Author: Dan Homan
Date: 14:54:30 06/27/01
Go up one level in this thread
On June 27, 2001 at 17:03:58, Bruce Moreland wrote: >On June 27, 2001 at 15:09:22, Dan Homan wrote: > >>On June 27, 2001 at 10:56:24, William H Rogers wrote: >> >>>It never ceases to amaze me how a small handful of people who may or may not >>>have ever gone to college or even studied advance computer science and make >>>statements that are 100% the opposite of some of the worlds greatest minds from >>>all over the world have stated. The top great thinkers from almost every >>>university in the world have defined A.I. and what it is suppost to do and yet >>>there are a few young people here would match their intelligence againts theirs. >>>Maybe the Harvard, Yale, Stanford, or people at M.I.T.s should look into getting >>>rid of all of their Phd's and call you guys. >>>A.I., as it has been losely defined is the ability of a device to solve or make >>>decisions regarding a specific problem. How the machine was designed or >>>programmed is not the point, it is what it does afterward when it is turned on >>>and ran. >>>Not trying to stir up more waves, but I have studied this matter for years. >>>If you disagree, good, but just don't rewrite the worlds greatest accepted >>>facts. >>>Bill >> >>Hi, >> >>I am not sure that there is this universally agreed upon definition of >>artificial intelligence that you quote. Perhaps the first expert in computers, >>Alan Turing, came up with the 'Turing Test' for artificial intelligence. His >>test is that you can ask the machine any twenty questions you want by typing >>into a terminal and the answers appear on the screen. If you cannot tell >>whether the answers were given by the machine itself or another human (who might >>be hidden in another room), then the machine is said to be 'intelligent'. > >The Turing test seems stupid to me, and I have no idea why it is touted as some >sort of standard of anything. It measures a program's ability to generalize in >the domain of light conversation. That's a fine thing to try to write a program >to do, and it's a hard problem, but it doesn't make sense to draw a line and say >that programs that can successfully make light conversation are intelligent and >those that can't aren't, in some absolute sense. There is no universal constant >H, which is defined as the capabilities of a human, that also defines >intelligent behavior. > >A dog can exhibit intelligent behavior, without coming anywhere close to being >able to pass the Turing test. Likewise, there are many humans who cannot >successfully impersonate a particular human. If you disbelieve this, just watch >"To Tell the Truth" on TV. > >AI is not about making something that can solve the Turing test. There are lots >of AI problems that have nothing to do with being able to simulate light >conversation. > >bruce > I didn't say Turing's test was a good one... in fact I pointed that chess programs would fail a limited version of his test because they were too successful at certain kinds of positions.... Bill ranted that people won't just accept the "expert's definition" of artificial intelligence, so I pointed out Turing's definition was quite different, and that he is clearly an expert. On the question of whether Turing's test is a good one or not, I guess that depends on what you would like to see out of artificial intelligence. If we want a generalized intelligence that can learn the nuances of language and the tons of other stuff necessary to fake being a human, then Turing's test is clearly a good one. If you just want the fastest driving route from Boston, MA to Charlottesville, VA, then I agree that his test is not after the right thing. - Dan >> >>This standard seems quite different than the one you quote above as the expert >>certified definition. In fact, the bar set by Turing is much higher than simply >>solving a problem. I am not saying that your definition is wrong; I just want >>to point out that even experts disagree (or at the very least define their terms >>differently). >> >>It is interesting to note that there was an extended discussion back in r.g.c.c. >>several years ago about whether chess computers could pass a 'very limited >>Turing Test', where the scope of the questions is restricted to chess positions >>with a single best move and the expected answer is that best move. I forget >>who, but someone pointed out that the computers could not pass even this limited >>test, because if you fed them 20 long, complicated mates, the computer would >>respond much more accurately (and quickly) than any human could be expected to. >> >> - Dan
This page took 0 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.