The role of representation in understanding cognition continues to be one of the most hotly contested debates in contemporary philosophy of cognitive science. Though sometimes appearing irresolvable due to fundamental differences in intuition, the conversation has developed somewhat in recent years, gravitating towards a set of well-defined themes that have brought clarity to long-standing disagreement. Such themes include the functional criteria required for a cognitive mechanism to operate in a recognizably representational manner, how content determination can be understood in a way that accords with explanatory practice in cognitive science, and whether representational descriptions contribute to explanation in a way not reducible to non-semantic, causal descriptions.
With Representation in Cognitive Science, Nick Shea further develops these themes, expanding on his own work over the past two decades, as well as advancing broader trends in the literature on cognitive representation. Like many of his contemporaries who defend the indispensability of representation talk in cognitive science, Shea concentrates his analysis on the explanatory role of subpersonal representation (roughly, unconscious states involved in information processing), in contrast to personal-level representation (roughly, doxastic and conscious states). The key idea of the book is that representational explanations pick up on patterns in vehicle–world relations that non-representational explanations miss, providing a distinct representational scheme of explanation. The nature of this pattern is the existence of a natural cluster in which ‘stabilizing processes’—such as evolution and learning—affix to ‘robust outcomes’—those resulting from broadly goal-directed behaviour—and a set of internal mechanisms bearing exploitable relations to distal features of the environment. There are two key moving parts in Shea’s positive account, which he titles ‘varitel semantics’: the first is the notion of a ‘task function’, namely, a function that is both robust and the target of a stabilization process; the second is the notion of exploitable relations. Much of the book’s eight chapters are dedicated to elaborating these ideas and how they interrelate. The result is a comprehensive, albeit technical, contribution to the representation debate that serves to progress contemporary representational theories of mind and, in the process, illuminates more precisely where outstanding disagreement might lie. In what follows, we summarize each chapter before turning to some comments on the book as a whole.
Shea uses Part 1 to set clear constraints on what he hopes to achieve with this book. He begins the introductory chapter with a clear and comprehensive literature review, which will help readers new to the debate gain a working understanding of the philosophical issues surrounding cognitive representation. Shea’s target is the question of content determination: ‘what makes it the case that a representation has the content it does?’ (p. 9). Shea highlights the importance of the question, as an answer would mark a significant step towards developing a full understanding of cognition, under a representational theory of mind (RTM). Indeed, given the prevalence and success of representational explanation, Shea does not defend but rather assumes RTM. His aim is sharply focused on establishing a realist attitude towards vehicles of content.
Following a brief but useful tour of the existing approaches to the problem of content determination, Shea associates himself with a broadly teleosemantic approach. He notes several long-standing issues for teleosemantics, such as the difficulty of identifying consumers of cognitive representations, identifying a notion of aetiological function that is neither too broad nor too narrow, and the challenge posed by the possibility of ‘swampman’, a creature physically identical to an adult human but lacking an evolutionary history. Finally, Shea identifies the principal desideratum his theory must fulfil: a convincing account of the way content gains explanatory purchase in scientific approaches to subpersonal cognition. This is the fundamental question that realists must answer. Without a convincing response, anti-realists are free to dismiss representation talk as, at best, a ‘gloss’ on non-semantic explanations. At worst, the anti-realist may argue that content attributions are arbitrary and misleading, raising difficult questions for RTM that make eliminativism an appealing alternative.
In Chapter 2, Shea sets out his relevant underlying commitments and also indicates some constraints on his theory. Here he makes a convincing argument that relying on intuitions is an uncertain way of making progress in understanding subpersonal mechanisms. Shea also defends his commitment to externalism by arguing that the explanatory value of mental content derives from the way it enables us to understand the functioning of a system without referring only to intrinsic properties of the system.
Shea closes Chapter 2 by looking ahead at what is novel about his theory of content determination, which he terms ‘varitel semantics’. Shea identifies two main conditions that must be met in order for content to be explanatorily useful in a given context. These are (1) a robust, stabilized function, and (2) an exploitable relation with the external world. Shea argues that each condition can be fulfilled in a variety of ways (hence varitel semantics). This development of an effective syncretism of disparate theoretical approaches is a significant contribution to the debate. Whether varitel semantics stands or falls depends upon whether Shea is successful in persuading others that the theory preserves the advantages and avoids the pitfalls of each theory, or vice versa.
It is in Part 2 that Shea sets out the details of varitel semantics. As a sibling of traditional teleosemantic accounts, normative notions of success and failure are central to understanding the explanatory import of content. Shea outlines his theory of function in Chapter 3, providing a naturalistic analysis of these concepts. He sets out his pluralist approach, acknowledging three separate ways that a task function can be ‘stabilized’ in an organism: natural selection, persistence of the organism, and learning through feedback. An example of a robust, stabilized function is the ‘dancing’ behaviour of bees. The behaviour is robust because it occurs across a range of contexts in response to different stimuli, and has been stabilized by natural selection because hives who practise dancing reliably harvest greater volumes of nectar, thus improving their fitness. At this point, Shea sensibly foreshadows a response to the lurking challenge of swampman, which threatens any aetiological theory of function. Shea indicates that his emphasis on success and failure (notions impossible to apply to swampman) force him to bite the bullet and accept that at the moment of creation, swampman’s internal states do not possess task functions. However, once stabilizing processes such as survival and learning supply selection pressure, then normative notions such as success and failure gain explanatory purchase.
Chapters 4 and 5 detail the kinds of exploitable relations that are necessary for content to have explanatory purchase. Chapter 4 focuses on correlational information, with a detailed study of the way this relationship is put to work in contemporary theories, the associated criticisms, and how they can be avoided. Using a notion of ‘unmediated explanatory information’, Shea aims to show how correlational information can be made a respectable bearer of meaning in order to avoid defending the unattractive consumer-semantics of other teleosemantic theories. If one isn’t persuaded that correlational information is always an appropriate exploitable relation, Shea’s pluralism permits him to appeal to an alternative: structural correspondence. This is the subject of Chapter 5. In a parallel move, Shea defines ‘unmediated explanatory structural correspondence’ and explores several case studies that demonstrate how structural correspondence can be instantiated. Shea also tackles some niche problems in this area, defending the principled separation between correlational information and structural correspondence, as well as the problem of approximate instantiations. By the conclusion of Part 2, the reader is left with all the main details of Shea’s position, giving the impression of a powerful and carefully conceived theory.
Chapter 6 opens Part 3 with an examination of how Shea’s varitel semantics deals with common challenges faced by theories of content at the subpersonal level. Chief among these are accusations that content is indeterminate and that reliance on historical factors for understanding content undermines its explanatory value. Of particular interest is the way Shea balances insistence that historical factors are necessary for the explanandum targeted by representational explanations—successful behaviour—while avoiding conflating misrepresentation with malfunction. Shea also returns to the swampman thought experiment, arguing that while swampman initially lacks subpersonal content according to varitel semantics, this does not preclude it possessing conscious states nor gaining subpersonal content over time, thus partially defanging worries associated with denying swampman content.
Chapter 7 shows how to understand ‘descriptive’ and ‘directive’ content—better known as ‘indicative’ and ‘imperative’ content—under varitel semantics. Following a broadly teleosemantic approach, exploitable relations can play a descriptive role, a directive role, or both, depending on the details of the representation’s relation to a set of conditions in explaining a task function. Descriptive representations refer to those cases where explaining a task function depends on a representation producing a certain condition. Directive representations refer to those cases where explaining task functions depends on a certain condition already obtaining when the representation causes behaviour. The chapter includes an extensive discussion of how Shea’s view compares to existing accounts and how it applies to several case studies. While perhaps rather niche for some readers, Chapter 7 will interest those working in the weeds of the representation debate and anyone concerned with the technicalities of how varitel semantics relates to teleosemantics and similar accounts.
With the positive account in place, Chapter 8 closes the book by returning to the primary question of how and what content explains. Fundamentally, for Shea, representation occurs where vehicles with exploitable relations to distal entities contribute to stabilization and robustness. Along the way, Shea discusses, among other things, whether there is a role for content when a non-semantic causal description is always available, which he answers in the affirmative, arguing that representational descriptions capture vehicle–world patterns that would otherwise be missed; and whether content is truly causally efficacious, which he remains neutral on, suggesting his account shows semantic properties to be at least explanatory but not necessarily themselves causally efficacious.
The closing chapter’s discussion of causal efficacy and explanatory traction again invokes broader issues within philosophy of science. For example, as Shea notes (pp. 208–9), determining whether content properties are truly causally efficacious requires addressing broader questions about special science properties, as well as the role of historical factors, given their part in varitel semantics. Further philosophical work could also be done to secure the overall inference that Shea wishes to draw. Given that he aims to justify his realism about ‘vehicles of content’ by providing a robust defence of the explanatory value of positing contents, Shea might strengthen his case with some discussion of the merits and pitfalls of such abductive inferences. Overall, we remained somewhat unconvinced that varitel semantics alone delivers the robust realism about representation that Shea aspires to, given these unresolved points. However, Shea is not alone within the representation debate in under-emphasizing these broader issues concerning realist inferences. As such, this is not a reason to fault this book in particular, but is rather a symptom of the representation debate as it currently stands, which is often somewhat detached from broader debates about realism and explanation in philosophy of science.
As a note on methodology, the book stands out by devoting a great deal of attention to detailed case studies across cognitive science, as Shea himself observes (p. 197). On the one hand, these case studies can be rather dense and may be daunting for those not steeped in the literature. On the other, they provide the book with an impressive level of rigour, continually demonstrating how Shea’s account applies to a range of scientific contexts. This approach is of value to both philosophers and scientists. For philosophers, the case studies provide a useful reference point for the debate, and one can imagine others returning to these in future papers. We expect also that researchers working in cognitive neuroscience will be satisfied that Shea’s philosophy is fully informed by the latest empirical work, and will welcome the sound defence of their practice.
Detailed treatment of core issues in the debate, such as correlational information and structural correspondence, are somewhat dense but provide a reliable overview for those involved in the discussion. Balancing accessibility, argument, and precision is a challenge for any book dealing with the perennial and complex topic of cognitive representation, and especially so for one offering a thorough defence of a pluralist approach. In general, however, Shea presents his case with clarity. Moreover, the paragraph-by-paragraph summary at the end of the book provides the reader with a useful reference point and a tool to both quickly review the book’s content and more easily discern how the chapters fit together. This is a practice that would greatly benefit philosophy as a whole if it were more widely adopted.
With its impressive command of the literature and rigorous defence of an account that advances several promising trends in the representation debate, Representation in Cognitive Science will undoubtedly enter the canon in the coming years. It is therefore essential reading for anyone involved in the debate. Moreover, given its examination of issues in explaining cognition more generally, it will be of interest to many working in areas adjacent to philosophy of cognitive science.