Vision, Learning, and Development 1 Argument and Apology 3 What I Mean by Vision, Learning, and Development 3.1 Vision

It seems to be a common feeling that animals learn to see, and this feeling, together with the re-emergence of computer learning paradigms that mimic many forms of human learning, has raised hopes that learning is the key to the computer vision problem. Indeed, it seems clear that Nature does not \program" all our visual capabilities into the genome, and we certainly know that programming a computer with a closed-form solution to the vision problem is a daunting task. This aim of this informal and elementary report (basically a term paper) is to cast doubt on the idea that biological systems learn to see. The complex process of development, beginning at fertilization and ending with a mature individual, could be considered to have genetic (\nature") and learning (\nurture") processes as logical endpoints or opposite poles. This report mostly considers what goes on between those endpoints, and is meant to raise the possibility that some of the least understood processes in biology are responsible for visual capabilities. In the mid to late 1980s, \computer learning" emerged from a decade of neglect and became one of the hot research areas of the 1990s. For the purposes of this paper and to save argument, I am willing to stipulate that work in this area has largely succeeded, by which I mean and that several good computational models of learning exist. I am concerned here only with one putative application of learning, viz. the application of learning in \learning to see". The idea of learning to see is popular with us computer vision researchers for two reasons. First, we have not been able to solve the vision problem in closed form, and the idea of explicitly programming in (into some genome, as it were) all the necessary visual capabilities seems increasingly daunting. Second and therefore, there seems to be a strong urge among us to believe that successful (i.e. biological) vision systems learn to do vision and therefore that computer learning can play a central, helpful role in the development of computer vision systems. This report is to suggest that the sort of learning that computer scientists have been modeling, and indeed that which psychologists usually study, may be inadequate to learn vision. The specialized neural structures that support vision certainly do develop in conjunction with experience of one sort or another. However, development is a powerful and little-understood set of interacting processes that …