Subpersonal, non-conceptual content in Marr’s theory of vision

Basileios Kroustallis

e-mail:

Non-conceptual content seems to be an exception to the representation by means of concepts, expressed in propositional attitudes: it involves as a necessary condition that the subject does not necessarily have to possess the concepts normally characterizing this content (Cussins, 1990; Crane, 1992). Perception seems to justify this notion; subjects see without necessarily believing what they see (the case of illusions), and this is more than having a certain sensation; it is a kind of representation. Non-conceptual content is a different means of the mind representing the world.

Although the generic structure of this content remains to be articulated and defended (see, for example, Peacocke, 1992), its possible extension is a separate issue. Non-conceptual content in the form of ‘subpersonal’ content has been attributed to implicit mental processes which cognitive science posits to a representing subject. The name has been derived from Dennett’s (1969) ‘subpersonal’ and ‘personal’ levels of explanation, and it denotes the semantic analogue of unconscious cognitive processes; David Marr’s (1982) computational theory of vision, involving successive stages of specialized mental processes, is usually considered an example of subpersonal, non-conceptual content.

However, even if the existence of non-conceptual content is assumed, there are independent problems for subpersonal visual content that need to be addressed. Two main criticisms have been raised: first, this kind of implicit, unconscious processes that the theory postulates may be only syntactically, and not semantically evaluable. McDowell (1994) calls the latter ascription ‘as-if’ content. Similarly, Egan (1996) attests that even in the computational level of Marr’s theory, where environmental constraints are discussed, this is rather a ‘function-theoretic’ characterization, and not a means of subpersonal representation.

Although the latter point seems justified, Marr’s theory is primarily explained in his algorithmic level, and in the work of specific computational mechanisms. Bermudez (1995) talks of a functional role that characterizes processes semantically according to environmental constraints: specific processes are selected or excluded from a set of possible computational solutions. However, this functional role may work differently: the same outcome results from two consecutive processes, but only the second of them is directly characterized according to a certain constraint. The detection of zero-crossings (sudden light intensity changes in an image which may denote object boundaries) comes in two stages: first, the second derivative of a filtered image is computed and certain segments are detected. In the second stage, the results of different image filters are considered, and the image segments with the same value in the appropriate filters are selected; the ‘new’ segments are meant to be the edges of a visual scene, according to the constraint of spatial coincidence. Although the second stage is no more than a comparison of previously measured segments, the same elements are now characterized as ‘edges’. The example can be extended to the whole of Marr’s theory, and shows that the functional role does not consist in the exclusion of functionally irrelevant states, but in the different characterization of the same states under successive processes.

The second criticism against this kind of content has been made prominent by McDowell (1994). He asserts that since vision at some point reaches a personal level and observers consciously see an object as such, there should be a connection between subpersonal and personal conceptual content; but all there can be is a relation between different subpersonal parts. The frog’s eye may inform the frog’s brain, but not the frog.

This conception assumes that all non-conceptual processes in vision are subpersonal; nevertheless, the subpersonal is not coextensive with the non-conceptual in Marr’s theory of vision. Subpersonal, early stage processes are both non-conceptual and unconscious; however the final stage of non-conceptual (but conscious) content is the construction of the 2½D sketch, a three-dimensional viewer-centered representation. Furthermore, this does not interact with conceptual content in a bottom-up fashion, the latter being at the receiving end of non-conceptual processes; rather, top-down activation of 3D memory stored object models instantiates one of them, according to the information provided by the non-conceptual 2½D sketch representation. There is an issue of how the subpersonal interacts with the personal and produces awareness before the construction of the 2½D sketch, but this is not ipso facto a problem of content.

Marr’s theory may not be the only explanation of human vision in its specific stages; however, the issue of cognitive penetration that it raises (bottom-up vs top-down in vision) is widely supported by researchers (cf. Pylyshyn, 1999; Barsalou, 1999). On the other hand, non-conceptual content in general may face different problems, such as the nature and its function in the explanation of behavior (Bermudez, 1995). Nevertheless, it is important to ensure that subpersonal content is not loaded with additional problems that hamper its acknowledgment as a kind of non-conceptual content.

References:

Barsalou, L. (1999). Perceptual symbol systems. Behavioral and Brain Sciences, 22, 577-660.

Bermudez, Jose Luis (1995). Nonconceptual content: from perceptual experience to subpersonal computational states. Mind and Language, 10, 333-369.

Crane, Tim (1992). The nonconceptual content of experience. In Tim Crane (ed.), The contents of experience: essays on perception, Cambridge: Cambridge University Press.

Cussins, Adrian (1990). The connectionist construction of concepts. In M. Boden (ed.), The philosophy of artificial intelligence, Oxford: Oxford University Press.

Egan, Frances (1996). Intentionality and the theory of vision. In Kathleen Akins (ed.), Perception. New York: Oxford University Press.

Dennett, Daniel (1969). Content and consciousness. London: Routledge & Kegan Paul.

Marr, David (1982). Vision. New York: W.H. Freeman.

McDowell, John (1994). The content of perceptual experience.Philosophical Quarterly, 44, 190-205.

Peacocke, Christopher (1992). A study of concepts. Cambridge, MA: The MIT Press.

Pylyshyn, Zenon. (1999). Is vision continuous with cognition? The case for cognitive impenetrability of visual perception.Behavioral and Brain Sciences, 22, 341-365.