perhaps it is true that the differences found in the cognitive psychological literature to which Adams and Aizawa (2001) and Rupert (2004) draw attention are irrelevant. Maybe what really matters is complementarity. Here is how Menary puts it,

It is important to cognitive integration that external manipulation do something different to brain processes … Otto’s use of his notebook is cognitive because he manipulates the vehicles (sentences) in his notebook to complete a cognitive task. … Otto’s manipulation of external vehicles is not cognitive because it is similar to Inga’s biological memory, but because Otto and his notebook constitute “an integrated system for holding and manipulating information during the performance of complex cognitive tasks” (Baddeley 2000, p. 78).26

Menary’s third sentence here apparently rejects the claims we cited from Clark and Chalmers (1998). Menary, far from denying the differences between Inga and Otto, energetically embraces them. Now these important and relevant differences are supposed to be part of what makes Otto’s use of his notebook cognitive. Whatever processes conspire to accomplish a cognitive task are supposed to be cognitive processes.

The discovery of complementarity is a peculiar basis upon which to argue for extended cognition. Why suppose that if a process of type A and a process of type B conspire to bring about some effect, then, really both processes are of the same type, either A or B? Perhaps there are some cases in which this kind of reasoning delivers a plausible result, but there are clearly cases where it does not work. Consider distillations in which two liquids with different boiling points might be separated. (See Fig. 3.) Consider n-octane which has a boiling point of around 126°C and n-decane which has a boiling point of around 174°C. By applying heat to a mixture of n-octane and n-decane, one can raise the temperature of the mixture to the boiling point of n-octane, so that it evaporates, while the n-decane remains in a liquid state. The n-octane vapor can then proceed up a column and into a condenser, while a coolant will reduce the temperature of the vapor below the boiling point of n-octane, thereby causing it to condense. Upon its condensation, the liquid n-octane can drain down an inclined column into a collection flask. In a distillation, we have complementary types of processes, evaporation and condensation, which work together to achieve a desire effect, an effect that neither process alone could produce. But, in distillation, we do not have a case in which the process of evaporation extends into the second process of condensation, or vice versa. Nor does distillation provide us with an instance of the process of evaporation becoming an instance of the process of condensation or vice versa.
Fig. 3

Illustration of a simple distillation apparatus: a a heat source, b a flask containing a mixture of n-octane and n-decane, c a condenser with coolant ports, and d a flask of distilled n-octane

Yet, Clark and Menary might see some hope for their view in this example of distillation. The process of distillation, they could rightly say, is simply a combination of a process of evaporation—which is not, itself, the process of distillation—and a process of condensation—which is also not the process of distillation. By analogy, they might argue that cognitive processes are made up of combinations of other non-cognitive processes, non-cognitive processes in the brain, non-cognitive processes in the body, and non-cognitive processes in the extra-organismic environment. The complementary actions of such processes are exactly what extended cognition is all about. And surely, they could continue, it is not at all unusual to find scientifically valid types of processes that are made up of disparate types of subprocesses. Think of the process of mitosis as being made up of the subprocesses of prophase, metaphase, anaphase, and telophase. The process of human digestion might be broken up into processes of mastication, swallowing, acidification, emulsification, and absorption. Many biochemical processes, such as glycolysis, involve many distinct subprocesses, such as phosphorylation, dehydration, and isomerization. So, again, we should not be surprised to find that cognitive processes are like other scientific processes.

There are numerous reasons to abandon this hope. In the first place, it could well be that Clark and Menary’s view is that cognition is a composite process made up of non-cognitive subprocesses just as distillation is a process made up of non-distillation subprocesses. But, to have that as a view does not amount to having a case for the view. Contrary to what Clark and Wilson and Menary propose, complementarity does not give one an argument for extended cognition; it merely provides a way to state the idea. In the passage from Wilson and Clark cited above, complementarity is supposed to ground an argument for the hypothesis of extended cognition, not merely constitute a statement of it. So, the analogy does not deliver an argument. We still need some reason to think that “cognitive processing” refers to processes in the whole brain-body-world ensemble, rather than to processes in the brain. Cognitivism might provide a basis for doing this, but Clark and Menary have shied away from cognitivism.

In the second place, and more seriously, the complementarity in the distillation example is still disanalogous to the complementarity in the extended cognition cases. Think of Otto suffering from Alzheimer’s disease before he takes up his use of a notebook and after he takes up the notebook. Prior to his use, his cognitive processes do not extend into the notebook. After he takes up the notebook, his cognitive processes are supposed to extend into the notebook. But, the analogy between cognition and distillation breaks down here. Prior to Otto’s complementation with the notebook, the processes in Otto’s lone brain might be likened to evaporation in the distillation case. But, keeping to the analogy, after complementation, we do not get evaporation extending into the condenser or the whole of the distillation apparatus, and so we should not suppose that we get cognition extending into Otto’s notebook. This is just to reiterate the very problem that the distillation example was meant to illuminate.

In the third place, where the process of distillation is only made up of the same two subprocesses, evaporation and condensation, Clark and Menary evidently suppose that cognitive processes could be made of any number of distinct kinds of complementary subprocesses. People can complement their relatively fixed cognitive apparatus with sun dials, analog watches, digital watches, hour glasses, warning lights, warning buzzers, sticky notes, PDAs, calculators, pocket calculators, personal computers, an abacus, a telescope, a microscope, magnifying glass, contact lens, eye glasses, sun glasses, hearing aids, cochlear implants, tape recorders, cameras, video tape recorders, Flash movies, animated gifs, rolodexes, index cards, mathematical equations, Fitch style deduction notation, musical notation, libraries, maps, and on and on. This is the motley of things to which we were referring in Adams and Aizawa (2001), the motley that Menary (2006), thinks he can avoid, but does not explain how.

In the final place, the advocates of extended cognition have no satisfactory theory of what cognition is that corresponds to the theory of what distillation is.27 Rather than embrace cognitivism as offering a theory of the cognitive, both Clark and Menary suggest that cognitive processes are what enable one to complete cognitive tasks. Familiar sorts of counterexamples reveal the inadequacy of this idea. Suppose that engaging in an entertaining conversation in an online chat room is a cognitive task. This, however, can be accomplished by so-called “chatterbots” that apply many simple rules. So, in response to the input “I am feeling…,” a chatterbot might reply, “Why are you feeling…?” In response to a particular celebrity name, such as “Halle Berry,” the chatterbot might reply, “I love Halle Berry! Don’t you?” Such chatterbots are often deployed to online chat rooms as a kind of prank or sometimes, more malevolently, to phish for personal information from visitors to the chat room. Yet, to date, among the many chatterbots that can successfully pass for human, there are many that simply do not think. Such chatterbots are simply successors to Joseph Weizenbaum’s computer program ELIZA that could pass for a Rogerian psychotherapist. Chatterbots are more practical versions of a computer program that is a mere vast lookup table of all possible conversations. Chatterbots and vast lookup tables are not cognitive agents and do no cognitive processing. So, Clark and Menary would have to say a lot more to validate the idea that whatever processes one uses to complete a cognitive task is thereby cognitive and still more to flesh out how there could be a familiar kind of science of extended cognition.

The short of the matter in this section is that, aside from providing a non-question-begging reason to think that, as a matter of contingent empirical fact, the cognitive is typically realized in the neural, cognitivism encourages right thinking in delimiting a proper subject for cognitive science. It helps us avoid falling victim to implausible complementary arguments. It encourages us to challenge the extended cognition view that claims that complementary brain, body, and environmental processes add up to a cognitive process. What would it be about these complementary processes that make them cognitive? In virtue of what do they bear the mark of the cognitive? Surely it cannot be that we should embrace some form of operationalism. Some four decades into computational theories of the mind, we should all be very familiar with the failings of operational definitions of thinking.


In this and earlier exchanges regarding the hypothesis of extended cognition, we have taken a decidedly reactionary strategy of invoking cognitivism in defense of the orthodoxy. This makes sense, since cognitivism is so central to cognitive science. It is the orthodox view of the nature of cognition. It begs no questions against extended cognition. It provides a principled basis for discriminating the cognitive from the non-cognitive. It supports right thinking regarding the delimitation of a field of cognitive science and avoiding the coupling-constitution fallacy. It resists certain challenges to even one of its boldest elements, the idea that there is non-derived content. It is a theory that has a lot going for it. It is a theory that Clark and Menary have been willing to take seriously in debates over extended cognition.

There are, however, those who are ready to dismiss cognitivism, or at least many of its components, out of hand. There are ecological psychologists, influenced by J. J. Gibson, who reject the need to hypothesize mental representations. There are also phenomenologists who reject mental representations, although for very different reasons. And there are those who are simply ready to celebrate a “post-cognitivist” cognitive science, whatever that may become. It would, indeed, be interesting to see what case one could make for or against the hypothesis of extended cognition, if one did not rely on cognitivism. We suspect that much of the argumentation would still be invalid and the hypothesis still far from secure. But, making good on such suspicions will have to await other works.


See, for example, Clark and Chalmers (1998), Haugeland (1998), and Rowlands (1999).


It is, of course, true that many advocates of extended cognition reject cognitivism. See, for example, Haugeland (1998), Thompson (2007), Wallace (2007), and Gomila and Calvo (2008). Yet, the fact that there is this disagreement does not mean that an appeal to cognitivism in making a case against extended cognition begs the question against extended cognition. One begs the question when one assumes, without argument, what one is trying to prove. But, cognitivism is not assumed without argument. The case for cognitivism lies in its success in explaining various features of cognition. Were the mere existence of different views of P sufficient to guarantee that one side or the other is begging the question, then every debate would have to involve begging the question.


For discussion of this, see Adams and Aizawa (2001, 2008, 2009) and Aizawa (2010).


This seems to be the kind of extended cognition that Hutchins (1995a, b) pursues.


For example, van Gelder (1995, 1998)


Sutton (2004) proposes “an integrated framework within which different memory-related phenomena might be understood” (p. 188). Such an integrated framework may or may not be a science.


In the review of Clark (2008); Fodor (2009) also appeals to non-derived content in challenging the hypothesis of extended cognition.


Some philosophers have been worried by our term “intrinsic representations.” How, they ask, can anything be intrinsically a representation? But, in this context, “intrinsic” means only that the representation does not get its content from any prior existing content. In other words, for us, it means just the same thing as “non-derived representations.” We do not draw any theoretical differences between “intrinsic representations” and “non-derived representations.”


This point is made in Adams and Aizawa (2008), but also in Aizawa (2010), in reply to Rowlands (2009).


Exactly the same point can be made about the example of Robert Wilson (2004) of playing a children’s board game called “Rush Hour.”


Clark (2005) lists three problems for intrinsic content. These are that it is unclear that it exists, that it is not necessarily limited to brains, and that its existence would not compromise the case for extended cognition. Since Clark does not argue for the first of these claims, and since we do not think that intrinsic content is necessarily limited to brain, we here limit ourselves to addressing Clark’s third claim beginning at Clark (2005, p. 5). For further elaboration on these three points, see Adams and Aizawa (2008).


Clark apparently assumes that the items in Fig. 2 represent the items in Fig. 1. For present purposes, we can abide by that assumption. Of course, it is also possible that Fig. 2 does not represent Fig. 1, but instead both Figs. 2 and 1 have the same content, namely, the same social conventions. We can abide by that assumption as well. Given our observations above that the derived/non-derived distinction is orthogonal to the specific content an item bears, it is fine if both figures have the same content and that one has derived content and the other non-derived content.


See Adams and Aizawa (2010).


It is unclear to us just how Menary would tell a plausible story about why we use tools, if it is not that we are more cognitively limited without them.


Indeed, Adams and Aizawa are on the books arguing that Fodor’s theory does not work. See, for example, Adams and Aizawa (1994).


For some reason, Menary seems to have presented the same dilemma twice in succession. We do not really see the difference between these two dilemmas.


Of course, one does not always speak one’s mind. One might say that “Bill Clinton was the greatest president of the twentieth century” because one thinks that Bill Clinton was the greatest president of the twentieth century or because one does not think this, but is being sarcastic. Nevertheless, there could often enough be enough usefulness in having a close matching between what one thinks and what one says to sustain a match.


As an aside, some advocates of extended cognition have embraced this idea and have employed it for yet another argument, a “complementarity argument,” for extended cognition. To put it crudely, the brain, body, and environment carry out complementary processes, so all of them are cognitive. Aside from the prima facie implausibility of this kind of argument, there is a prima facie tension between the coupling-constitution arguments that try to argue for the sameness of processes, where the complementarity arguments are predicated on the differences among processes.


Although we have encountered this argument in discussion with Susan Hurley, Tony Chemero, and Michael Silberstein, something like this line is found in Rockwell (2005) and Hurley (2010). Block (2005) notes that Alva Noë is inclined to say this kind of thing as well.


See Adams and Aizawa (2008) Chapter 7.


See Clark (2008, 2010). For simplicity of exposition, we will eliminate the portions of the thought experiment that are directed to Robert Rupert’s view.


It is, of course, an oversimplification to say that the whole of the brain realizes cognitive processes, since there may well be glial cells or blood vessels or other such structures that do not, but it is an oversimplification we tolerate merely for the sake of simplifying the exposition.


The distinction we are here drawing between hippocampal processing and extra-hippocampal processing is, in important respects, analogous to the distinct Clark recognizes between vision for action and vision for perceptual experience. See Clark (2008), Chapter 8. Hippocampal and extra-hippocampal processing will differ at the psychological level as do visual processing for action and visual processing for perceptual experience. Moreover, they will be localized in different parts of the brain, just as visual processing for action is localized in the dorsal stream, where visual processing for perceptual experience is localized in the ventral stream.


In addition to the complementarity reply, there is also the suggestion that we should not look at fine-grained similarities between Inga and Otto; we should instead look at coarse-grained similarities. There is also the suggestion that we should not look for scientific regularities, but instead “common sense” regularities. See Clark (2008), p. 88. These will not be addressed here merely for lack of space.


Sutton (2010) explores this position in terms of a move from “first wave” extended mind to “second wave.”


Like Menary, Wilson and Clark (2009), p. 13, also relate complementarity and the idea of accomplishing cognitive task: “In most cases where we are tempted to speak of cognitive augmentation, the same rule of thumb seems to apply: we find cognitive augmentation where new resources help accomplish a recognizable cognitive task in an intuitively appropriate manner, e.g., by enabling the faster or more reliable processing of information required by some goal or project.”


For further discussion of this, see Adams and Aizawa (2008) chapter 5.


Copyright information

© Springer Science+Business Media B.V. 2010