Philosophy and the Brain Oxford (1988) by J.Z. Young

Author:J.Z. Young
Language: eng
Format: epub
Published: 2012-12-23T16:00:00+00:00

25. Some theories of vision

i. How can visual representations best be studied?

Workers in artificial intelligence (A.I.) and in cognitive psychology have approached these problems of vision from a theoretical point of view. They ask ‘What are the functions of vision? What needs to be represented? How should it be represented?’43 The answers to these questions are far from simple, and workers in A.I. often find it necessary to express them in mathematical form or to test them by computer models. Vision is a very complicated process and sophisticated analysis is needed to investigate it. The studies of these workers are providing the groundwork that will be needed if we are ever to understand the visual centres of the brain. Such work is also the basis of the attempts to make machines that will perform the visual tasks that are needed in order to allow robots to ‘see’ as they go about their work.

When workers first started to try to make machines that would recognize objects, they were unpleasantly surprised to find that enormous amounts of computation were needed. Moreover, as Sutherland has emphasized, it is very hard to identify the features of images by which objects are recognized.44 Early attempts to make visual recognizing machines have had to use conventional sequential computers in which data are fed very fast through one central processor. The brain does not work like this: it uses many parallel channels, which exchange information through a very complex communication network. Electrical engineers are now beginning to devise such systems, helped by the great decrease in the cost of logic circuits. Such systems make use of arrangements similar to those of the visual cortex, where there is a series of representations of an image, in each of which the neurons extract a different feature, say contour, colour or ocular disparity. The general plan is that computation between the units of each area can then test hypotheses about the probable nature of the object represented by the image. It is not clear how identification of it is then to be achieved, either by the machine or a brain.

Such difficulties have been among the problems attacked by students of artificial intelligence. The issues involved can be described as largely theoretical and philosophical, and the studies illustrate how these precede and overlap the practical and scientific questions. We may illustrate them by describing some results produced by the group at the Massachusetts Institute of Technology including the late D. Marr, T. Poggio, and M. K. Nisihara. We shall follow the account given by David Marr in his book Vision and supplement this by some criticisms and suggestions made by A. Sloman of Sussex University.

ii. Marr’s analysis of vision

Marr begins by emphasizing that vision like any other complex process has to be understood at different levels. The principles involved in processing the visual information will not immediately appear from studying the actions of individual neurons, any more than the behaviour of the gas in a bottle is best studied by solving thermodynamic equations for the particles in it.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.