The Massive Redeployment Hypothesis - Neural Reuse and Recycling

The Adaptable Mind: What Neuroplasticity and Neural Reuse Tell Us about Language and Cognition - John Zerilli 2021

The Massive Redeployment Hypothesis
Neural Reuse and Recycling

Neural reuse theories comprise what Anderson describes as “an emerging class of theories,” which, “taken together . . . offer a new research-guiding idealization of brain organization” (Anderson 2010, p. 246). Anderson’s own hypothesis builds on the assumption that evolution might prefer the reuse of neural circuitry over the development of new circuitry de novo (Anderson 2010, p. 246). From this assumption, three predictions are thought to follow, the most obvious being neural reuse itself. “A typical brain region [should] support numerous cognitive functions in diverse task categories.” Second, older brain areas should, ceteris paribus, be reused more than newer ones, because “having been available for reuse for longer,” they are likelier candidates for integration into recently evolved functions.2 Third, recently evolved functions should be more distributed than older ones since it should on the whole prove easier to utilize available circuits than to devise special-purpose circuitry afresh, “and there is little reason to suppose that the useful elements will happen to reside in neighboring brain regions.” Conversely, “a more localist account of the evolution of the brain would . . . expect the continual development of new, largely dedicated neural circuits” for every cognitive innovation or significant increase in cognitive power.

Anderson has tested these predictions in a number of studies, with conspicuous success (2007a; 2007c; 2008). For instance, the typical cortical region was found to be implicated in fully nine domains extending from action, vision, and audition through language, mathematics, memory, and reasoning. This illustrates an important feature of reuse; that is, the possibility (in principle) of congruously overlapping regions—just the same circuits exapted for one purpose can be exapted for another, provided sufficient intercircuit pathways exist to allow alternative arrangements of them. The same parts put together in different ways will yield different functional outcomes, just as “if one puts together the same parts in the same way one will get the same functional outcomes” (Anderson 2010, p. 247, my emphasis) (see Fig. 3.1).


Figure 3.1 Two cognitive functions indicated by solid and dashed lines, organized in the top figure the way that an anatomical account of modularity would predict, and in the bottom figure in accordance with how neural reuse sees the matter. Anatomical modularity maintains functional dissociability and localization for gross or high-level functions with few if any overlapping units. Reuse suggests overlapping units that form different patterns of connection.

(Michael L. Anderson, “Neural reuse: A fundamental organizational principle of the brain,” Behavioral and Brain Sciences, Volume 33, Issue 4, August 2010, pp. 245—266. Reproduced with permission of Cambridge University Press.)

Regarding the second prediction, that older areas are more likely to be reused than recently evolved regions, if we make the simplifying assumption that older areas lie at the back of the brain, Anderson’s results confirm the expectation. Anderson reports a negative correlation between the position of a brain region along the Y-axis in Tailarach space3 and the number of tasks that activate the region. The results were replicated using different data sets, with Anderson evaluating them in this vein:

Although the amount of variance explained in these cases is not especially high, the findings are nevertheless striking, at least in part because a more traditional theory of functional topography would predict the opposite relation, if there were any relation at all. According to traditional theories, older areas—especially those visual areas at the back of the brain—are expected to be the most domain dedicated. But that is not what the results show. (2010, p. 247)

As for the third prediction, that recently evolved functions ought to generate more distributed patterns of activation than older ones, Anderson’s (2007a) findings suggest that language could well be the paradigm, supported by more distributed activations than visual perception and attention, and indeed than any other domain that was tested (Anderson 2008). Results such as these showing widely scattered activations across the brain for putatively late-developing functions are incidentally consistent with the degree of specialization for local circuits that neural reuse actually presupposes. For neural reuse is of course consistent with a certain kind of specialization: as the very word “redeployment” suggests, it presupposes the existence of comparatively fixed neural circuits whose functional contribution may be preserved across multiple task domains. The metabolic costs of maintaining long-distance connections would presumably encourage the reuse of local flexible (“poly-functional”) circuits, if any were around; “[t]hat this is not the observed pattern suggests that some functionally relevant aspect of local circuits is relatively fixed” (Anderson 2010, p. 247; but cf. Anderson 2014, pp. 15—16, 104).4 Anderson’s earliest formulations of the redeployment hypothesis accounted for this fixity by introducing an important distinction, following Bergeron (2007, 2008), between stable, low-level computational “workings” (or cortical “biases”) and diverse higher level cognitive “uses.” Workings are represented in the numbered units of Fig. 3.1, while uses are represented by the functional composites formed from these units. Workings are really very tiny regions of cortex that make a specific computational contribution to higher-level cognitive tasks or “uses.” We might say that workings represent an essential functional contribution across all task categories, considered in isolation of neural context (although Anderson has since moved away from essentialism), and that uses are the higher level cognitive functions enabled by the composite of several workings. (More on this in Chapter 4.)

Stable structure for local circuits is a feature of a closely related account of neural reuse, one that posits reuse or recycling as a developmental solution (in contrast to Anderson, for whom reuse was, originally, primarily an evolutionary solution). How are readily transmissible cultural practices whose phylogenetic emergence is too recent for evolutionary hardwiring to explain, such as reading and arithmetic, to be neurophysiologically accounted for? Early developmental neuroplasticity might be one way, but in supposing that local circuits might be too rigid for the effects of experience to overcome, Dehaene (2005) gives priority to “neuronal recycling.”

Of course, neuroplasticity is not something either Dehaene or Anderson would wish to deny. Dehaene goes as far as accepting (as I think one must) that novel uses that depart significantly from existing cortical biases cannot simply be exapted from them: a high-level use that places a significant cognitive burden on existing circuits, themselves better suited for other uses, must in the end disrupt those circuits and the alternative uses to which they might be put. More cognitively demanding functional acquisitions therefore require more neuroplasticity. This brings us to a potentially thorny issue. Just what is the relationship between reuse and plasticity? There might seem to be a tension between the fixity necessary to get reuse off the ground on one hand, and the plasticity necessary for reuse to play an interesting role in learning and evolutionary novelty on the other. In fact there is no real problem here (Anderson & Finlay 2014). Anderson describes neural reuse as a change in use without a change in working, and plasticity as a change in use resulting from a change in working (Anderson 2010, p. 297). There is no real problem here because some forms of neuroplasticity (such as Hebbian synaptic plasticity) do not require flexible units before they can effect a change in use, given that they involve only adjustments to connection strength (“a change in use without a change in working”); besides that, neural units are not quite so “fixed” as Anderson’s own (2010) remarks suggest, allowing for more drastic forms of neuroplasticity (such as synaptogenesis and the like) to partly override the natural biases of particular brain regions (“a change in use from a change in working”). I will revisit these themes in Chapter 6 (see p. 93).

It might be just as well to point out one other respect in which the story of neural reuse is compatible with the known biophysical constraints and possibilities of neural circuits. Neural reuse is really an ideal solution to what might be called the scaling problem (Zador 2015, p. 43; Bullmore & Sporns 2012, pp. 337—339). “The scaling problem” refers to the dilemma that, as the number of neurons increases, undoubtedly conferring advantages in the form of increased behavioral flexibility and intelligence, the number of neurons that must be connected before such advantages can materialize grows quadratically larger. Thus, in a small 10-neuron circuit, only 100 connections are required, but in a larger circuit consisting of perhaps 100 million neurons, anywhere up to a thousand billion connections might be required. It is not easy to see, from the point of view of engineering and design, how an ever larger brain can be wired up efficiently when the number of neural connections required soon becomes immense. This consideration has actually been played as an argument in favor of modularity, but it could just as well be pressed into the service of neural reuse, which delivers fixed low-level cognitive workings that operate autonomously in something like the way of traditional modules (see § 5.1).