ol2 (12/01/82)
Searls assumes that an English book describing the way Chinese characters may be combined is possible. True, this is possible for a dictionary-type "knowledge" of the language but it does not capture its grammar. To look intelligent, our 'monkey' would have to respond properly to the characters handed to it. This means the book would have to specify the meaning of character sequences since not all grammatic sequences are appropriate at any time. (Recall Chomsky's famous sentence about green colorless dreams...) When we do that, we can program this knowledge into a machine. Anyone who has tried to program a machine to understand text knows that the whole subject of text analysis is closely linked to representation of knowledge, i.e "understanding" the text. When we get *that* sophisticated, the book we hand over might really be "intelligent". Of course, the prime counter-example is the 'doctor' program. However, this program could be very easily fooled if you knew how it worked, and it deliberately chose the part of a 'shrink' who says very little and reflects the answers back at you. I guess the lesson is that in a limited context we could mimic intelligence without any real intelligence, but if a machine truely behaves intelligently, it *is* intelligent, and intelligence is describable mathematically. Sorry for the heavy volume, Opher Lekach pyuxjj!ol2