Deflating Mental Representation presents Frances Egan’s Jean Nicod lectures, which offer a sharp and scientifically informed exposition of deflationism—her alternative to both realism and fictionalism in the psychological sciences. Egan’s writing is exceptionally clear, making the book essential reading for anyone interested in realism in science, the notion of belief, and the nature of perception.

The book consists of four interconnected chapters, each of which also stands on its own. The first two develop Egan’s deflationary account of mental representation and computational models of the mind, while the final two apply her framework, first to common-sense psychology and then to perception.

According to realism about representation in cognitive science, representational language should be taken literally: mental representations are genuine components of the mind, comparable to maps. By contrast, fictionalism—a form of eliminativism—treats talk of mental representation as a convenient fiction, much the same way as scientists metaphorically describe subatomic particles as tiny planets.

In chapter 1, Egan argues that realism is mistaken because although there are vehicles of thought that can be called ‘representations’, their content is not essential to them. The same kind of mental state that represents a fly in one system might have a different meaning, or none at all, in another. Yet fictionalism, she says, also falls short. While it’s true that mental states lack essential content, they can still be individuated as vehicles of computation. Egan concedes that these vehicles possess a kind of content, which she terms ‘mathematical content’. What she rejects is the idea that computational states or processes inherently have external or ‘distal’ content. This latter kind of content, which she calls ‘cognitive’, she treats as merely a heuristic gloss.

Consider, for example, Marr’s well-known account of vision, in which the visual system is described as computing a smoothing function—the Laplacian of the Gaussian of the retinal array—to help identify edges in a scene. According to Egan, this implies a mapping between the physical states of the visual system and the states posited by Marr’s model. This mapping captures the system’s causal organization. Saying that the system takes intensity values at points in a scene as input and calculates rates of intensity change as output means that when the system occupies a physical state interpreted as intensity values, it transitions to another state interpreted as rates of change. The inputs and outputs, then, represent the arguments and values of the computed function. Egan argues that this mathematical content is essential to the computational characterization of the visual system: it specifies what the system does.

Does this mean the visual system literally represents numbers? Not exactly. In chapter 2, Egan explains that the relationship between a computational or mathematical model and the physical system it represents is analogous to the relationship between numerical measurement scales and the properties they measure. Physical measurement involves mapping physical magnitudes—such as mass, length, or temperature—onto elements of a numerical scale. The abstract relational structure shared by the physical domain and the numerical scale grounds this mapping in a non-fictional way. However, there is no meaningful relationship between a system’s physical state (say, having a particular mass) and the number that represents it (say, seven grams), which Egan calls its ‘numerical representative’.

One might wonder, then, whether ‘mathematical content’ deserves to be called content at all, given that Egan’s claim amounts to saying only that there is a mapping between a mathematical (or computational) description and the states of a physical system. While this mapping does specify the system’s causal organization and predict its behaviour, it offers little explanatory value for treating the system as representing numbers. Moreover, realists typically press a further question: why does the mapping succeed in capturing the system’s causal organization and predicting its behaviour? The realist’s brief reply (to which we will return) is that the system genuinely computes vehicles that possess content.

As for cognitive content, Egan argues that its status as a mere heuristic gloss stems from the problem of indeterminacy (chapter 1). For instance, does the visual system represent edges or rather light discontinuities at various locations? Does the anuran toad represent moving wormlike stimuli (there, now), as Neander (2017) suggests, or nutritious flying objects (there, now), as proposed by Shea (2018)? According to Egan, neither interpretation is strictly correct. Each is defensible, depending on one’s explanatory aims. If the goal is to understand how the anuran visual system evolved, then ‘nutritious flying object (there, now)’ is more appropriate. But if the focus is on how the toad’s visual mechanism functions in the present, ‘moving wormlike stimuli (there, now)’ better fits the neural and stimulus conditions. Thus, for Egan, the attribution of cognitive content is inherently parochial, shaped by our own explanatory interests rather than by anything intrinsic to the system itself.

Like Ramsey (2020), I find the indeterminacy argument unpersuasive. The fact that we cannot specify exactly what the toad’s neurons represent does not imply that they fail to represent anything, nor that their representational content depends entirely on our explanatory interests. Not every attribution of content is equally valid; what counts as a plausible content depends crucially on how the mechanism operates, within what system, and in what environment. Moreover, the claim that the same kind of state might have carried a different meaning, or none at all, in another system does not entail that in the toad the state lacks genuine meaning.

Egan’s claim that representations in cognitive systems lack essential content—because the same kind of mental state could have had a different meaning, or no meaning at all, in a different system—seems, in fact, to beg the question. It may be that the auditory system also computes the Laplacian of the Gaussian of some sonic array, but whether those states are genuinely of the same kind as those in the visual system remains an open issue. What matters is what such states represent and what they are designed to represent within the system in which they are embedded. After all, an aneroid barometer can function as either an altimeter or a tire gauge, depending on what it measures and in what context.

In chapter 3, Egan revisits the measurement analogy to extend her deflationary approach to common-sense psychology. She offers a fresh perspective, portraying common-sense psychology as a social and regulative system that is enforced rather than merely employed. At the same time, she argues that it is a shallow theory: it makes no commitments regarding how propositional attitudes are realized in the brain. Contrary to Fodor, common-sense psychology does not presuppose the existence of a sentence-like cognitive architecture.

According to Egan, ascriptions of beliefs with content function mainly as a useful gloss for identifying causal complexes that aid in predicting and explaining behaviour. In this respect, they resemble everyday measurement practices. To say that an object has a temperature of 20°C is not to claim that the object possesses a particular numerical property; rather, it indicates that the object has a property located at a specific point on a numerical scale. Egan explains: ‘We gloss the physical magnitude temperature using the real numbers, but we don’t assume that the magnitude (that is, temperature itself) has all the properties and relations of the representing scheme—the reals—that we use to think about it’ (p. 92).

Properties of a model need not be properties of the system it models. Determining which properties of the representational scheme we can attribute to the empirical domain requires empirical investigation. Egan continues:

Analogous points can be made about our commonsense practice of specifying beliefs in terms of that-clauses. Just as in the measurement case, where the numerical specification doesn’t pick out a numerical property of the object, similarly in the belief case, the that-clause serves to specify the belief but without justifying the conclusion that the belief has the linguistic properties of the that-clause used to specify it. To say that a subject believes that it is raining is not to say that she stands in some substantive relation to something that means it is raining. Rather, it is to say that the subject possesses a causal complex that is specified in terms of the that-clause. We represent the attributed attitude—we gloss it, in my terminology—in linguistic terms, though without commitment to the idea that the attributed attitudes really have these linguistic properties. In so doing, we provide a way of representing the attitudes for our practice of predicting, explaining, and regulating behavior. (pp. 92–93)

While the measurement analogy is illuminating, choosing a scale for describing an object’s temperature (for example, Celsius or Fahrenheit) appears more arbitrary than selecting which propositional attitude to ascribe to a subject when predicting and explaining behaviour. It remains unclear what the equivalent of ‘choosing different scales’ would be within the domain of common-sense psychology.

We can agree with Egan that not every property of a model must correspond to a property of what it represents, and thus that ascribing a belief to a subject does not make the belief itself linguistic. Yet, when we ascribe a belief such as ‘it is raining’ to someone, we do more than indicate that the subject possesses a causal complex described by the that-clause. For example, we need to distinguish a belief that it is raining from a hope that it is raining. Furthermore, ascribing such a belief identifies the specific causal complex that produces rain-related behaviour; it would be strange if saying someone believed it was raining instead picked out the complex that produces uncontrollable laughter. Returning to an earlier point, a realist will ask why ascribing a belief in rain correctly singles out the causal complex that leads to rain-adjusting behaviour. A plausible realist answer is that the believer bears a substantive relation to a representation of the fact that it is raining. Thus, more detail is needed to develop a robust anti-realist interpretation of common-sense psychology.

In the final chapter, Egan addresses perception and introduces external sortalism, her version of adverbialism about perceptual experience. According to adverbialism, having a perceptual experience of something red, for instance, should be understood as perceiving redly—that is, as an episode of sensing in a particular manner. Adverbialism rejects both naïve realism, which treats a veridical perceptual experience as a relation between a subject and the perceived physical object, and representationalism, which sees perceptual experience as a relation between the subject and what the experience represents. Given Egan’s deflationism, her embrace of adverbialism is unsurprising: she seeks to deny any essential role for a relation between subjects and contentful states in understanding perceptual experience.

Adverbialism is not a widely held view in contemporary philosophy of perception and is often criticized for issues like the ‘many properties problem’, which Egan addresses (p. 131). Nonetheless, I found Egan’s version surprisingly compelling. Drawing on Sellars (1975) and Smart (1959), she interprets the phrase ‘perceiving redly’ as undergoing the kind of experience that typically occurs when one is confronted with something red. Notably, Smart employed this analysis to avoid invoking phenomenal properties in cases of afterimages and to defend a materialist view of the mind. When we see a red afterimage, Smart argues, there is nothing red present either in the external world or in our brains; rather, we are simply undergoing the experience that would normally accompany perceiving something red.

The challenge is that Smart’s position—and Egan’s by extension—is compatible with both representationalism and naïve realism, because part of Smart’s argument is to remain neutral regarding the nature of the experience we normally have when confronted with something red. Exactly what that experience consists in remains unexplained. Egan might argue that we should refrain from attempting such an explanation, leaving it to the empirical sciences. Whether this stance is convincing, however, remains open to debate.

In sum, Deflating Mental Representation is a clear and compelling work that will appeal to anyone interested in exploring alternatives to both realism and fictionalism in cognitive science.

Acknowledgements

Thanks to Caro Flores and Bill Ramsey for helping me think through some of the issues raised in this review.

Nico Orlandi
UC Santa Cruz
norlandi@ucsc.edu

References

Neander, K. (2017). A Mark of the Mental: In Defense of Informational Teleosemantic, MIT Press.

Ramsey, W. (2020). ‘Defending Representation Realism’, in J. Smortchkova, K. Dołęga and T. Schlicht (eds), What Are Mental Representations, Oxford University Press, pp. 54–78.

Sellars, W. (1975). ‘The Adverbial Theory of the Objects of Sensation’, Metaphilosophy, 6, pp. 144–60.

Shea, N. (2018). Representation in Cognitive Science, Oxford University Press.

Smart, J. J. C. (1959). ‘Sensations and Brain Processes’, The Philosophical Review, 68, pp. 141–56.