The Adaptable Mind: What Neuroplasticity and Neural Reuse Tell Us about Language and Cognition - John Zerilli 2021
Multiple Realization Revisited
Saving Faculty Psychology: Debunking the Argument from Multiple Realization
8.2.1 Preliminary remarks
Two primary empirical arguments have been advanced at one time or another in support of the view that cognitive states are multiply realized. One argument proceeds from evidence of extensive neuroplasticity in the brain. The other proceeds from an account of convergent evolution. In the first part of this section, I shall briefly address these arguments for MR and the existence of MR kinds. In the second part, I provide a conspectus of the most forceful arguments against MR developed in recent years.
8.2.2 Empirical arguments for multiple realization
8.2.2.1 Neuroplasticity
Neuroplasticity has in recent times been thought to provide compelling evidence for the MR of mental states. Shapiro (2004) and Polger (2009) review this evidence and find that it does not provide evidence of MR. Polger (2009, p. 470) concludes that “contrary to philosophical consensus, the identity theory does not blatantly fly in the face of what is known about the correlations between psychological and neural processing.”
As we saw in Chapter 2, there is more than one kind of brain plasticity, including, inter alia, cortical map plasticity and synaptic plasticity. Very roughly, the former occurs when different brain regions subserve the same function at different times in an individual’s history—say, after brain injury or trauma—and it is this plasticity that is most often regarded as supporting MR. “Synaptic plasticity” refers to the strengthening or weakening of connections between neurons, and it is believed to have a role in learning and memory (and quite possibly, therefore, in cortical map plasticity). I shall restrict myself to the first kind here.
Evidence supporting a modest identity relationship between psychology and neuroscience includes the following (Polger 2009, pp. 467—468):
•Cortical maps do not migrate wildly; i.e., they do not simply “ ’jump’ to recruit unused but non-adjacent cortical areas,” and when they appear to do so, it is generally to exploit the structural features common to different sites (see, e.g., the discussion of EB in Chapter 6).
•Recovered functions are frequently suboptimal—genuine MR would require the same psychological state to be underwritten by different neurological states; suboptimality is evidence that the psychological states are in fact different, and therefore evidence of difference underlying difference, not difference underlying sameness, as MR requires.1
•Functional studies of the rewired ferrets whose visual cortex was induced to project into auditory cortex suggest they “developed processing structures—in particular, columnar organization—that is [sic] typical of visual processing” within auditory cortex. Since auditory cortex came to resemble visual cortex (Shapiro 2008, p. 518), this is not evidence that auditory regions learned to handle visual domain tasks while still in their auditory configuration, as genuine MR would require here. After all, auditory regions have specific neural configurations and connection patterns. If these are not preserved when the ferret’s auditory cortex begins processing visual input, it cannot be a case of MR. What we have is a case of sameness underlying sameness, not difference underlying sameness, as MR requires. Put another way, the studies are reporting a genuine case of crossmodal plasticity (or perhaps supramodal plasticity), not MR (see §§ 2.4.2—2.4.3).
In fact, it is unsurprising that neuroplasticity has not been able to deliver up the expected argosy of empirical support for MR, and this for two reasons. Firstly, many of the most sophisticated brain-imaging techniques to date have not been able to yield high-resolution mappings of the neural configurations implicated in rehabilitation after injury. Neither positron emission tomography (PET) nor fMRI measures neural activity and network configuration directly. What they measure, in fact, is blood flow, which can hardly tell us much about whether the functions in question are multiply realized. Even if MR were pervasive, these methods would not yield coherent, interpretable results. Why, then, assume that a recovered function with a new location should have a different realization basis, unless of course there is some other indication that makes it likely (e.g., simultaneous preservation of the old and recovered function in the new area)? Secondly, without standardized criteria for evaluating sameness/difference judgments, how can we be sure that neuroplasticity even speaks to MR? I conveniently neglected this consideration when making the preceding points, and in fact it is only in recent years that more careful attention has been paid to questions of sameness and difference in debates about MR. I take this up in § 8.2.3.4.
8.2.2.2 Convergent evolution
Evolutionary considerations, particularly the idea that convergent evolution is likely to generate psychological similarities as well as behavioral similarities (such as flight in birds and bats) in morphologically unrelated species, have been thought to weigh in favor of MR. But the issue cannot be decided a priori. MR is an empirical hypothesis in the end and must sooner or later come to terms with empirical evidence. In fact, evolutionary considerations might actually tell against MR. Here I cite only one case, detailed at greater length in Bickle (2003, 2010). The process by which short-term memory becomes long-term memory is known as “memory consolidation,” and Bickle cites evidence supporting the likelihood of there being shared molecular mechanisms for memory consolidation across biological taxa as diverse as fruit flies, sea slugs, and mice. One might not think this sort of evidence admits of any far-reaching consequences for human psychology or MR generally, but if the instance is seen to follow from certain “core principles of molecular evolution,” it assumes a larger significance. One such principle holds that the amino acid sequences of specific proteins in functionally important, “constrained,” domains change much more slowly than in functionally less important domains. This principle implies the existence of “universally conserved” molecular mechanisms across distantly related biological species, albeit those found deep down in cellular physiology and intracellular signaling pathways, just as the fly/slug/mouse pathway attests.
In the end, any psychological kind that affects an organism’s behavior must involve the cell-metabolic machinery in individual neurons. In the brain, causally speaking, that’s where the rubber hits the road. But that’s the very machinery [that] tends to be conserved evolutionarily across existing biological species. Random changes to its components, especially to amino acid sequences in its proteins’ constrained domains (almost) inevitably are detrimental to an organism’s survival. (Bickle 2010, p. 258)
Far from being exceptional, “molecular evolution suggests [such mechanisms] should be the rule.” So at least this empirical argument, rooted in considerations of evolutionary plausibility and molecular evolution, predicts that the MR thesis is false at the molecular level, if not at the systems level (see also Hawrylycz 2015, pp. 8—9; Koch 2015, p. 26; and Zador 2015, p. 43).
8.2.3 A conspectus of recent arguments against multiple realization
8.2.3.1 Outline of arguments
The most powerful arguments against the MR hypothesis as presented in the recent literature include:
•the argument from comparative psychology (Bechtel & Mundale 1999);
•the argument from grains (Bechtel & Mundale 1999); and
•the argument from context (Bechtel & Mundale 1999; Shapiro 2000; Polger & Shapiro 2008; Shapiro & Polger 2012; Polger & Shapiro 2016).
8.2.3.2 The argument from comparative psychology
Bechtel and Mundale (1999) appeal to “neurobiological and cognitive neuroscience practice” in the hope of showing how claims that psychological states are multiply realized are unjustified. Essentially, theirs is an argument from success: cognitive neuroscience’s method assumes MR is false, and the success of that method is evidence that MR is false. They argue that it is “precisely on the basis of working assumptions about commonalities in brains across individuals and species that neurobiologists and cognitive neuroscientists have discovered clues to the information processing being performed” (1999, p. 177).
Bechtel and Mundale examine both the “neuroanatomical and neurophysiological practice of carving up the brain.” What they believe this examination reveals is, firstly, that the principle of psychological function plays an essential role in both disciplines, and secondly, that “the cartographic project itself is frequently carried out comparatively—across species” (1999, p. 177), the opposite of what one would expect if MR were “a serious option.” It is the very similarity (or homology) of brain structures that permits generalization across species; and similarity in the functional characterization of homologous brain regions across species only makes sense if the claims of MR are either false or greatly exaggerated. For instance, “[e]ven with the advent of neuroimaging, permitting localization of processing areas in humans, research on brain visual areas remains fundamentally dependent on monkey research” (1999, p. 195). Brodmann’s famous brain maps were based upon comparisons of altogether 55 species and 11 orders of mammals. If MR were true, “one would not expect results based on comparative neuroanatomical and neurophysiological studies to be particularly useful in developing functional accounts of human psychological processing” (1999, p. 178). They also argue that the ubiquity of brain mapping as a way of decomposing cognitive function points to the implausibility of the MR thesis. The understanding of psychological function is increasingly “being fostered by appeal to the brain and its organization” (1999, p. 191), again, the opposite of what one would expect “[i]f the taxonomies of brain states and psychological states were as independent of each other as the [MR] argument suggests” (1999, pp. 190—191).
Aizawa (2009, pp. 500—503) detects a tacit claim in Bechtel and Mundale to the effect that unique cross-species localization of functions in the brain entails their unique realization. This is thought to be a non sequitur. It is true that, strictly speaking, what much of their paper succeeds in showing is the unlikelihood of “multiple localization,” but two things can be said in response. Firstly, the criticism to some extent misses the point of their analysis. Bechtel and Mundale have deliberately opted for functional localization; i.e., brain activity in the same parts or conglomerate of parts across species, as the relevant standard by which to judge the sameness or difference of brain states, and they have done so in deference to cognitive neuroscience practice. Localization is for them the appropriate standard to adopt because it is at the right grain of description (see next section, 8.2.3.3). Secondly, it is not actually easy to police the distinction between localization and realization, for neural localization is an important dimension of neural organization. Aizawa complains that Bechtel and Mundale do not provide independent evidence in support of any such proposition, but, all in all, it does not seem to be a particularly tendentious one.2 It is true that, for all we know, functions could be localized in the same region in closely related species, and yet be organized very differently. But surely we would need some reason to think that this is likely.
A more serious criticism of their argument is that it speaks only to species that are homologous—it might demonstrate that the ubiquity of MR is questionable as long as we restrict our gaze to primates and rodents (i.e., mammals generally, related by common descent), but surely it does not succeed in showing that octopuses and humans realize their psychologies in anything like the same way (Kim 2002; Shapiro 2008). Nevertheless, the arguments grounded in methodological and comparative considerations offer an impressive refutation of MR among the many that have been attempted in the past 15 or so years. Clearly there is some notion of similarity at stake here that, while largely unarticulated in Bechtel and Mundale, ultimately needs to be reckoned with—at least among homologues (cf. Gillett 2002, 2003; Polger & Shapiro 2008).
Next, I present two distinct but related arguments, the argument from grains and the argument from context. Bechtel and Mundale have something to say about both, whereas Shapiro confines his analysis, as far as we shall be concerned, to the import of context alone.
8.2.3.3 The argument from grains
Bechtel and Mundale (1999, pp. 178—179, 201—204) resort to grains as a way of making sense of what they perceive to be the entrenched, almost unquestioning consensus prevailing around MR. They think this can be traced to the practice of philosophers’ appealing to different grain sizes in the taxonomies of psychological and brain states, “using a coarse grain in lumping together psychological states and a fine grain in splitting brain states.” When Putnam went about collecting his various specimens of pain, he ignored the many likely nuances between them. At the same time, he had few compunctions about declaring them different at a neurological level. His contention that pain is likely to be an MR kind can only command our respect if we can be sure that when he was comparing his specimens from a neurological point of view, he was careful to apply no less lenient a standard of differentiation than he applied when comparing his specimens from a psychological point of view. Bechtel and Mundale maintain that when “a common grain size is insisted on, as it is in scientific practice, the plausibility of multiple realizability evaporates.” As their examples of neuroanatomical and neurophysiological practice attest, scientists in these fields typically match a coarse-grained conception of psychological states with an equally coarse-grained conception of brain states. Despite the habit of philosophers’ individuating brain states in accordance with physical and chemical criteria, a habit no doubt originating with Putnam, this is not how neuroscientists characterize them. The notion of a brain state is “a philosopher’s fiction” (1999, p. 177), given that the notion neuroscientists actually employ is much less fine-grained; namely, “activity in the same brain part or conglomerate of parts.”
8.2.3.4 The argument from context
A not-unrelated factor that Bechtel and Mundale think might help explain the prevailing MR consensus in philosophy and the mind sciences is that the hypothesis itself is often presented in a “contextual vacuum.” The choice of grain is always determined by context, with “different contexts for constructing taxonomies” resulting in “different grain sizes for both psychology and neuroscience.” The development of evolutionary perspectives, for instance, in which the researcher necessarily adopts a coarse grain, contrasts with the much finer grain that will be appropriate when assessing differences among conspecifics. “One can adopt either a coarse or a fine grain, but as long as one uses a comparable grain on both the brain and mind side, the mapping between them will be correspondingly systematic.”
Shapiro (2000), more alive than most to the acuteness of the metaphysical problem central to MR, has provided perhaps the most useful and philosophically perceptive treatment of sameness/difference judgments as they relate to MR. He states: “Before it is possible to evaluate the force of [the MR thesis] in arguments against reductionism, we must be in a position to say with assurance what the satisfaction conditions for [the MR thesis] actually are” (2000, p. 636). For him: “The general lesson is this. Showing that a kind is multiply realizable, or that two realizations of a kind are in fact distinct, requires some work” (2000, p. 645), and “[t]o establish [the MR thesis], one must show that the differences among purported realizations are causally relevant differences” (2000, p. 646).
Shapiro’s concerns revolve around what motivates ascriptions of difference, and therefore sameness. The issue is important because the classic intuition pump that asks us to conceive of a mind in which every neuron has been replaced by a silicon chip depends on our ascription of an interesting difference between neurons and silicon chips, apparently even where silicon chips can be made that contribute to psychological capacity by one and the same process of electrical transmission. His answer, too, like Bechtel and Mundale’s, depends ultimately on context—in particular, the context set by the very inquiry into MR itself.
Shapiro (2000, pp. 643—644) argues that “the things for which [the MR thesis] has a chance of being true” are all “defined by reference to their purpose or capacity or contribution to some end.” This is the reason why carburetors, mousetraps, computers, and minds are standard metaphors in the literature of MR. They are defined “in virtue of what they do,” unlike, say, water, which is typically defined by what it is—i.e., its constitution or molecular structure, and accordingly not an MR kind. Genuine MR requires that there be “different ways to bring about the function that defines the kind.” Truly distinct (indeed multiple) realizations are those that “differ in causally relevant properties—in properties that make a difference to how [the realizations] contribute to the capacity under investigation.” Two corkscrews differing only in color are not distinct realizations of a corkscrew, because color “makes no difference to their performance as a corkscrew.” Similarly, the difference between steel and aluminum is not enough to make two corkscrews that are alike in all other respects two different realizations of a corkscrew “because, relative to the properties that make them suitable for removing corks, they are identical.” In this instance, differences of composition can be “screened off.” Naturally, there may be cases where differences of composition will be causally relevant—perhaps rigidity is the allegedly MR kind in question, in which case compositional differences will necessarily speak to how aluminum and steel achieve this disposition. Each case must simply be judged on its own merits, indeed, in its own context (as Bechtel and Mundale might put it). Thus, unlike the two corkscrews identical in all respects save color, which do not count as distinct realizations, waiters’ corkscrews and winged corkscrews are enabled to perform the same task by virtue of different causally relevant properties, and therefore do count as genuinely distinct realizations of a corkscrew, one based on the principle of simple leverage, the other relying on a rack and pinions (see Fig. 8.1).
Figure 8.1 A waiter’s corkscrew (a) and a winged corkscrew (b). Each contributes to the capacity of cork-removal in different ways.
The problem for the MR advocate is this: it is apparent that each of the examples just given involves a straightforward mechanism that renders its classification reasonably intuitive (though I dare say still subject to the odd disagreement!). With the brain, the situation is inestimably more complicated, so why should the MR advocate feel so confident that there is more than one truly distinct way to build a mind, given that we know comparatively little about how brains produce minds in the first place? One might even think that the brain’s functional properties would need to be understood at least as well as the corkscrew’s before one could venture an opinion about whether truly different brains can produce similar minds.3
8.2.4 Conclusion
The primary empirical arguments for MR resting on weak evidence at best, the functionalist has more work to do before she can, by her lights with a clean conscience, look askance at the neural evidence and its implications for cognitive theorizing.