GUEST EDITOR'S INTRODUCTION: HERE'S THE AI

Knowledge representation (KR) is sometimes called the glue that binds A1 together. It is about capturing meaningful descriptions of the world with formal marks and seems to be an easy notion to understand. There is a problem, however. Leaders in the area of knowledge representation (in particular, nonmonotonic logic) have reiterated that the literature has become highly inaccessible to the lay A1 practitioner as well as to quite a few KR experts. It seems difficult, without considerable investment of time, to at once grasp the state of the art yet also understand foundational issues. Can the theory be presented to the implementer in a meaningful way? That is, can we say exactly what’s in a link? That means being able to specify details and being precise about how this will be done. There isn’t agreement on this question. The wellknown division between “scruffies” and ‘‘nests," (performance theories versus competence theories) exists in A1 but also in vision, linguistics, and philosophy. Of course, nothing is wrong with procedural “ad hocery” but it is difficult to generalize. Some members of the community have argued in defense of logic as the foundation for artificial intelligence while others have responded with sound and complete critiques of pure reason. Proceduralists still argue that people are bad logicians yet much more successful in the real world than theorem provers. A recent difference of opinion historically concerns the kind of formalism that’s in the links. Though the logicist tradition can be traced to AI’s beginnings, the probabilist trend is relatively new. Observe how few entries 011 probabilistic methods appear in the 1980 Handbook of Artificial Intelligence. Perhaps these approaches were circumscribed out because the information necessary to assign probabilities usually isn’t available to programs or programmers or because there are difficulties handling statements with quantifiers. In any case, a spirited debate between the logicist and probabilist schools during the mid-80s spoke of the logic mafia, the formal straightjacket, the inability of humans to reason well with probabilities and the intractability of most interesting probabiiistic calculations. Most of these polemics have now subsided--performance versus competence, logic versus probability, probability versus other uncertainty formalisms, Bayesian probability versus other interpretations of probability-but recently criticism and perhaps some cynicism has been directed against the entire KR enterprise. Asking “where’s the AI?”, some suggest that this study will produce computationally sophisticated philosophers of marginal utility. This is an unfortunate view for at least two reasons. One, philosophy has much in common with the A1 goals of automating human performance. Furthermore, it has a history dating back to antiquity and a wealth of experience to offer. Two, the knowledge representation enterprise has been very productive with respect to explaining reasoning patterns studied widely in the 80s. It is timely therefore that someone write clearly and fundamentally in defense of philosophy as it relates to artificial intelligence. The Taking Issue forum in this issue of Computational Intelligence follows in the spirit of Drew McDermott’s “Taking Issue with Logic” forum’