Reply to Mace and Roskies
Caitlin Mace and Adina Roskies (hereafter, M&R) argue that the identification of vehicles in neuroscience depends intimately on the prior assignment of content, and so content cannot be relegated to an extra-theoretical gloss. Identifying neural vehicles is notoriously difficult and, they claim, subject to indeterminacy.
To identify computational vehicles, experimenters look for neural signals whose activity appears to be correlated with some experimental variable taken to be the representational content, such as the onset of a cue, a stimulus feature, or some aspect of behavior.
I agree that identifying vehicles requires the theorist to look for an ‘experimental variable’ such as a stimulus feature or aspect of behavior with which neural activity is correlated. But these experimental variables should not be identified with representational content, whether content is construed as essential to the theory, or, as in my view, a heuristic gloss. These experimental variables are objective determinants of content, not content itself. M&R mischaracterize my view when they say “… Egan will say the relation between the vehicle and experimental variable is not substantive, which we take to mean is outside the theory proper…” Not so. Characterizing the relation (typically a causal relation) between vehicle and experimental variable is an essential part of the theory. Whether these experimental variables fully determine content or require supplementation by appeal to such pragmatic considerations as the theorist’s explanatory goals is the issue separating thoroughgoing representational realists from a content pragmatist such as myself. M&R have nothing to say about this crucial question.
I agree with M&R that identifying representational vehicles is a considerable theoretical achievement. They describe the process as follows:
The vehicle is provisionally taken to represent that experimental variable and some kind of computational transformation is posited. There is an iterative refining via continuing negotiation of the vehicle, the representational content ascribed, and the function attributed. This iterative process may change the experimenter’s conception and/or identification of the vehicle itself…
But there is nothing here that is inconsistent with my account. Indeed, in the book I argue that a provisional content attribution can serve as a placeholder in an incompletely developed computational theory, prior to the specification of the structures and processes that make up what I am calling the “theory proper” (37-38). For example, a visual theorist may hypothesize that cells in primary visual cortex respond to, and so represent, edges, and then go on to investigate which structures are involved (the vehicles) and attempt to characterize precisely their stimulus conditions. Once that is done the theorist may revise the initial content attribution – the representational characterization – to more accurately gloss the causal process characterized by the theory. The fact that a provisional content attribution played a role in the discovery of the vehicle, enabling the theorist to formulate hypotheses about the causal roles of the structures she is investigating, does not entail that content is an essential property of that vehicle or that it should be considered part of the theory proper.
M&R have no interest in considering how the various functions served by content attribution in my account motivate my content pragmatism. Rather, they accuse me of being disingenuous, of relegating content to a gloss simply to avoid the notorious naturalization problem. Their ad hominem charge is unfounded. M&R ignore my detailed argument that content attribution in cognitive science depends, in part, upon appeal to theorists’ explanatory interests and goals. It is these pragmatic elements that supplement the objective determinants of content that motivate treating content attribution as a gloss. This is not a “verbal maneuver,” it is a substantive issue. In fact, I argue in the book that an explanatory gloss accompanying a theory and serving various expository and heuristic functions is a general feature of scientific theorizing and not an ad hoc feature of my deflationary account of computational cognitive science. An important motivation for my distinction between theory and gloss, one that M&R overlook, is that theorists want their theories to have universal application, i.e., to hold in environments or explanatory contexts where a particular intentional gloss, especially one that may have figured in the theory’s construction, would have no application.
In conclusion, there is no reason to think that the account is unstable, at risk of collapsing either into thoroughgoing representational realism on the one hand or pragmatism about both vehicle and content on the other. The two components of my account – vehicle realism and content pragmatism – are independently motivated.