The Evidence for Representations
Ken Aizawa
It is commonly said that one reason to postulate mental representations is in order to explain behavior. I take this reasoning to be an instance of abductive reasoning, not to be identified with inference to the best explanation.[1] Empiricists have long been skeptical of abductive reasoning that is supposed to provide evidence for unobservable entities. This skepticism as applied to mental representations perhaps reached a high point in psychology with B. F. Skinner’s Verbal Behavior:
It has generally been assumed that to explain behavior, or any aspect of it, one must attribute it to events taking place inside the organism. … The difficulty is that the ideas for which sounds are said to stand as signs cannot be independently observed. (Skinner 1957, 5).
There is obviously something suspicious in the ease with which we discover in a set of ideas precisely those properties needed to account for the behavior which expresses them. We evidently construct the ideas at will from the behavior to be explained. There is, of course, no real explanation. (Skinner 1957, 6).
Although the emphasis is not upon experimental or statistical facts, the book is not theoretical in the usual sense. It makes no appeal to hypothetical explanatory entities. (Skinner 1957, 12).
Skinner’s influence has continued to this day, often through the work of Daniel Dennett and those attracted to his idea of representation talk as providing a summary of real patterns in behavior, rather than entities that might explain behavior.
In the first place, my paper, “The Evidence for Representation,” is meant to push back against the empiricist prejudice against abduction. It reviews some of the Nobel Prize winning work of Alan Hodgkin and Andrew Huxley, making the case that it relies upon abduction. The implicit suggestion is that empiricists should not be too quick to dismiss scientific work that scientists value highly. The paper also briefly sketches a theory of the reasoning underlying that work. Although one sample of scientifically acceptable abductive reasoning is far from establishing the reasonableness of just any sort of abductive reasoning anywhere, it does argue against an empiricist blanket dismissal of abduction.
In the second place, the paper draws attention to the scientific practice of “coordinating” abductive reasoning about representations with single-cell neurophysiological recordings. Consider one example from the paper. Based on multiple studies of rats navigating mazes, in 1948, Edward Tolman postulated cognitive maps that he took to explain the navigation behavior.[2] Cognitive maps are often assumed to be representations. In 1971, John O’Keefe and Jonathan Dostrovsky recorded from hippocampal cells that, subject to some qualification, they took to represent a rat’s position and orientation on a platform. In time, these cells came to be called “place cells.”
(Aizawa 2025) provides a much more extensive discussion of Hodgkin and Huxley’s abductive reasoning as well as a more detailed theory of this abductive reasoning. It also contrasts that theory of abductive reasoning with Peircean and “Neo-Peircean” theories of abduction, with Harman’s theory of inference to the best explanation, with Lipton’s theory of inference to the best explanation, and with the New Mechanist theory of “matched interlevel experiments.”[3] Nevertheless, it is very early days in the project of understanding the long neglected role of abductive reason in science.
Returning to unexplored issues in “The Evidence for Representations,” there is much more to be said about the possible limitations on abductive inferences based on behavior. Maybe Hodgkin and Huxley’s work shows that abductive inferences are not always objectionable, but maybe there are limits on their acceptability. If so, what are those limits? Presumably it is not simply that empiricist’s aversion to unobservables. But if not that, then what? More also needs to be said about the scientific interpretation of single-cell recording results. Are these interpretations simply “observations” (as suggested in (Thomson and Piccinini 2018)) or are there abductive inferences based on these results? Aren’t rates of action potentials firings abductively inferred from what appears on an electronic measuring device? Aren’t representations abductive inferred from, say, a correlation of an environmental feature F and a single-cell firing rate R? Finally, nothing at all has been said to develop a theory of how behavioral results are “coordinated” with single-cell experimental work. What does this even mean? That is a black box at this point. It is “early days” in philosophical thinking about these issues, but the first step would seem to be the appreciation of the lay of the land.
Aizawa, K. 2025. Compositional Abduction and Scientific Interpretation: A Granular Approach. Cambridge: Cambridge University Press.
Craver, Carl F, Stuart Glennan, and Mark Povich. 2021. “Constitutive relevance & mutual manipulability revisited.” Synthese 199 (3-4):8807-8828. doi: https://doi.org/10.1007/s11229-021-03183-8.
Harman, G. H. 1965. “The inference to the best explanation.” The Philosophical Review 74 (1):88-95.
Lipton, Peter. 2004. Inference to the Best Explanation. London: Routledge.
Peirce, Charles Sanders. 1878. “Deduction, induction, and hypothesis.” Popular Science Monthly 13:470-482.
Skinner, B.F. 1957. Verbal Behavior. New York, NY: Appleton-Century-Crofts.
Thomson, Eric, and Gualtiero Piccinini. 2018. “Neural representations observed.” Minds and Machines 28:191-235. doi: https://doi.org/10.1007/s11023-018-9459-4.
Tolman, Edward C. 1948. “Cognitive maps in rats and men.” The Psychological Review 55 (4):189-208. doi: https://doi.org/10.1037/h0061626.
[1] Abduction, as I understand it, is inferring some hypothesis H because it explains some evidence E. IBE, however, goes beyond this is promising a theory of how one is supposed to choose between rival explanations, H’, H’’, H’’’, …, of E. Abduction, as I understand it, does not constitute a theory of this hypothesis choice. For more differences between abduction, as I understand it, and IBE, see (Aizawa 2025).
[2] (Tolman 1948).
[3] (Peirce 1878), (Harman 1965), (Lipton 2004), and (Craver, Glennan, and Povich 2021).