Comments on An inquiry into computer understanding by Peter Cheeseman

I share Peter Cheeseman’s conviction that probability ideas have much to offer workers in artificial intelligence. I fear, however, that the tone of Cheeseman’s article will alienate rather than persuade those who are not yet fond of probability. I myself am offended by Cheeseman’s dogmatic dismissal of non-Bayesian methods of probability judgment. In the following comments, I will spell out some of the many points where I disagree with Cheeseman, and, more importantly, I will try to supply a broader perspective on the issues and problems that he raises. Most of the issues have been the subject of long debates, and most of the problems have been the subject of considerable study. Readers who are encountering the issues and problems for the first time deserve some signposts pointing to the existing literature.