Story of Psychology - Morton Hunt [310]
The chief argument in favor of the AI approach to visual perception is that there is no projector or screen in the brain and no homunculus looking at pictures; hence the mind must be dealing not with images but coded data that it processes step by step, as a computer program does.
Fifteen years ago the chief argument against the AI idea was that no existing program of machine vision had more than a minuscule capacity, compared with that of human beings, to recognize flat shapes, let alone three-dimensional ones, or to know where they are within the environment, or to recognize the probable physical qualities of the rocks, chairs, water, bread, or people it was seeing. But since then there have been extraordinary developments in machine vision. Formerly limited to two-dimensional representation, it is now capable of 3-D, and methods of identifying shapes and distances have greatly improved. Robots guided by machine vision now run operations in a great many factories; AI systems using machine vision have guided driverless automobiles across the desert, avoiding obstacles and ravines; security systems can now match a seen face to a photograph of that face, and so on.
Having said all that, it remains true that machine vision has only a very limited capacity, compared with that of human beings, to recognize all sorts of objects for what they are; it doesn’t understand, it doesn’t know, it doesn’t feel. Basically, that’s because it isn’t hooked up to the immense information base of the human mind: its vast store of mental and emotional responses built in by evolution, its immense accumulation of learned meanings of perceptions, its huge compilation of interconnected information about the world. As remarkable as the achievements of the designers of machine vision are, their work has led to a greater understanding of how to make machine vision work but not to a deeper understanding of how human vision works.
The other school of thought about how cognitive perceptual processes work has long relied and continues to rely on laboratory studies of human thinking rather than machine simulations of thinking. This view, going far beyond the Helmholtz tradition that perception is the result of unconscious inference from incomplete information, includes conscious thought processes of other kinds. Its leading exponent in recent years was Irvin Rock (1922–1995) of the University of California at Berkeley. His book, The Logic of Perception, was described in the Annual Review of Psychology as “the most inclusive and empirically plausible explanation of perceptual effects that seem to require intelligent activity on the part of the perceiver.”79
Rock, though an outstanding perception psychologist, was far from outstanding in his early undergraduate years; in fact, in an intellectual family he was the black sheep. But during World War II his unit was dive-bombed by enemy planes, he felt sure he would be killed, and “I vowed to myself,” he said, “that if I survived I would try to do more with my life than I had until then.”80 After the war he became a top-notch student. He began graduate school in physics but switched to psychology when he realized that there was greater opportunity in that young field for a significant contribution