Making Believers out of Computers

To someone unfamiliar with Artificial Intelligence, all of the activity that is taking place at this IJCAI conference must be very puzzling. And the current fuss about what is called a "knowledge-based system" (KBS) must be a total mystery. For one thing, there is an amazing amount of talk about knowledge. There are even advertisements in The AI Magazine reminding us of the value of knowledge. How bizarre! A stranger to AI must wonder who exactly it was that thought that lack of knowledge was better. On closer examination, the idea of a KBS is not totally vacuous. The idea is not just to construct systems that exhibit knowledge, but to represent that knowledge somehow in the data structures of the program, and to have the system perform whatever it is doing (diagnosing diseases, controlling a power plant, explaining its behaviour, or whatever) by manipulating that knowledge explicitly. Now as anyone who reads the ads can tell you, we can achieve expert-level performance by following these principles and applying vast amounts of domain-dependent knowledge. While not totally vacuous, this strikes me as a form of buck-passing. What we have done, in effect, is to replace the problem of doing computer medical diagnosis (say) by the problem of getting a computer to apply large amounts of domain-dependent knowledge. Unfortunately this is a type of information processing that we know very little about. How is it possible at all? How does the behaviour of a KBS vary with the amount of knowledge and the kind of knowledge made available to it? What are the limits of KBSs? These are