Computationalism and the Extended Mind

May 2008

In The Extended Mind, Andy Clark and David Chalmers discuss the possibility of extended minds; coupled systems in which a human organism produces cognition, beliefs, and thoughts with the aid of external entities. Clark and Chalmers give the example of Otto, a man suffering from Alzheimer’s who uses a notebook to record his thoughts and beliefs. The authors argue that our notion of belief ought to be adjusted so that Otto and his notebook constitute a cognitive system possessing beliefs that are similar in all important ways to normal beliefs. In Why There Still Has to Be a Language of Thought, Jerry Fodor argues that our thoughts have a constitutive, evaluable structure analogous to the structure of language and to the relations among entities in the world. I offer three objections to Clark and Chalmers’s concept of the extended mind, ultimately arguing that extended mind semantics allows us to make counterproductive and counterintuitive claims about minds, beliefs, and more. I will show the similarities between Clark and Chalmers’s claims about language’s role of relating the mind to entities in the world, and Fodor’s ideas of mental representations and computationalism. Although I find important differences between Clark, Chalmers, and Fodor, I will offer a plausible account of extended minds that is friendly to the representational and evaluational aspects of Fodor’s computationalism. I ultimately conclude that although extended mind semantics is sometimes useful, the line between the mind and the rest of the world should remain in the skull for the time being.

Clark and Chalmers’s argument for the extended mind begins with the notion of an epistemic action. An epistemic action is an action that alters the world in way that is a cognitively beneficial for an agent (8). The authors suggest that the brain has evolved in “the reliable presence of a manipulable external environment” (11), and that humans have naturally adapted to manipulate their environment to reduce cognitive load. One of the most salient examples of epistemic action given is of the philosopher who thinks best by writing her thoughts down on paper (11). By writing down her thoughts, the philosopher manipulates her environment through epistemic action. In turn, the cognitive process of examining her thoughts is aided by an external entity—pen and paper.

Clark and Chalmers call a process in which an agent utilizes an external entity a coupled process, and state that “this sort of coupled process counts equally well as a cognitive process, whether or not it is wholly in the head” (8). In the case of the writing philosopher, the human and the pen and paper are a coupled process because they “jointly govern behavior in the same sort of way that cognition usually does” (8). If the pen and paper are removed, the philosopher will suffer the same decrease in “behavioral competence…as [she] would if we removed part of [her] brain” (8), so we are urged to agree that cognitive processes involving external entities are on the same footing as wholly internal cognitive processes occurring in the agent’s brain.

Next, the authors wish to show that not only does some cognitive processing occur in the environment, but that “beliefs can be constituted partly by features of the environment,” (12) and if this is the case, then “mind extends into the world” (12). Clark and Chalmers give the example of Otto, an Alzheimer’s sufferer who must use a notebook to remember (among other things) that a particular museum is located on 53rd street (12). Otto wants to go to the museum, so he consults his notebook to “remember” the address and goes to 53rd street. The authors claim that even before Otto consults his notebook, Otto has the belief that the museum is on 53rd street because the notebook plays the same role for Otto that memory plays for other people, except in poor Otto’s case, the information “lies beyond the skin” (13). The authors also claim that we ought to say that Otto has the belief before consulting his notebook because any alternate explanation involving intermediate beliefs about the contents of the notebook introduces “one step too many, [adding pointless complexity]” (13) to our descriptions of Otto’s mental life. Clark and Chalmers conclude that “the notion of belief ought to be used so that Otto qualifies as having the belief in question. In all important respects, Otto’s case is similar to a standard case…of belief” (14); therefore, Otto’s beliefs and thus his mind are extended to his environment.

I offer three objections to the argument that Otto’s belief that the museum is on 53rd street before consulting his notebook is cognitively “on par” with belief as traditionally conceived. First, simply ask Otto if he believes the museum is located on 53rd street and he will likely respond that he has no belief about where the museum is; he will *not say that his notebook believes the museum is on 53rd street, because then he would not need the notebook! How can we convince Otto that he has a belief that he believes he does not have? And if Otto is not convinced that he has the *belief, why should we be convinced that he has it? To go further and say that Otto plus the notebook form a “coupled system” that has a belief reminds us of the systems objection to Searle’s Chinese Room argument and the absurdities that it entailed.

My second objection is that complex explanations of Otto’s beliefs involving intermediate beliefs about the notebook do not introduce unnecessary steps because Otto is not a properly functioning human being. As an Alzheimer’s sufferer, it is reasonable that explanations of the nature of his beliefs require extra steps, so we should not change our collective notion of belief just to simplify Otto’s case.

My third objection is that Clark and Chalmers’s notions of extended minds have some consequences that we would not want to admit into our philosophies of mind and epistemology. The authors offer three central criteria for ascriptions of extended belief in Otto’s case: the notebook is a constant in Otto’s life, the information in the notebook is directly available without obstacle, and upon retrieving the information, Otto automatically endorses it (17). The authors mention a fourth criterion—that the notebook’s information is a consequence of an earlier conscious endorsement by Otto—but they de-emphasize this criterion because Clark and Chalmers are focused on “active coupling” (11). Suppose we replace Otto’s notebook with a telephone book. A telephone book could easily satisfy the criteria of constancy, accessibility, and authority, yet we would not want to say that for every person listed in the book, Otto *knows *that person’s phone number.

To the first objection, Clark and Chalmers would likely respond that Otto would report his beliefs differently if he were allowed to consult his notebook, and that Otto should be allowed to consult his notebook just as normal people are allowed to consult their memories. This still places abnormal constraints on Otto’s beliefs that make the Otto-notebook system appreciably deficient. Clark and Chalmers would respond to my second objection by offering other examples of coupled systems that do not involve corner cases, such as neural implants of the future. My third objection stands as a general problem for the authors—I think extended mind semantics may be useful in some situations, but it allows us to say too many things about minds and persons that we do not want to allow.

Clark and Chalmers offer some interesting thoughts on the emergence of language as a sort of mental glue used to construct extended cognitive systems. These ideas have great import for the Language of Thought hypothesis and the brand of computationalism offered by Jerry Fodor in Why There Still Has to Be a Language of Thought. It seems that Clark, Chalmers, and Fodor have compatible explanations for the semantic structure of language. Clark and Chalmers suggest that,

Language…appears to be a central means by which cognitive processes are extended into the world. Think of a group of people brainstorming around a table, or a philosopher who thinks best by writing, developing her ideas as she goes. It may be that language evolved, in part, to enable such extensions of our cognitive resources within actively coupled systems. (11)

In addition to the evolutionary view of language, Clark and Chalmers also suggest that language serves as a “computational artifact” (12) that allows an individual brain to develop “in a way that complements the external structures [in the world]” (12). Compare these statements to the Language of Thought hypothesis, which holds that mental representations have a constituent structure that mirrors the structure and complexity of entities in the world (Fodor 136). Fodor suggests that mental representations are, for lack of a better analogy, to be thought of as abstract syntax trees or “parse trees” (114), and that we transition between mental states by performing operations on these representations (145). This notion of thinking as performing operations on mental structures is, in essence, computationalism.

I find a couple of potential difficulties in reconciling Clark and Chalmers’s extended mind with Fodor’s ideas of mental representation and computationalism: if the mind is extended, are our mental representations extended as well? How do we account for our structured mental representations and the operations we perform on them being extended into the world? Returning to the example of the philosopher who writes down her thoughts, do we suppose that her mental representations are somehow partially contained in the pen and paper? If this is the case, where is the locus of her thought—in her brain or somewhere outside her skull? Only the philosopher’s brain has the “hardware” upon which her thoughts are realized, while the pen has only ink and plastic, and the paper only wood fibers, so active thinking does not seem to be extended. The pen and paper contain representations, but not mental representations. Under Fodor’s computationalism, our minds operate on only mental representations—not representations in ink—to produce thought, so the representations of the pen and paper must be translated into mental representations before our minds can do anything with them.

It is obvious that pen and paper are cognitive aids, and in many cases, speaking of these aids as extensions of our minds is helpful and sounds natural; however, when we scratch the surface and wonder how thoughts are structured and evaluated if the mind is extended to external entities, we quickly find that the notion of the extended mind is no longer helpful. Furthermore, even if we doubt the Language of Thought hypothesis or computationalism, the notion that our thoughts occur, persist, or are active in features of our environment seems like much more than we need to explain the use of cognitive aids like pen and paper.

Despite these difficulties, there is no reason to believe that Clark and Chalmers would not agree with Fodor about the computational aspects of the “non-extended” mind, but we still need a convincing story about how metal representations and computationalist thought occur in an extended mind. I offer a story that I hope Clark, Chalmers, and Fodor will agree is plausible. Fodor wants mental representations to correspond to the world in a way that mirrors the semantic structure of language (Fodor 135), and Clark and Chalmers want language to facilitate the extension of the mind to the environment. In an extended, computational mind, we might suppose that the mind keeps reference tokens that delegate cognition to external entities; this is not to say that the external entities cognize, but just that the mind cognizes less by relying on the environment more. The philosopher who writes down her thoughts is taking immensely large and unwieldy mental representations and committing more manageable sub-representations to paper, thus making it easier to think about the problem at hand. We could say that the philosopher adds reference tokens to her mental representations, which are evaluated by consulting the paper and reconstructing mental representations from the ideas recorded there. This way, Clark and Chalmers can posit coupled cognitive systems while remaining compatible with Fodor’s mental representations and computationalism.

Although extended mind semantics is sometimes useful, we can easily think of examples where the notion of extended minds gets us into trouble by allowing us to make claims about minds that we do not want to be able to make. Due to the lines that are crossed by positing extended minds, it is fair to say that the notion of extended minds is sufficient but not necessary for describing the phenomenology of cognitive aids. Even highly sophisticated cognitive aids such as neural implants are best described not as parts of mind, but as useful entities which interact with the (distinct) mind; the tight, cognitive coupling noted by Clark and Chalmers should be seen as a demonstration of the plasticity of the brain, not as the extension of the mind into the world. With Otto’s notebook or the philosopher’s pen and paper, there are cases in which explanations of cognition in terms of extended minds may be useful, but there are many more cases in which such explanations generate absurdities.