The purpose of this post is to give a brief overview over the complexity of the discussion surrounding the hypothesis of extended cognition (HEC) and functionalism. I give some key definitions, outline some motives from multiple realisability (MR) and the martian intuition (MI) and then rehearse the problems with these concepts in relation to functionalism and the HEC. To make it a bit my own, I also indulge myself by proposing to give up a strong version of the MI in order to stick to a notion of MR that is only applicable to actual world creatures. This is meant to propagate some idea of a posteriori functionalism that could prevent falling into the trap of a radical HEC.
The idea behind MR, i.e. that mental states can be realised in a variety of ways, for instance pain as C-fibre stimulation as well as equivalent Octopus realisers, and hence functionalism is that they free us from bio- chauvinism about the mind (Wheeler, 2010: 88). Functionalism is a metaphysical concept of cognition that assigns particular mental states to functional roles they serve and hence what effect they bring about and what part they take in the more holistic cognitive system. (Levin, 2008; Braddon-Mitchell & Jackson, 2007; Shapiro, 2008: 6). The number of citations here already indicates that there is much disagreement about what exactly the intricacies and implications of this more general definition of all versions are.
One of the possible implications is the HEC (Clark, 2008: 88). This hypothesis suggests that cognitive processes are not confined to the biological system of an agent, but can extend to and incorporate the environment. For instance an iPhone can under certain conditions become part of someone’s cognitive process by serving the its functional role (Clark, 2008). This is not undisputed: Sprevak (2009) insists that functionalism logically entails the HEC, Wheeler agrees calling it a “footnote to Putnam” (Wheeler, 2010: 8), while Shapiro (2008) and Rupert (2004: 34) hold that there is no clear connection between functionalism and the HEC if not even a mutually exclusive relationship. In essence the discussion is “muddled” (Walter, 2010: 87), but for the purpose of this essay I shall follow Sprevak’s line of reasoning.
Which brings me to a last definition important to that line: The parity principle (PP), which assists the HEC and especially Clark by claiming that the cognitive-ness of internal and external processes need to be judged equally by their relevant functional similarities and not on the basis of their location (Clark & Chalmers, 1998: 8; Walter, 2010: 86; Sprevak, 2009: 505; Wilson, 2004: 195). This principle now assumes functionalism (Walter, 2010: 96).
Sprevak (2009) abstracts a problem from this principle: If we were to imagine a martian who uses a similar process to Otto’s notebook then according to the PP we are to grant it the label ‘cognitive’. This in turn means that Otto’s notebook on earth also deserves this label. However, one can lead this argument ad absurdum by pointing out that you can put increasingly odd things into a martians head until the most bizarre processes count as cognitive. This according to Sprevak leads to a radical HEC where even going online would extend my memory into the sum of human knowledge. Therefore functionalism and the HEC are false.
Some have argued that this strong MI is absurd itself: Just because of the “in-the-head-ness” of a process found a Martian we cannot imply that it is cognitive. In fact, to do so violates the PP that Sprevak appeals to (Wheeler, 2010: 19-20)! In essence, you cannot just put things in some Martian’s head and call them cognitive. This is however not how Sprevak goes about arguing against the HEC. First, he establishes that martian processes are cognitive and then through the PP argues that radical HEC processes must be too (Walter, 2010: 298). Again, for the purpose of this essay, I will follow Sprevak’s line of reasoning since his construction of the argument already allows the ordinary version of the MI to bring about a radical HEC.
In order to resolve this, if we accept both the MI as well as the HEC, the next step must be finding a more refined definition of functionalism, that 1) defines functional roles finely enough not to allow radical martians / the radical HEC, and 2) coarse enough not to circumcise MR, i.e. to still allow for the HEC, mental states in animals or even modest Martians. In other words, the problem is one of grain. Define it too finely and martians as well as beings on an appropriate “cognitive bar” are excluded (Clark, 2008: 89, 93), define it too coarsely, and the radical HEC follows (Sprevak, 2009). The struggle between these two positions mirrors the positions of Clark against Adams & Aizawa (A&A).
A&A famously argue for fine grained definitions by stating that the external processes that Clark advocates need to exhibit a ‘mark of the cognitive’ that we are used to within contemporary cognitive sciences (A&A, 2001: 48; Aizawa, 2005). In other words, Clark’s HEC is not cognitive because its processes are incommensurable with what we know about cognition, for instance because Otto’s notebook does not show negative transfer (A&A, 2008; Rupert, 2004: 413). The afore mentioned relevant functional similarities between external and internal processes are not what we ought to look at.
What do we look at then? A&A have an answer to this question. They bring forward different definitions of cognition. After all, it is cognition that is supposed to be extended, so Clark better come up with a good notion of it further than “whatever contributes to intelligent behaviour” (A&A, 2008: 73, 85). This would lead to a dispute between Clark and A&A over intrinsic content that this essay will not go into.
Instead, I propose picking just one argument out of this discussion to reexamine the MI and PP: In response to Clark’s complaint that there is no reason to exclude external structures from having intrinsic content (Clark, 2005: 1), Aizawa (2005: 4) states that he does not believe that this is unintelligible, but simply as “matter of contingent fact” not happening. I find this response very intriguing and would like to examine it more closely as I believe it shifts the discussion towards more empirical considerations.
Suppose in the actual world, cognition required some neuronal mark the way A&A support and radical martians are not possible. MR would take the form of a concept that applies only to actual organisms. Would it still make sense to support the original MI in this case? After all, our motivation for MR and functionalism was to allow for instance animals to have mental states, why include impossible martians? The immediate problem here of course is epistemic, i.e. we cannot know what sort of martians are possible, but A&A maintain that it is still “perfectly reasonable” (Aizawa, 2005: 6) to exclude them under an empirical theory of cognition that would form only around actual creatures. Indeed, supporters of coarse-grained functionalism often base their arguments around the premise that tomorrow we could find a martian with some wibbelly- wobbelly-brain organised in a way that allows functionally similar cognitive processes (e.g. Clark, 2005: 5; Wheeler, 2010: 4), yet A&A note that although this is metaphysically possible, it is sensible to “conjecture that only neuronal processes are in fact so organised” (A&A, 2008: 69).
What A&A are advertising here is the idea of empirical or a posteriori functionalism as opposed to the common-sense functionalism that Clark employs to support the HEC (Andersen, 2007: 3), and they are not alone in demanding a more “scientifically informed account” of what it means to be a part in a cognitive system (Wheeler, 2010: ch. 3, 13), since Clark only develops a “rule of thumb” to defeat bio-chauvinists (A&A, 2008: 77). I disagree here with A&A since to me it seems not at all obvious that the HEC is scientifically uninformed, in fact, Wilson even suggests that today the HEC and common-sense functionalism are a necessary part of many psychological explanations (Wilson, 2004: 212).
A&A with this figuratively say ‘show me a Martian and I will reconsider, for now let us stick to what actual, observable and neuronal creatures have in common to abstract an appropriate level of grain’. Yet, I believe the question what is observable forms the strongest counterexample to A&A’s argument: Doolittle (1994) points out that there is a great deal of functional convergence in evolution, i.e. different incommensurable structures evolving to entertain the same functionalism role as for example enzymes in different species. Even more problematic I find evolutionary psychological variations and unusual realisers that occur among species on a similar level of cognitive sophistication (Pinker, 1997). These variations that A&A may insist need to be included might be enough to force grain to be so coarse to allow for the HEC and even some Martians (Sprevak, 2009: 512-13).
Moffett (2010) would insist that this leads to the conclusion that A&A’s argument towards a posteriori functionalism is false, however, I believe their are on to something: If it were possible to abstract a posteriori what all cognitive systems have in common, then there would be a framework for what martians where possible and perhaps a more holistic cognitive science with room for the HEC (Rupert, 2008: 13), which brings us back to the problem of grain and how to organise such a science in light of the immense difficulty of finding adequate and filtered functional definitions (Jackson & Petitt, 1993; Moffett, 2010: 84). However, his endeavour shall be left for another post.