A Mark of the Mental: In Defense of Informational Teleosemantics is Karen Neander’s long-awaited first book. It achieves three main goals. First, it establishes teleosemantics as a leading candidate for naturalizing intentionality. Second, it develops a highly original and innovative version of teleosemantics, informational teleosemantics (IT), which stands in contrast to consumer-based accounts. Third, it offers plausible solutions to the various content indeterminacy problems that have historically plagued the topic. Throughout the book, Neander exhibits the deft interweaving of philosophy of biology and philosophy of mind that has marked her career.
Stated roughly, IT says that the content of a (sensory-perceptual) representation is whatever is supposed to cause it (149). More precisely, at its most basic, IT holds that a sensory-perceptual representation, R, is about C if there is a mechanism that has the function of producing R’s in response to C’s (151). So, for example, a specific midbrain activation pattern in a toad might be about long wriggling things, if there is a mechanism in the toad’s brain that has the function of producing that pattern in response to long wriggling things. In this, IT differs from consumer-oriented views, which hold that a representation’s content depends on the mechanisms that utilize (consume) it. For example, in Millikan’s consumer-based view, a toad’s midbrain activation pattern is about worms because that is what it must correspond to for the consumer mechanism (which includes bits of the hindbrain involved in the tongue snap) to fulfill its function in the historically normal way (e.g., Millikan Reference Millikan1989, 290). In this respect, Neander positions herself within a tradition that includes Stampe (Reference Stampe1977), Dretske (Reference Dretske and Bogdan1986), and the early Fodor (Reference Fodor1984).
A crucial distinction between IT and its predecessors is that IT is, in the first place, a theory of sensory-perceptual representations. It is not a general theory of concepts; it is a foundation for that theory. IT also incorporates elements of causal and resemblance theories of meaning. Moreover, while one could accept IT without accepting the selected effects (SE) theory of biological function, SE is an important part of the book’s intellectual backdrop.
Chapter 1 poses the book’s main question: how do sensory-perceptual representations get their contents? By virtue of what is a bit of neural activation in my visual cortex, say, a representation of a blue square rather than a red triangle, or nothing at all? Neander spends much of the chapter distinguishing this question from a host of related questions having to do with the nature of representational vehicles, natural information, and derived intentionality.
In particular, Neander suggests that we reinterpret the traditional distinction between original and derived intentionality. Traditionally, minds possess original intentionality; artifacts derived. She recommends a different way of thinking about the distinction: we should say that something has derived intentionality if it gets its intentionality from something else that has intentionality; otherwise, it has original intentionality (9). Framed in this way, there is a serious scientific question about which kinds of mental representations have original intentionality. She argues that beliefs, plausibly, lack original intentionality, since they derive their intentionality from the intentionality of concepts. Many concepts lack original intentionality, too; they merely borrow it from other concepts (plausibly, my concept of bear food derives its meaning from the concept of bear and the concept of food). One possibility is that only sensory-perceptual representations have original intentionality (10). Hence the book’s focus on these sorts of representations.
Neander notes that, once we have a theory of original intentionality in place, we are still a long way off from a complete theory of human representation. Our theory of derived intentionality will have to carry a large part of the burden in explaining how concepts and beliefs get their contents. One might think this limitation is a weakness of her theory, but it is not. First, one well-established methodology in philosophy is to figure out how a theory applies to the simplest cases (e.g., how representations of colors, lines, and shapes get their contents) before tackling the complex ones. From an evolutionary perspective, that strategy makes a lot of sense, since IT is meant to apply to some of our phylogenetically distant ancestors. Second, she gives some productive and useful leads about the emergence of concepts, particularly in chapter 8’s discussion of second-order similarity.
Chapter 2 aims to demonstrate that cognitive neuroscientists do, in fact, posit nonconceptual representations and that such posited representations are error-permitting. They are the sorts of things that can represent correctly and that can misrepresent. She describes an unusual visual system defect that causes those afflicted to misrepresent, in a systematic way, the location of visual objects. (For example, if a square is presented in the upper-right quadrant of the visual field, the person might represent it as located in the upper-left quadrant.) This is a kind of error-permitting, nonconceptual representation.
Chapter 3 is one of two chapters devoted to biological functions. She raises a “methodological conundrum” about functions. Biologists often distinguish “why” questions (why do mammals have kidneys?) and “how” questions (how do kidneys filter the blood?). Many philosophers of biology say that, while SE functions are relevant for thinking about why questions, they are useless for thinking about how questions. For the latter, we need something like Cummins’s causal role functions. But if that is right, why should physiologists and other experimental biologists care about SE functions at all?
Neander offers the following solution. SE functions are vital to the work of physiologists, not just because they explain how traits evolve, but because they help physiologists make good generalizations. When physiologists explain how traits work, they must generalize; they abstract away from the idiosyncratic details of this or that organism. To do so, they describe a kind of idealized system in which all the parts and processes perform their normal-proper functions. (Whatever else these normal-proper functions are, they are the sorts of things that admit of a function/dysfunction distinction and a function/accident distinction.) Moreover, she argues, SE functions are highly plausible candidates for identification with these normal-proper functions. If so, SE functions are pivotal to the physiologists’ mission. I agree with her conclusion and have also argued—though on different grounds—that we should reject the style of pluralism that tries to restrict SE functions to evolutionary biology alone, to the exclusion of large parts of genetics, neuroscience, and ecology (Garson, Reference Garsonforthcoming a).
Chapter 4 gives the first argument for IT, the “methodological argument.” Cognitive neuroscientists, she claims, are implicitly committed to something like IT; so the success of cognitive neuroscience lends support to IT (87). But why is cognitive neuroscience implicitly committed to IT? The argument seems to be this: cognitive neuroscience postulates natural-factive information. It also postulates normal-proper functions. It even postulates mechanisms that have the (normal-proper) function of doing things with (natural-factive) information. She believes that anyone who admits this last fact is, effectively, admitting the correctness of IT: “normative aboutness is posited once the normal-proper functions of cognitive systems are wedded to the natural-factive information they have the function to operate on” (85). I am not convinced by this argument and will return to it below.
Chapter 5 gives the start of a more persuasive argument for IT. The rest of the book develops the argument. The argument, in short, is this: first, we have strong, pretheoretical reasons (reasons that are independent of one’s favored theory) to hold that the content of a perceptual representation is some configuration of surface properties. For example, we have good pretheoretical reasons to hold that, when a toad sees a worm, the content of the relevant neural representation is something like there’s a small elongated object moving parallel to its longest axis rather than there’s a worm or there’s a nutritious meal. Second, IT yields precisely that content, and many other theories do not (or, if they do, they suffer from other internal problems). So, we should accept IT.
Let me be more specific. When a toad sees a worm, there are several different stages of visual processing, but the one most relevant to this story involves a certain class of midbrain cells, the T5-2 cells. High-frequency T5-2 activity corresponds very well to worms. The content ascription that Neander asks us to entertain is this: high-frequency T5-2 activation means there’s a small elongated object moving parallel to its longest axis, at such-and-such location in the visual field.
This raises a question long familiar to philosophers of biology. What makes the T5-2 activation about some configuration of surface features, rather than worms, or nutritious meals? Her answer is that content ascriptions should be constrained appropriately by our knowledge of the underlying neural machinery. The visual mechanisms that allow toads to capture worms are causally sensitive to features of the world such as shape, size, and motion (114). Those mechanisms are not causally sensitive to the presence or absence of being nutritious. Nor are they causally sensitive to species membership. So T5-2 activation does not have the content there’s a nutritious meal or there’s a worm. (Incidentally, the debate in the 1990s about the content of the frog’s visual representation of the fly was never a matter of mere intuition mongering, as some have complained. It was, and still is, about which content ascription hangs together best with the commitments of cognitive neuroscience and other relevant scientific disciplines.)
She supports her claim by considering what she calls the localization content of the T5-2 activation: T5-2 cells have fairly large receptive fields. As a consequence, toads cannot localize worms very precisely just by relying on T5-2 cells; they need more processing. Let us suppose, for the sake of argument, that a toad’s T5-2 cell has a receptive field that corresponds to a fairly large region of the visual field, say 30 degrees. Now, suppose someone claimed that the content of the T5-2 activation is there’s a worm at such-and-such location, where the location is specified very precisely, say, to within two degrees of the visual field. That would be an implausible content ascription. Why? Because T5-2 cells lack the capacity to respond to stimuli with that degree of precision. (Among other things, that content ascription would make some of the toad’s behaviors inexplicable. For example, why must the toad continue to engage in orienting behavior, even after a T5-2 cell has activated?) In sum, contents should be “tied to the information-processing functions of the mechanisms that produce [them]” (114).
Chapter 6 is the second of the book’s two chapters that deal squarely with biological functions. IT holds (again, at its most basic) that a representation R is about C if there is a system that has the function of producing R’s in response to C’s. The feasibility of IT hangs crucially, then, on the existence of response functions. Sometimes a function of a trait is just an effect. Consider the function of the heart, which is to pump blood, or the function of the zebra’s stripes, which is (at least on one current theory) to deter biting flies. But sometimes, the function of a trait is not just to produce an effect, but to produce an effect in response to a certain stimulus. The function of the immune system is to release an army of white blood cells, but only in response to infection. A function of the pineal gland is to release melatonin, but only in response to the dimming of light.
Some have looked askance at the very idea of a response function. They think that a trait’s functions are a subset of its effects; so, when we state the function of a trait, we should cite only its effects. We should not also cite its causes. Neander shows that there is nothing suspect about response functions and that it is hard to see how certain traits evolved without them (e.g., the pineal gland’s ability to release melatonin would not do much good if it was not triggered by the dimming of light). She also develops her thoughts on the notion of (natural-factive) information, preferring what she calls a “simple causal” theory of information to a theory based on correlations or conditional probabilities.
The last three chapters of the book offer solutions to various content indeterminacy challenges. Teleosemantics’ critics have often argued that teleosemantics cannot deliver the sorts of fine-grained contents that humans and other animals are capable of representing. These content indeterminacy challenges are conventionally divided up into two: the “vertical” and the “horizontal” problems. Suppose a worm crawls into a toad’s visual field and causes a T5-2 activation. The vertical problem is this: what makes the T5-2 activation about a worm (say), rather than a nutritious thing, or a small elongated object that moves in a direction parallel to its longest axis? (We will assume, for the sake of argument, that all three properties were coextensive in the toad’s ancestral environment.) The horizontal problem asks why we should say that the T5-2 activity is about worms at all (or some suitably distal object) rather than worm-shaped retinal impressions or certain patterns of optic nerve firing. In this part of the book, Neander shows that there are actually six distinct content indeterminacy challenges, and she offers plausible solutions to them all.
Chapter 7 deals with the vertical problem and two related problems. IT says that the content of a representation is whatever in the world is supposed to cause it. Fortunately, causation is property-sensitive (159). So the toad’s visual system can be causally sensitive to certain configurations of properties involving size, length, and motion, without also being causally sensitive to properties such as species membership or nutritiousness (even if they are all coextensive). So T5-2 activation need not be about flies or nutritious things. (Note that her argument here presupposes the essential correctness of her argument in chap. 5 that the right content ascription is something like there’s a small elongated object moving parallel to its longest axis. One goal of chap. 7 is to show how IT delivers this content.)
In chapter 8, Neander expands her theory of content to include relations of what she calls second-order similarity. Sometimes, she holds, a representation has a content not because there is a mechanism that was selected for producing that representation in response to that content, but because there is a mechanism that was selected for producing the analog of that content. What is a content’s analog? Suppose there is a set of representations in the mind and a set of objects in the world. Suppose that the relations of similarity and difference that obtain between the representations mirror the relations of similarity and difference that obtain between the objects. Then, a certain representation is an analog of a certain object in the world if the representation stands in the same relations to the other representations, as the object does to the other objects. Neander uses a doctor’s pain measurement scale as a simple example. Here there is a set of representations, namely, the numbers from 1 to 10. There is also a set of pains, namely, the pains from least to most intense. The number 10 is the analog of the worst pain imaginable, since 10 stands in the same relation to the other numbers (namely, being the largest) as the worst pain imaginable does to the other pains.
A benefit of extending the theory in this way is that it helps teleosemantics solve an outstanding problem, namely, the problem of novel contents. How can people represent novel properties, that is, properties that neither they nor their ancestors have ever encountered? Neander considers Hume’s missing shade of blue. She thinks, as Hume did, that even if neither I nor my ancestors ever encountered a certain shade of blue, I could plausibly imagine exactly what it would look like. That is because my system for representing colors is able to produce the analog of the missing shade of blue. She also uses these second-order similarity relations to solve two other content indeterminacy problems, namely, how I can represent C, rather than Q, when C is a determinate of Q (say, scarlet a determinate of red), and vice versa. This chapter also begins to broach the problem of conceptual representations in addition to nonconceptual ones.
The final chapter tackles the last indeterminacy problem, the horizontal problem (alternatively, the problem of distal content). Why is T5-2 activation (say) about distal objects rather than more proximal stimuli, such as retinal impressions or optic nerve stimulation? More generally, if C and Q are two members of a sequence of stimuli that causes R, and Q is more proximal than C, why is R about C rather than Q? (Note that this problem is orthogonal to the vertical problem. In the vertical problem, we assumed that all candidate contents were suitably distal—for example, worm, elongated moving thing, nutritious thing. Here, we are asking why we are entitled to that assumption.) Dretske (Reference Dretske and Bogdan1986) pointed out the problem for teleosemantics and developed an ingenious solution to it, to which I will shortly return.
Neander develops a new solution that exploits informational asymmetries between C and Q (222). According to her simple causal account of information, a singular event E carries information about another, F, if F caused E. So Q carries information about C (e.g., retinal impressions about worms), but not vice versa. She thinks that between two candidate contents, Q and C, if Q carries information about C, but not vice versa, then C is the correct content (also see Neander Reference Neander, Ryder, Kingsbury and Williford2013). She closes the book with some speculations on the mechanisms that let people perceive contents as distal. What allows me to perceive this chair as distal, that is, as an object out there in the world?
I want to focus on three problems for IT as a whole, and particularly for her solution to distal content. The first concerns the methodological argument outlined in chapter 4. There, she indicates that cognitive neuroscientists are already implicitly committed to something like IT. After all, they posit information, they posit functions, and they posit mechanisms that have the function of manipulating information. I agree that they do posit all of those things, but that falls short of a tacit commitment to any special theory of representation. Surely Neander has proven that IT does not require any ingredients that cognitive neuroscientists are not already using. Its ontology is not more expansive than the ontology that cognitive neuroscientists take for granted. But that should be a constraint on anyone’s theory of intentionality, whether IT, a consumer-oriented view, Fodor’s asymmetric-dependence view, or even Cummins’s “picture” theory. It seems to me that we need some further premise to show that cognitive neuroscientists are committed to the claim that being the product of a mechanism that has such an information-carrying function is just what it is for a representation to have a content.
There are two more problems, both of which concern distal content. First, some commentators have worried that Neander’s solution gives overly distal contents (Price Reference Price2014, 590; Artiga Reference Artiga2015; Schulte Reference Schulte2018). Schulte develops a particularly vivid thought experiment. Suppose there is a kind of bug that turns red whenever it eats potassium. Suppose, moreover, that toads have mechanisms for detecting the redness of bugs, and those mechanisms evolved precisely because toads need the potassium those bugs contain (it is otherwise pointless to eat the bugs). Whenever they see a small moving red speck, they enter a specific neural state. Neander’s solution seems to imply that the content of the neural state is there’s a potassium-rich thing rather than there’s a small red thing. After all, the bug’s redness carries information about its potassium richness, but not vice versa. So Neander should either amend her view to exclude such hidden properties or accept that sometimes sensory-perceptual representations are about hidden states and not just surface features of objects.
A second problem with Neander’s solution to distal content is that it is unmotivated; in my view, Dretske (Reference Dretske and Bogdan1986) not only identified the problem but adequately solved it. Dreske’s view is that if a creature has only a single source of information about the world (say, an earthworm with nothing more than a cluster of photoreceptors near the surface of its skin to distinguish light and darkness), then there is no principled solution to distal content. In order to get a solution, a creature needs multiple streams of informational input about its world. Merriam’s kangaroo rat, for example, has different ways of detecting mojave rattlesnakes; it can exploit visual, tactile, and auditory information. (It avoids rattlesnake strikes by leaping into the air.) Roughly, the rat represents snakes, and not retinal impressions, because it possesses multiple epistemic routes to the same distal source.
Neander criticizes Dretske’s approach on the grounds that it does not confer appropriately distal contents onto representations that are restricted to a single sensory pathway. Suppose a kangaroo rat sees and hears a rattlesnake. Auditory and visual pathways converge to form a multimodal representation, R, of the snake. Now let us focus on an earlier stage of perceptual processing and consider the representation RV that is formed in the visual pathway, which precedes and causes R. We want to say that RV, taken all by itself, represents a property of the snake, and not a property of the retina. Dretske’s theory does not explain why that should be so: “[Dretske’s] strategy will not confer determinately distal contents on the sensory-perceptual representations that belong to the specific epistemic routes that we are in effect abstracting from” (227).
Here is a sketch, though, for how we can modify Dretske’s view to get around this worry (see Garson [Reference Garsonforthcoming b, chap. 12] for discussion). We should say that the visual representation, RV, somehow inherits its distality from the distality of R. The idea is this: if that visual pathway were the only means the kangaroo rat had for detecting rattlesnakes, then RV would not have determinate content. (It might have been that way 5–10 million years ago, around the time kangaroo rats evolved.) But given that the pathway is now, at this stage of evolution, nested in a multimodal representation system that is specialized for detecting snakes, we can say that the representations that are formed within the visual pathway, and that are restricted to the visual pathway, are still about a property of snakes and not a property of retinas. If that is right, then there is no need for a solution that exploits informational asymmetries between the representation’s causes, a solution that is problematic on other grounds.
In summary, Neander has given us a pathbreaking work on naturalizing semantics, one that represents decades of careful reflection on the topic. It outlines an ambitious agenda for teleosemantics that should keep philosophers of biology and mind busy for years to come. It demonstrates the vitality and relevance of philosophy of biology for understanding the mind. Finally, it is essential reading for anyone interested in how the brain makes meaning.
