Extended Cognition (Part II)

by Carl Pierer

TetrisAfter having presented Clark and Chalmers' extended cognition hypothesis as well as two lines of argument against the hypothesis, the last article at this place ended with an intuitive, bad gut-feeling and a promise to develop this feeling into a full blown argument. Before making good on that promise, this article will start with a brief recap of the arguments presented so far.


Clark and Chalmers' argue in their famous “the extended mind” paper that when a person uses tools or the environment to facilitate a particular cognitive process, this person and her tool constitute a coupled system. Indeed, Clark and Chalmers suggest that in such a coupled system the cognition extends, i.e. it is not confined to the brain/skull-boundary. The argument works as follows: suppose the cognitive process in question is to decide whether a certain shape that appears on the screen will fit into a given slot (as in the classic Tetris game). The person can use a computer to rotate the shape and decide whether it will fit or not. Now, this is clearly an external process. But imagine that in the not so far future, a person will have a neural implant with exactly the same functional structure as the computer and she can use the implant to rotate the shape and check whether it will fit (or she can use the traditional method of rotating it mentally). Clark and Chalmers think that as there is no difference between the computer and the neural implant. Further, whether the person in the near future choses the implant or the traditional method does not matter for the process to count as cognitive. Therefore, the only thing that distinguishes the computer-scenario from the neural-implant-one is that the former involves the use of a tool external to the brain/skull-boundary. But since precisely this is at question, this difference cannot be invoked to support the claim that using the computer is non-cognitive. Thus, using the computer is cognitive and so cognition extends.

Clark and Chalmers' argument relies on the parity principle:

If, as we confront some task, a part of the world functions as a process which, were it done in the head, we would have no hesitation in recognizing as part of the cognitive process, then that part of the world is (so we claim) part of the cognitive process.

This seems to follow directly from the basic functionalist idea that what it takes for a process to count as cognitive is its functional structure, rather than its physical instantiation.

In the previous article, two lines of argument against this view were presented. The first is taken by Adams and Aizawa. They suggest that any process that is to count as a cognitive process has to bear the “mark of the cognitive”. They think that it is not theoretically impossible for cognition to extend, but as a contingent matter of fact there is no process involving the external world that bears the mark of the cognitive. It was mentioned in passing that their suggested “mark” is closely modelled on human cognition. The second line is taken by Sprevak, who argues that the hypothesis of extended cognition provides a counterargument to the view from which it is derived, i.e. functionalism. He attacks Adams and Aizawa's argument on the grounds that their “mark of the cognitive” is too closely modelled on human cognition and deny processes to be cognitive on the grounds of being instantiated differently – a violation of the basic functionalist idea. At the same time, he suggests that functionalism entails extended cognition and further that a moderate (Clark and Chalmers') version of extended cognition is impossible. Instead, if functionalism is accepted the conclusion that any process is cognitive follows.


Richard Menary (2010), arguing in favour of extended cognition, proposes to take a closer look at what Clark and Chalmers call a coupled system. According to them, whenever an external tool is used in such a way that by their initial (Tetris) argument it would count as part of the cognitive process, then the tool and the person using it constitute a coupled system:

All the components in the system play an active causal role, and they jointly govern behaviour in the same sort of way that cognition usually does. If we remove the external component the system's behavioural competence will drop, just as it would if we removed part of its brain. Our thesis is that this sort of coupled process counts equally well as a cognitive process, whether or not it is wholly in the head. (Clark & Chalmers 1998, p. 8)

Menary argues further that there are two possible interpretations of “causal coupling” and that proponents and opponents of extended cognition differ on their interpretation.

On the first, external components have an asymmetric influence over internal processes. For instance, using the computer to decide if the shape will fit. Both the mental and the processes in the computer “play an active causal role, and they jointly govern behaviour”. Furthermore, if the computer was taken away, the competence of the system would drop. It would be more difficult to decide if the shape will fit and take more time. However, there is no need to say that because this is the case, the computer is part of the cognitive process. It would be fallacious to base this claim on the causal connection

On the second, “the inner and outer features have a mutually constraining causal influence on one another that unfolds over time” (Menary 2010, p. 4). It is not just that the external part provides inputs that are then processed on a higher level by internal processes. Internal and external have an equal standing in governing behaviour. For instance, when using the computer to rotate the shape, the result of the digital rotation will inform the decision whether the shape fits just as a mental rotation would. (cf. Menary 2010)

Adams and Aizawa, Menary writes, work with an asymmetric understanding of causal coupling, whereas Clark and Chalmers understand the influence to be symmetric. This essay will suggest reasons to think that Sprevak's development of the extended cognition hypothesis into a counterargument to functionalism can work only on the first, the asymmetric interpretation.

Sprevak argues that the idea of extended cognition is derived from two basic principles of functionalism:

(F1) If an organism counts as sufficiently like us on a coarse-grained global functional comparison, then it is a cognitive agent.

(F2) If a cognitive agent contains a representation-manipulating process that is significant for guiding its action (in appropriate ways) when employed, then that process is one of its cognitive processes. (Sprevak 2009, p. 521)

(F2) is related to the parity principle. (F1) is at the heart of the idea that the actual physical instantiation of a cognitive process does not matter to its status as cognitive. These two principles jointly entail that any process will count as cognitive, Sprevak thinks. The argument runs as follows: Take a task X. Person A accomplishes X by using the (external) instrument Y, call this process Z. Now imagine a Martian B, who is “sufficiently like us on a coarse-grained global functional comparison”, then by (F1), B is a cognitive agent. Further, B's brain (or part thereof) has the same structure as the instrument Y. B uses her brain (or the part that is like Y) to accomplish the task X. This is then a cognitive process by (F2). Thus, by the parity principle, A's use of Y has to count as cognitive as well – because the only difference between A's use of Y and B's use of Y is that in the former case Y is external while it is internal in the latter. Moreover, since X and Y are arbitrary variables, it is possible to use this argument to establish that any process Z is cognitive.

If Sprevak's argument was sound, this would be a very fine problem for functionalism. Hardly anybody would like to assert that any process whatsoever is cognitive. Of course, it could be argued that if B's brain has the same structure as an instrument Y, then B might not be sufficiently like us to count as a cognitive agent. This, however, runs the risk of modelling cognition too closely on human cognition and to violate basic functionalist intuitions; namely, that it shouldn't matter how a process is physically instantiated. Yet, even if it is granted that B is a cognitive agent, i.e. that (F1) obtains, there is a different, more problematic issue with (F2).

(F2) requires “a representation-manipulating process” to be significant for guiding the agent's action in order for that process to count as cognitive. The terminology here is a bit unfortunate as it is not quite clear what Sprevak means: either the “representation-manipulating process” is a cognitive process in itself or it is only a component of the cognitive process. If he means the former, then his argument is attacking a straw man. Clark and Chalmers (in the Tetris scenario) do not claim that the computer's digital rotation of the shape is a cognitive process, they argue that it is part of the larger cognitive process of determining whether the shape will fit.

To see why the second meaning is problematic for his argument, it is much clearer to use Clark and Chalmers' term of a component of the cognitive process for what Sprevak calls “representation-manipulating process”. Then, it becomes evident that Sprevak's requirement for what it takes for that component to be part of the cognitive process is different from Clark and Chalmers'. Sprevak requires it to be significant in guiding the agent's action (in appropriate ways), whereas Clark and Chalmers have the stronger constraint that, moreover, it ought to have an active causal influence and that a removal of the component would result in a deterioration of the agent's performance.

It seems likely that Sprevak works with what Menary calls the asymmetric interpretation of causal influence, whereas Clark and Chalmers have a symmetric interpretation in mind. If this is the case, then his argument is subject to the same discussion as Adams and Aizawa's (see, for instance, Menary 2010). To be as favourable as possible to Sprevak's argument, it can be assumed, however, that he works with the same interpretation as Clark and Chalmers.

On this view, the component needs to have a symmetric causal influence on the agent for it to count as part of the coupled system and thus as part of the cognitive process. This restriction severely limits the generality of Sprevak's argument. For now, we can no longer substitute just any significantly action guiding use of an external instrument Y in the formulaic argument outlined above. Instead, the instrument Y has to have symmetric causal influence. This means its inputs to the internal parts of the process have to have the same standing as the internal proceedings. The argument has become toothless: it does no longer work to show that – on accepting extended cognition – any process will count as cognitive. Rather, the class of possibly cognitive processes will be limited to the ones involving a coupled system where the parts have a symmetric causal influence on each other. This is precisely the result Clark and Chalmers would like to get at.


In conclusion, the move to turn extended cognition into a counterargument against functionalism appeared very threatening at first. However, there seem to be reasons to believe that the success of this move is based on a certain vagueness of Sprevak's term “representation-manipulating process”. The aim of this essay was to cast doubt that any possible interpretation of this term will deliver the desired support for this move. If this was successful, then the hypothesis of extended cognition should still be an appealing one that is compatible with and entailed by functionalism.


Adams, F., & Aizawa, K. (2001). The Bounds of Cognition. Philosophical Psychology, 43-64.

Clark, A., & Chalmers, D. (Jan. 1998). The Extended Mind. Analysis, 7-19.

Menary, R. (2010). Introduction: The Extended Mind in Focus. In R. Menary, The Extended Mind (pp. 1-22). MIT Press Scholarship Online. doi:10.7551/mitpress/9780262014038.003.0001

Sprevak, M. (2009). Extended Cognition and Functionalism. Journal of Philosophy, 503-527.