Research Program

One strand of effort, mainly philosophical in character, is in epistemology. We all recognize that our knowledge is insecure: new evidence could change not only our general theories about the world, but even our accepted values of physical constants. One response to the insecurity of accepted knowledge is to eschew acceptance: to say that to the extent that we are rational, we do not accept statements, but only assign appropriate probabilities to them.

What are these appropriate probabilities? According to many, they are subjective and arbitrary, except for being real-valued and subject to the axioms of the probability calculus. This strikes me as too skeptical. But the effort to ground probabilities in our knowledge of frequencies demands that we accept, for example, generalities concerning frequencies.

There is thus a complex and delicate nest of issues to be disentangled in a reasonable manner. I am convinced that a reasonable reconstruction can involve interval valued evidential probabilities, and a probabilistic principle of acceptance. But there are many problems to be dealt with.

A second strand of effort, closer to computer science, concerns induction as it is involved in mining databases for general knowledge. Is this possible? Is it possible with the techniques and tools now employed? Work along these lines, with Len Schubert and Greg Carlson, is currently focussed on mining general knowledge from ordinary written documents.

A longer project, mainly in computer science, involves the development of a statistical or inductive assistant. This would be a program that would not merely do statistical computations, but would accept and store potentially relevant general knowledge and provide guidance in the attempt to learn from samples.