ON THE WAY PEOPLE AND MODELS DO IT KENNETH MARK COLBY* Model builders in the field of artificial intelligence who attempt to simulate human mental processes often face the question: Granted that your model is doing the same thing people do, how do you know the model is doing it the same way people do it? This question has, at various times, seemed odd, slippery, and vexing to me. Let us try to be clear about what a "same way" question might request. We first assume that the questioner and model builder have come to agree that people and models are "doing the same thing" under some initial set of described observations; for example, they are playing chess or solving problems or conversing in English or acting paranoid. Assume we agree on the descriptions ofclasses of patterned actions on which we are focusing our attention. Now, are the individuals involved (people and models) carrying out these patterned actions in the same way? This apparently simple adverbial question conceals more than a few difficulties . Consider the words "dog" and "cat." Are they the same or different? They are the same because: (1) they are letter concatenations ofthe same length; (2) they are English words; (3) they are terms for animals; (4) they have a vowel in the middle position; (5) etc. They are different because: (1) they have different letter components; (2) they refer to different animals; (3) they have a different vowel in the middle position; (4) they sound different; (5) etc. Thus, designating two individuals as the same or different can be played as an endless game. We have to say they are the same in respect to something, and for every property they share we can cite a property they do not. Basic to this game ofsame and different is an interest in forming general classes which free us from dealing with an otherwise overwhelming variety of experienced particulars. We place individuals in classes knowing they are not identical (otherwise we wouldn't term them "individuals ") but treating them as if they were by blurring certain dis- *Department of Psychiatry and Biobehavioral Sciences, Neuropsychiatrie Institute, University of California at Los Angeles, California 90024. This research was supported by grant MH27132-02 from the National Institute of Mental Health. Perspectives in Biology and Medicine ¦ Autumn 1977 | 99 tinctions. For an imprecision of reference we gain a precision of inference . What we want from a classification are inductive groups about which new and reliable inferences can be made. It is time, labor, and resource saving to generalize about classes, not only for cognitive efficiency but also for pragmatic purposes—for example, in medicine when we wish to carry out treatments in relation to members of an equivalence class. The term "way" usually means a method or manner of doing something . Suppose you and I are playing chess. (1) Are we doing the same thing? Yes. (2) Are we doing it the same way? Yes and no. Yes, because chess is constituted by rules and ifwe were not both following these legal rules we would not be playing chess. No, because as unique individuals we cannot behave in exactly the same way. It has seemed worth repeating for several centuries that no two blades of grass and no two people are alike. Again, much depends on what is agreed on as the description level. If it is a truism that every individual is unique, we must remember that to be unique does not mean that an individual shares no properties with others; it isjust that he does not share all his properties with others and does not behave exactly like anyone else. The number of shared properties is description relative. At some level of description you and I are playing chess in the same way and under another description we are not. Suppose when you analyze a chess position, you examine the pieces first whereas I look at the pawn structure first. Does this difference make a difference? If we arrive at the samejudgment about the position as to who has the overall advantage, this different "way" is not significant for that judgment, but it might be significant for something else...
Read full abstract