8.10: The Bounds of Cognition

The purpose of this chapter was to introduce Pylyshyn’s (2003b, 2007) theory of visual cognition. This theory is of interest because different aspects of it make contact with classical, connectionist, or embodied cognitive science.

The classical nature of Pylyshyn’s theory is found in his insistence that part of the purpose of vision is to make contact with perceptual categories that can be involved in general cognitive processing (e.g., inference and problem solving). The connectionist nature of Pylyshyn’s theory is found in his invocation of artificial neural networks as the mechanisms for assigning and tracking indices as part of early vision. The embodied nature of Pylyshyn’s theory is found in referential links between object files and distal objects, the use of indices to coordinate vision and action, and the use of indices and of referential links to exploit the external world as a scaffold for seeing and visualizing.

However, the hybrid nature of Pylyshyn’s theory of visual cognition presents us with a different kind of puzzle. How is this to be reconciled with Pylyshyn’s position as a champion of classical cognitive science and as a critic of connectionist (Fodor & Pylyshyn, 1988) and embodied (Fodor & Pylyshyn, 1981) traditions? The answer to this question is that when Pylyshyn writes of cognition, this term has a very technical meaning that places it firmly in the realm of classical cognitive science, and which—by this definition—separates it from both connectionist and embodied cognitive science.

Recall that Pylyshyn’s (2003b, 2007) theory of visual cognition was motivated in part by dealing with some of the problems facing purely cognitive theories of perception such as the New Look. His solution was to separate early vision from cognition and to endorse perceptual mechanisms that solve problems of underdetermination without requiring inferential processing.

I propose a distinction between vision and cognition in order to try to carve nature at her joints, that is, to locate components of the mind/brain that have some principled boundaries or some principled constraints in their interactions with the rest of the mind. (Pylyshyn, 2003b, p. 39)

The key to the particular “carving” of the system in his theory is that early vision, which includes preattentive mechanisms for individuating and tracking objects, does not do so by using concepts, categories, descriptions, or inferences. Time and again in his accounts of seeing and visualizing, Pylyshyn describes early vision as being “preconceptual” or “non-conceptual.”

This is important because of Pylyshyn’s (1984) characterization of the levels of analysis of cognitive science. Some of the levels of analysis that he invoked—in particular, the implementational and algorithmic levels—are identical to those levels as discussed in Chapter 2 in this volume. However, Pylyshyn’s version of the computational level of analysis is more restrictive than the version that was also discussed in that earlier chapter.

For Pylyshyn (1984), a computational-level analysis requires a cognitive vocabulary. A cognitive vocabulary captures generalizations by appealing to the contents of representations, and it also appeals to lawful principles governing these contents (e.g., rules of inference, the principle of rationality). “The cognitive vocabulary is roughly similar to the one used by what is undoubtedly the most successful predictive scheme available for human behavior—folk psychology” (p. 2).

When Pylyshyn (2003b, 2007) separates early vision from cognition, he is proposing that the cognitive vocabulary cannot be productively used to explain early vision, because early vision is not cognitive, it is preconceptual. Thus it is no accident that when his theory of visual cognition intersects connectionist and embodied cognitive science, it does so with components that are part of Pylyshyn’s account of early vision. Connectionism and embodiment are appropriate in this component of Pylyshyn’s theory because his criticism of these approaches is that they are not cognitive, because they do not or cannot use a cognitive vocabulary!