group6 (12/03/82)
Ever since I read D. R. Hofstader's "G:odel, Escher, Bach: an Eternal Golden Braid" I have held the opinion that many researchers in Artificial Intelligence,at least those trying to mimic human intelligence, are going about it in the wrong way. Instead of trying to make super-smart chess programs or other "expert" programs, I also believe we need to examine the hardware that we are trying to copy, instead of trying to do it using hardware designed for tasks that are *completely* different. I don't know of too many people whose brain is much like a PDP-10, even if it runs an "intelligent" operating system. Because the architecture of the human brain is so different from the architecture of today's conventional computers, it is understandably difficult to get them to do anything even remotely approximating human intelligence. I believe an understanding of some important aspects of thought and reasoning is to be gained by modelling the mind with several n-illion similar (but *not* identical) circuits (call them "newrons"), connected in a random network. Some of the circuits should be connected to the external world, so that our "mind" can do more than contemplate itself. After this electronic tangle has been allowed to run for a while (?) we may observe the development of some kind of behavior. This experiment is similar to the "primordial soup" study done a while back by Stanley, in which the supposed mixture of compounds present in the early atmosphere and oceans was subjected to radiation and electrical discharges for an extended period. After a few weeks, the apparatus contained several of the simpler amino acids used in all known life forms. Any comment? Has this type of thing been tried before? Am I beating a dead cow? Am I way off base? -Dave Decot -...!decvax!cwruecmp!group6 (By the way, how about a new group net.ai, net.psychology, net.??? for this kind of discussion?)
soreff (12/06/82)
Self organising systems have been tried in AI, and by and large haven't worked. There was a great deal of work on "Perceptrons" growing out of ~McChollach (I know the spelling is wrong) and Pitts's work on neurons in the 50's. It turned out that the networks used had some intrinsic problems (Minsky proved some theorems which put bounds on the kinds of patterns they could recognize, bounds below human performance). There is also a lot of evidence that neural systems start out with a lot of structure, so random nets are probably not the way tp go. -Jeffrey Soreff
zrm (12/08/82)
While the central nervous system starts out with a lot of organisation, that organisation is also very mallable. Cases of asymptotic agenesis of the corpus callosum behave like normal controls in the same sorts of experiments where where commisurotomy patients exibit their remarkable "split brain" behaivior. Both are missing the corpus callosum, but the subjects born without it have developed other means of hemisphere to hemispere communication. That is, while the organ specialised to send information from one side to the other is gone, if it was not present at birth, other parts of the brain take over its function. The most important thing about out brains may not be their structure, but the way in which they develop. Cheers, Zig
jcz (12/13/82)
References: cwruecmp.312 Reminds me of how the freshmen CSC majors here at state used to resubmit the same job over and over in the hope that transmission errors would fix the bug. --jcz NCSU