Abstracts

Inference is as Inference does: The Evolution of Conscious from Unconscious Inferential Processes

ALISTAIR M. C. ISAAC University of Edinburgh

The concept of an “unconscious inference” has played a contentious role in the history of psychology. The basic idea is that certain low-level, typically perceptual processes are best understood as inferential in character. These processes are “unconscious” in the sense that we are not aware of them as they occur, and often they are encapsulated or modular, in the sense that they are (at least partially) impervious to top-down influence and can be given mechanistic explanations. At various points in the 20th century, this notion of “unconscious inference” has been criticized as theoretically vacuous, unfalsifiable, or an abuse of terminology. The aim of this talk is to clarify the sense in which unconscious inferences indeed constitute inferences and to argue for their theoretical significance in understanding the evolution of “high-level” human inferential capacities, as exemplified for instance by scientific inference.

In the late 19th century, Helmholtz and Peirce independently advanced the idea that perception should be interpreted as the outcome of an inferential process. Helmholtz was motivated by the demonstrable fact that stimulation of the sense organs underdetermines the nature of the stimulus. Peirce was motivated by considering the logical structure of perception—the causes of a stimulation cannot be deductively derived from its effects. The import for both authors was twofold: first, the relationship between input and output in an act of perception is the same as that between premises and conclusion in an (abductive or inductive) inference; second, this implies that the output, i.e. our perceptual experience of the world, has the same epistemological status as the outcome of other inferences, e.g. scientific theories.

This analysis was criticized at various points in the 20th century on two grounds. First: countenancing both unconscious sensations and unconscious inferences means explanations can be generated arbitrarily and thus the theory is unfalsifiable (Köhler, 1913). Second: these automatic processes do not have any of the features of conscious inference, thus calling them “inferences” is unilluminating and/or vacuous (Hamlyn, 1977; Hatfield, 2002). Nevertheless, the interpretation of low-level perceptual processes as inferences has again become popular, driven now by a unified analysis of low-level and scientific inferences as instances of Bayesian update (Friston, 2003; Gopnik, 2012; Clark, 2013).

I argue that low-level perceptual processes are structurally equivalent to inferences. Furthermore, this same structure is found in animal perception, demonstrating a continuity from animals to humans in low-level inferential capabilities. Contra criticisms of this approach, I argue that our best strategy for providing a scientific analysis of high-level inferences is by analogy with mechanistic theories of these better-understood low-level processes. If this project succeeds, it will provide evidence for an evolutionary continuity between animal perception and high-level human inference. Until such an approach is disproven, it at least provides a unified framework within which to compare and contrast human and animal inferential capacities.