park@usceast.cs.scarolina.edu (Kihong Park) (04/11/90)
Sorry to barge in on the argument like this. Not to beat around the bush, I would like to forward this comment. A better understanding of high-level aspects of "intelligent" phenomena, or for that matter any type of solid understanding of this elusive property which we so cherish is definitely called for. But can an understanding be reached by thought experiments and contemplation alone? Don't get me wrong. For what it's worth, I do consider myself a theoretician and do ravel in this aspect of investigation. But could it be that our approach is wrong? At least, it may be the slipperiest side of the mountain we wish to climb. Why not try more pragmatic approaches? Instead of trying to understand what it means to understand, we could try to direct the effort at understanding less "abstract" phenomena. For instance, in my mind, the second most amazing aspect about "intelligence" is its low-level nature. What we conventionally consider "intelligent" phenomena(inductive/deductive reasoning, huge self-awareness, abstract symbolic manipulation one-step removed from its physical basis, etc.) are traits that we would in most part attribute to this select class of beings called humans. But what about the commonalities we share with our "lesser" beings? Your dog surely wouldn't know how to play chess, but is knows to walk/run without bumping into things, it can distinguish different people and objects, it has "feelings" in the sense that it knows what it likes, and it also can be trained to do various tricks. What about a bee? It solves mind-boggling navigational problems, it knows to build "bee-homes", they have a sort of hierarchical social structure which they do abide by, etc. Surely, these things require some aspect of "intelligence". A fundamental substrate upon which higher forms of "intelligent" phenomena are maybe based? We are prone to zoom-in on the most interesting aspect. But is the correct theory inferable from the information conveyed by the tip of the iceberg alone? I for one have gotten the feeling that it's not. I think this is the wrong track. Has 3 decades of symbol-manipulation oriented AI research provided us with any fundamental insights into this phenomenon called "intelligence"? The most basic questions still stare us right in the face. I am not trying to make an all-out pitch for neural network type approaches. I am just saying that whatever computational model one uses, the "right" questions should be asked and dealt with. A pretty presumptious statement, but that's how I view it. I don't think trying to understand what it means to understand is solvable at this point in time where we don't even have a scant understanding of the substrate processes. I may be wrong. I may be crazy. Sounds like lyrics to a tune... Kihong Park. (park@cs.scarolina.edu)