In reply to McDermott

McDermott states that “. . . the premise . . . that a lot of reasoning can be analyzed as deductive or approximately deductive is erroneous. ” His arguments actually appear to express the extreme position that, unless deduction can be shown to underlie all the forms of reasoning we wish to implement in artificial intelligence systems, it should not be used at all (at least, not without apology) in their construction or analysis. We counter that many tasks are deductive or have substantial deductive components, and that, even if deduction is inadequate to perform the entire task, deductive techniques and analysis ought to be used in systems where they are applicable. Even if McDermott had offered some Godel-type proof that logic was incomplete for all reasoning, the efforts by Hayes, Hobbs, et al. to formalize commonsense knowledge should still be encouraged. The effort may, like Principia Mathematica, force us to examine our concepts more closely and thus detect deficiencies in our own thinking. It may also, like Encyclopedia Britannica, be forever useful as a compendium of knowledge that can be drawn upon whenever needed. In this relatively short rebuttal, our response will be limited to pointing out advantages and applications of deduction; we are not attempting to cover all the flaws in McDermott’s paper. Rather than attack his points specifically or argue with his reasoning, we shall present our own case.