Abstract

For several decades there has been considerable interest within artificial intelligence and cognitive science in building computer simulations of the processes of scientific discovery, in the spirit of constructing expert systems (automated discovery), or directed toward understanding how human scientists make discoveries. Here, we will consider both lines of inquiry, but with emphasis on the AI aspects.

Our central concern will be with specificity and generality. How do we compare the merits of discovery programs that are efficacious in one domain with those that have generality over several or many? How can we develop a common language to describe systems for different domains, thereby carrying the experience gained in systems into explorations of others?

Most of what we have to say is not peculiar to discovery systems, but applies to all kinds of expert systems. Hence, we will not entirely limit our discussion to discovery tasks. We will focus on computer programs as research vehicles, without implying that all worthwhile research on the subject should result in a computer program.

full paper