Concepts at the Interface is a scientifically grounded account of the nature of human conscious thinking, where thinking is understood in a broad sense that encompasses not just conceptual thought but the use of imagination in its various forms. It aims to explain how human thinking can be so flexible and also so powerful. Its primary focus is on the role of concepts and conceptual thought in unlocking information stored elsewhere in the mind in multiple formats and from multiple special-purpose bodies of knowledge. Hence ‘concepts as an interface’, serving to call up information of other sorts into what Shea calls ‘the playground of thought’.

In many respects the basic idea of the book is not new. The claim that abstract goals in a central executive operate in such a way as to call up information from a variety of special-purpose systems into a central workspace, where it can be sustained and manipulated, is decades old and will be quite familiar to most readers; as will be the fact that we frequently rely on various kinds of simulation that enable us to access implicitly encoded information about the world. But Shea elaborates these ideas carefully and systematically, drawing on a wealth of scientific detail.

While acknowledging that concepts play additional roles that are shared with other animals (for example, in perceptual recognition and categorization), Shea’s focus is on concepts as freely combinable representations that satisfy the so-called generality constraint. Concepts in the sense that he is interested in are representations that can be combined together with any other such representation of appropriate adicity to form a novel thought. (Roughly speaking, this means that any singular concept can be combined with any predicative one, and any two singular concepts can be combined with any two-place relational one, and so on.) It is this general-purpose re-combinability that explains the flexibility of human thought, while it is the capacity of such thoughts to access other kinds of information stored within the mind that explains its power.

After an introductory chapter laying out the main goals and assumptions of the book, the next three chapters slowly build the foundations for the account provided in chapter 5. Chapter 2 reviews the different kinds of representational structure that the mind contains. One is the general-purpose sort of structure afforded by combinations of concepts. The remainder comprise a variety of domain-specific non-conceptual representations, including analogue-magnitude representations of properties like number and temporal intervals, as well as the map-like structural representations of space and other abstract relations that are coded in the hippocampus and medial temporal cortex.

Chapter 3 discusses the sorts of computational processes that operate over these structures. The main distinction is between content-general and content-specific forms of inference. The former are constituted by broadly logical transitions among conceptual thoughts. (‘Broadly’ logical because these comprise all forms of inference that are content-general, including induction.) In contrast, most content-specific computations operate over non-conceptual representational structures, as when one adds together or subtracts analogue-magnitude representations of number (which human infants can do), or when one accesses a cognitive map to plan a novel route through space. But some transitions among conceptual representations are also content-specific, and themselves admit of two kinds: mediated and unmediated. One of Shea’s examples of the former is someone figuring out whether a chair will fit into the trunk of a car (a fully conceptual goal) by generating and manipulating an image of the chair and the shape and size of the trunk, thereby coming to recognize that the chair will (or will not) fit—a fully conceptual conclusion. Unmediated transitions among conceptual thoughts include such things as inferring from the concept of ‘dog’ to that of ‘barking’, from the concept of ‘red’ to that of ‘coloured’, or from the concept of ‘kill’ to that of ‘die’.

Chapter 4 then outlines the various ways in which information can be stored in the mind. These include purely conceptual information stored in semantic memory (that Paris is the capital of France) as well as mixed-format expectations of what a dog of a certain size might sound like when it barks. Then there are mixed-format representations of specific events from one’s past stored in episodic memory; theory-like bodies of knowledge about particular domains (folk-psychology, naïve physics); and a wealth of implicit statistical knowledge built into a whole range of specific systems, of the sort that might enable one to anticipate what the action resulting from a particular motor-instruction might look like, or what its further consequences in the world would likely be.

Chapter 5 then builds on the previous three chapters to explain how conceptual thought enables us to flexibly access information of multiple kinds, stored in many different formats, and processed using a variety of computational resources. Concepts are, in Shea’s memorable phrase, ‘plug and play devices’. We use conceptually formulated thoughts to direct attention and initiate motor-simulations, thereby pulling representations of various kinds into working memory and enabling domain-specific computations to operate over them. We use working memory as a ‘playground of thought’ in which freely combinable concepts can be used to unlock knowledge stored elsewhere in the mind, enabling us to explore possibilities and plan for both the short-term and the long-term. An especially interesting aspect of the chapter, I thought, was section 8, in which Shea suggests how combinations of concepts might serve to partition the information linked to the components, enabling only relevant information to be accessed. For example, the phrase ‘dog bites man’ activates only information about dogs in the agent role, whereas the phrase ‘man bites dog’ would only activate information about dogs as patients.

I have some concerns about this chapter, however. One is that Shea writes throughout as if it were the conceptual representations in working memory that control and activate the simulations that pull up new knowledge. But it is one’s background goals and intentions—which belong to the executive component of the overall working-memory system without figuring in the workspace itself—that direct attention, initiate motor-rehearsals, and direct queries to a variety of memory systems. While concepts themselves can figure among the contents of working memory, of course, and can spark yet more tacit inferences, these are pulled up into working memory as input ‘from below’ rather than executively directing what happens in the playground of thought ‘from above’.

Another concern about the chapter is that I think Shea’s framework could have been greatly enriched had he considered the role of the ventral attentional network in addition to the sort of top-down directed attention that is the immediate cause of entry into working memory. This network is largely right-lateralized and continually monitors both unattended external stimuli and representations that have become activated from any of the memory systems. These are evaluated for relevance to one’s current goals and background values, competing to attract top-down attention. For example, consider the well-known cocktail party effect: one is engrossed in a discussion with someone at a party when suddenly the sound of one’s name used in a nearby conversation ‘pops out’ of the background noise. This is a result of the work of the ventral attention network, which competes with the top-down network for control of the direction of attention, resulting in a switch of attention and entry of novel contents into working memory (in this case, one’s own name) when appraised as relevant or important enough. The functioning of this network could have helped Shea explain how the winnowing of information activated by the phrase ‘dog bites man’ can take place (for, of course, not everything stored in memory about dogs as biting-agents becomes conscious).

The remainder of the book explores the implications of the framework developed in chapter 5. Chapter 6 discusses how it enables human cognition to bypass the frame problem, because only a small set of contextually relevant information gets activated by a given goal or thought and is selected for entry into working memory. Here the content-specific inferential dispositions built into the domain-specific processing systems (which amount to built-in assumptions of relevance) play a role, for example. (This is yet another place where it might have been helpful to discuss the ventral attentional network, since it enables relevance to be computed ‘on the fly’, in addition to being built in.)

Chapter 7 then discusses two different ways that cognitive processes draw on, and are sensitive to, the meanings of the representations involved. The standard view (and the insight underlying the representational theory of mind) is that the computations that operate on representations to produce transitions among them can be arranged in such a way that starting from true representations as input they are likely to issue in true representations as output. This is exemplified in the sorts of domain-general computations discussed in Chapter 3, which are broadly logical in nature (including induction). But in a very different way it is also exemplified in the transitions that take place within domain-specific processing systems (such as expecting an object released above a desk to fall), as well as in unmediated transitions among concepts (like the inference from the concept of ‘dog’ to ‘barking’). For here the reliability of these domain-specific transitions has been built in by evolution or by previous learning.

Chapter 8 turns to metacognition of our own thoughts and thought-processes, discussing how awareness of the contents of the playground of thought makes it possible for us to learn new strategies and heuristics for thinking and problem-solving, as well as enabling us to estimate the reliability and usefulness of our concepts themselves (for inductive purposes, for example). Being aware of our working-memory contents, we can monitor the progress of our thinking and intervene when we sense that it is in some respect mistaken or insufficient.

The final chapter (chapter 9) reviews the basic picture presented in the book and reflects on its significance, discussing, in particular, how human thinking can be so flexible and so powerful.

There is much to applaud about this book. While the basic framework is likely to be familiar to most readers, and while it is wordier than it needed to be (there is a significant amount of repetition and each chapter concludes with an extended paragraph-by-paragraph summary), it is both richly grounded in cognitive science and frequently insightful.

Peter Carruthers
University of Maryland
pcarruth@umd.edu