Author: Andrew Dados
Date: 13:58:52 03/14/01
Go up one level in this thread
On March 14, 2001 at 16:36:46, Bruce Moreland wrote: >On March 14, 2001 at 14:20:08, Christophe Theron wrote: > >>On March 14, 2001 at 12:56:58, Bruce Moreland wrote: >> >>>On March 14, 2001 at 05:25:37, Andrew Dados wrote: >>> >>>>I doubt chess domain is wide enough for 'generalization' here. If a program >>>>could learn Thai chess in 5 min as all chess-playing humans do I would attribute >>>>it some 'intelligence'. >>> >>>I disagree -- I think it is wide enough. As I said, people spend tremendous >>>time mastering chess and still find more to learn. That's big enough. I agree >>>that a program that can handle more than one game without special purpose code >>>(no fair making a program that essentially has two modules) exhibits more >>>intelligence than one that can only play one game, but I think it's still good >>>enough. Fifty years ago it was certainly good enough. >>> >>>Please don't think that I argue that the exhibition of intelligence by a program >>>means that I argue that the program is human or even fractionally so. The >>>ability to exhibit intelligent behavior is very complex in humans, and I don't >>>think it's necessary for a computer to superset a human before its behavior can >>>be labelled intelligent. Otherwise the issue may be whether we can say that the >>>program is human. >>> >>>This may be part of the problem. Humans associate intelligence with humanity, >>>and they are very jealous of their humanity. >>> >>>I don't think that saying that a program solves a problem in intelligent fashion >>>is a threat to humans. >>> >>>>For now traversing Shannon tree with huge speeds and evaluation function >>>>'correct' in 99,96% or so I call 'good craftsmanship'. >>> >>>It takes a hard problem, one that provides humans with decades of challenge, a >>>problem that is spoken of in the same breath with art and music, and handles it >>>with sufficient skill that humans must take note. >>> >>>The AI field picked chess as hard solvable case that it could learn from. It >>>turns out that you can do chess pretty well without inventing something that >>>will win you a Nobel prize. >>> >>>That doesn't diminish the problem. That alpha-beta plus primitive eval can >>>perform so well is actually very interesting. >>> >>>I think the fact that an intelligent program can be created by a high school kid >>>in a few weeks is fascinating. >>> >>>bruce >>> >>>> >>>>-Andrew- >> >> >> >>Great serie of post Bruce. Very much to the point. >> >> >> >> Christophe > >I think that what I should have said is that the AI people aren't interested in >chess anymore because the solution is too simple, not the problem. That the >problem can be solved to such a degree with simple techniques is amazing. If >you could bring Turing or Knuth or any of those old guys forward from 1958 to >the modern day, and show them a modern program, without them knowing anything >about advances in hardware or programming techniques, they would immediately >declare the programs intelligent, and they would assume that the software must >be fantastically complicated and advanced. If you told them about the hardware >advances they might still believe this. > >The fact that a decent engine could be a few thousand lines of a pretty low >level language would stun them. > >I don't think that the simplicity of the solution means that the problem itself >is boring. It's simply amazing that such a complex problem can be solved so >simply. > >I think this must be what you get when you send researchers and product >developers to war with each other for 40 years. > >bruce I think chess can be easily mistaken for a complex problem while it is not. As you pointed most advances over last years were done thanks to speed increase rather then software. In a problem where full information is available your move is determined; you don't make 'decisions' or 'choices'. That is somehow obvious to me, however I fail to create good set of arguments to back up my point that chess programs are showing no intelligence. However if you call chess program intelligent exact same reasoning applies to program playing 3x3 tic-tac-toe. Computation cost of solving a deterministic model does not make a solution to it more 'intelligent', imo. -Andrew-
This page took 0.01 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.