merrill (12/02/82)
Having just discovered the USENET I am quite amused to see a raging and ongoing discussion on the Turing test with Searle prominently mentioned. Though I don't believe I have anything startlingly original to contribute to this, I do have some remarks I should like to make since I have at different times dwelt upon the two sides of the professional fence here. By this I mean to say that for close to 10 years, and until approximately 8 months ago I taught philosophy (logic, philosophy of science, philosophy of language, and even artificial intelligence) at both the undergraduate and graduate level. While a philosopher I was a bit embarrassed by the likes of Searle and (even worse) Dreyfus. They take a very dangerous position in that they offer a priori arguments about what computers CAN'T and COULD NEVER do. Almost all of us would agree that there is not now a computer (or better, a "system" -- hardware/software) that is what we would really call intelligent. But the arguments go beyond this. They hope to show that no POSSIBLE computer could ever be reasonably classed as intelligent. This is a particularly tricky course since we do not currently understand enough about human intelligence to state even a reasonable criterion of intelligence. Thus the conclusion being pushed is roughly "No matter what we eventually understand intelligence to be, computers won't have it!" This, I submit, is a silly thing to say, and no rational person should subscribe to it. Without going on at much greater length, let me state briefly what I think the appropriate view of all this is. Computers (and their software -- let's not continue to make a distinction for now) will become more and more sophisticated, more and more able to "fool" us about their intelligence. Simply as a practical matter of dealing with such systems we will adopt a model according to which they are intelligent -- it will just become too cumbersome to insist that they are not. We will ACT as though they are intell- igent because the model of intelligence we adopt will be the best way to account for their behavior, to guide our further research, etc. As this happens (or when it happens, as you like) we may say either: (a) Computers have become intelligent or (b) Our concept of intelligence and the meaning of the word 'intelligent' have changed such that we now (correctly) say that computers are (or at least can be) intelligent. Some will of course ask "But are they REALLY intelligent?" However, unless come clear sense (i.e., a precise criterion) can be associated with 'REALLY intelligent', AND we can be shown what talk of "REAL" intelligence gains us over simply talk of (simple?) intelligence, this question will be nonsense (in the sense that the 'REALLY' just won't have any identifiable meaning). Notice that my position differs significantly from those of my opponents. It is an empirical view -- time will tell whether it is true. Theirs is not. Indeed, their position can easily become (if it is not already) empirically meaningless if they deny the possibility of specifiying criteria by means of which we can determine whether an arbitrary object is intelligent. We can argue endlessly about what it MEANS to be intelligent, but we seem to be gathering a lot of inductive evidence that whenever we CAN say what (a kind of?) intelligence amounts to we can design a system that satisfies all objective criteria regarding that kind of intelligence. In short, then, my view is that in the future those who cling to a view of intelligence according to which computers are not intelligent will be clinging to an antiquated view devoid of any empirical content. It will become an article of faith rather than a meaningful description of the differences between humans and machines. But as I say, time will tell. Apologies about the length and rambling style of this, but I have grown weary of a priori arguments concerning computer intelligence. I am reminded of numerous medieval arguments concerning the impossibility of a vacuum. They didn't pan out either -- and for rather straightforward empirical reasons. Gary Merrill