Artificial intelligence: an empirical science

摘要

My initial tasks in this paper are, first, to delimit the boundaries of artificial intelligence, then, to justify calling it a science: is AI science, or is it engineering, or some combination of these? After arguing that it is (at least) a science, I will consider how it is best pursued: in particular, the respective roles for experiment and theory in developing AI.I will rely more on history than on speculation, for our actual experience in advancing the field has much to tell us about how we can continue and accelerate that advance. Many of my examples will be drawn from work with which I have been associated, for I can speak with greater confidence about what motivated that work and its methods (and about its defects) than I can about the work of others. My goal, however, is not to give you a trip through history, but to make definite proposals for our future priorities, using history, where relevant, as evidence for my views.