One approach to the science/philosophy interface involves using the analytical tools of philosophy to illuminate the foundations of some scientific discipline. The sciences make use of concepts like species, inflation etc., and one adopted goal of philosophy is ‘making sense’ of these notions: explaining what they mean, where they apply, how they cohere, etc. Another reverses this strategy, using established scientific concepts and theories to answer pre-existing philosophical questions.

These interfaces are not, obviously, independent. Well-understood scientific concepts are liable to contribute most to naturalistic philosophy. So, in the best cases, we find ‘boot-strapping’ between science and philosophy. Recent cognitive science presents an exemplary case of this sort of positive feedback, especially the strand centering on the representational picture of mind.

Foundational work in this tradition came from philosophersFootnote ¹, and philosophically-minded scientistsFootnote ², and explicitly aimed at resolving traditional philosophical questions. Most centrally: how is the rational causation of behaviour possible in a material world? In broadest outline, the answer pursued is: physical states of the system (e.g. the brain) can represent states of the environment, and while the rational relations between these states depend on these representational/semantic properties, their causal properties are determined by their physical properties. The functioning of the system depends on a correspondence between these representational and physical properties.

Representationalism has flourished across the cognitive sciences, wherein complex behavioural capacities are explained in terms of the processing and transformation of representations.Footnote ³ This is great news for the scientist. However, as always, for the philosopher, things aren't so peachy.

Firstly, the notion of representation itself poses major philosophical worries. Representation seems to bring with it puzzling abilities not found elsewhere. Minded creatures can represent entities to which we can't stand in any physical relationships, due to their lack of existence (as when I think about Macondo), and distinguish between one-and-the-same entity (as when I wonder whether Muhammad Ali is braver than Cassius Clay). For these sorts of reasons, a major project of philosophy of mind has been ‘naturalizing intentionality’: showing how these representational properties are ‘nothing-over-and-above’ some complex of ontologically unproblematic properties.Footnote ⁴

Secondly, and perhaps surprisingly, while representationalism's successes have been substantial, they have primarily been outside the areas that the philosophically-minded might have hoped. Staples of philosophical accounts of mind and action (beliefs, desires, and intentions, not to mention consciousness) have remained stubbornly untouched by most scientific work. The scientific gains have instead centered on the ‘subpersonal’ level: unconscious inference in the visual system, motor control, sentence parsing, etc. While fascinating topics for a psychologist, one could be excused for thinking that such phenomena shed little light on philosophical debates about rationality and mentality.

Enter Nick Shea. His ambitious Representation in Cognitive Science aims to develop an answer to the question ‘What is representation, as appealed to in the cognitive sciences?’ by focusing precisely on these subpersonal cases. He is, advisedly, cautious about how far such an account will get us towards a general philosophical account of representation, but suggests that his account may provide a starting point. In the remainder of this review, I shall outline his proposal concerning the former, before raising some concerns with the latter.

Shea synthesizes and elaborates the programs of indicator semantics and teleosemantics.Footnote ⁵ The former aims to ground representational content in correlations between representer and represented. The latter adds that representations do not merely correlate with their targets, but function to do so, in the evolutionary/biological sense. This cluster of views has provided the dominant approach to naturalizing intentionality in analytic philosophy. Shea aims to further this program, resolving many of the issues raised against it over the years.

Shea begins by asking: what must representations be, if they are to do what cognitive scientists need them to? The first step in asking this requires identifying the kinds of phenomena cognitive scientists aim to explain. Shea singles out ‘robust outcome functions’: behaviours and behavioural achievements reliably produced in different environments. Some dogs, for example, are reliably able to catch frisbees. This capacity calls for explanation. More carefully, for two kinds of explanation: why do these dogs behave this way? and how do they do so?

Paradigmatic answers to these why-questions appeal to history. Ontogenetically, perhaps this dog has received treats in the past for catching frisbees. Because this behaviour has been rewarded, the dog is now catching the frisbee for a reward. Phylogenetically, perhaps ancestral canids which caught nearby flying objects out-reproduced those which didn't. In either case, a history of selection contributes a normative element to the capacity: the outcome/behaviour is for some benefit, and can be successful or unsuccessful. Shea calls such historically selected outcome functions ‘stabilized’.Footnote ⁶

When a robust outcome function is stabilized, it becomes a ‘task function’.Footnote ⁷ Task functions, then, are capacities to reliably perform in certain ways which exist on account of the benefits that such performances accrue to the organism. They combine reliable behaviour with backwards-looking teleology.

Stabilization thus provides an explanation of why a system has a robust outcome function. The question then becomes how such an outcome is achieved. In many cases, the best explanation of how a system manages to robustly achieve a certain desirable outcome appeals to the algorithmic exploitation of correspondences between internal states of the system and features of the environment. Continuing our example, if some internal state of the dog were to correlate with the frisbee's location, the dog could exploit this correspondence in tracking, and catching, it. For Shea, when the best explanation of a stabilized task function appeals to the manipulation of states standing in such correspondences to features of the distal environment, these internal states represent those external features.

Shea identifies two classes of correspondence: correlation and structural correspondence. Correlation is cashed-out in the usual way: an internal state correlates with a distal feature when the presence of the internal state increases the likelihood of the presence of the distal feature. Structural correspondence exists when there is a mapping from the relations between internal states to the relations between external states. The paradigm is a roadmap: relations of distance and direction between points on a map correspond to the relative locations of the cities the map targets. When such correspondences between internal and external states exist, the former can be used by the cognitive system as a proxy for the latter. And when they are used, i.e. when the algorithmic processing of the system is systematically sensitive to variation in such correspondences, the internal represents the external.

In addition to the constraint that such correspondences must be used by, not merely present in, the system, Shea insists that the representation-grounding explanations must appeal to ‘unmediated’ correspondences. That is, the explanatory relevance of this correlation between internal and distal states to the system's functioning must not depend on further correlations. To use the standard example, the correlation between a frog's internal states and flies is transparently relevant to explaining its ability to catch flies. The correlation between internal states and ambient black dots, while perfectly real, is relevant only given the further correlation between flies and dots. So, the former correlation is unmediated, and thus relevant to the fixation of content, while the latter is not. An analogous constraint applies to structural correspondence. These two constraints provide Shea's responses to the major worries with naturalized intentionality (Chapter 6).

Thus concludes my brief and selective exposition. I hope I have captured enough of the elegance of Shea's theory to entice the reader into spending more time with it. The remainder of my remarks concern the prospects for using this theory as a ‘staging post’ (p. 225) in the construction of a general theory of representation. Shea is openly sceptical that his account will apply straightforwardly to ‘personal level’ psychology: beliefs, desires, and conscious states (p. 42). However, he may underestimate the extent of mental life which outstrips his explanatory resources. In particular, Shea's account seems well-suited for explaining representations whose content is exhausted by their function of tracking distal states. But many paradigmatic subpersonal representations seemingly feature content beyond mere correspondence to the environment.

Consider first colour perception. On the view most amenable to Shea's treatment, colour vision functions to identify reflectance properties of surfaces in the environment.Footnote ⁸ However, states of the visual system do not merely correlate with distal colours. Colour perception is categorical.Footnote ⁹ That is, our visual system transforms information about environmental colour, imposing on it an internally-specified structure. Considered objectively, surface reflectance properties, and the wavelengths they reflect, vary continuously. However, we do not ‘see’ this continuous variation: rather, we see colours as existing in ‘bands’ centered around paradigm instances. We see, for example, shades of green as more like one another than they are like shades of blue or yellow, even when the distance between the two in ‘colour-space’ is identical.

Another example comes from sentence recognition. When we hear sentences, we hear them as having grammatical structure. We hear ‘Visiting beaches can be fun’ and ‘Stampeding bulls can be scary’ as having different grammatical structures, even though the physical realizations of these sentences (sound patterns, pixel distributions, etc.) need not differ in relevant ways. In this case there seems to be nothing in the environment for our perceptual representations to correspond to.Footnote ¹⁰ These cases show that while correspondence to the environment may play an important role in explaining representational capacities, it leaves something out. Namely, representational capacities are rarely simple ‘mirrors’ of the environment. There is often a substantial contribution from the organism, determining which categories are used, how they relate to one another, etc. This internal contribution seems absent from Shea's discussion. These kinds of cases suggest that the explanation for why some representation has the properties it does does not always solely appeal to the fact that internal-external correspondences facilitate behaviour.

Shea convincingly argues that his account applies to many subpersonal systems. And philosophers and cognitive scientists should celebrate the existence proof of a naturalized semantic theory. The disagreement, if there is one, concerns how much of the subpersonal world is amenable to this approach, and thus how big a gap there is between what we now have and what we want: a general theory of content.

Shea identifies Ruth Millikan as a primary inspiration for this work. This influence is obvious in the book's content, but also in its style. Like Millikan's work, it is often a tough read. But it massively rewards the effort it demands. It is as rich and rigorous as any recent work of philosophy, and is already a focal point in the literature. It beautifully synthesizes the philosophical tradition and contemporary work in psychology, neuroscience, and ecology. I recommend this book to anyone interested in the mind and its place in nature.Footnote ¹¹