Slashdot Mirror
Chess Championship: Humans vs. Computer
An anonymous reader writes "X-bit labs has posted very interesting editorial called "Chess Championship: Humans vs. Computer".
During the last 10 years computers penetrated into various spheres of human life. In this article guys try to find out how well computers can
play chess and if it would be correct to say that artificial intelligence is superior to human mind. Interesting read."
The First "Cyborg Championship"?
Meanwhile, Garry Kasparov has arranged for an exhibition match with 23 year old GM Veselin Topalov of Bulgaria for June in which both players will have an (identical) laptop computer that they may consult during the games. The laptops will have databases preloaded by each player (therefore containing their own analysis and selections), as well as a tactical engine. Each game of the 6 game match will last only one hour, meaning that a large part of the strategy will be how much time spend on the computer! A number of analysts are calling this the "First 21st Century World championship" although of course it's only an exhibition. (http://www.uschess.org/clife/issue47/buzz.html)
It's from 1997, but I think they're right. The future does seem to be moving in that direction.
I recall reading an interview with former world chapion karpov who said that when he was learning chess, his teacher said that one day it would all be computers. One of the other students said, "So why are we bothering to do this then?" and the coach replied, "My computer will beat your computer." or something like that. Pretty soon it'll all be down to which computer is better and which person can better control it. I'm sorry I can't better quote the interview. It was in the ChessLife about the Karpov v. Kasparov x3d match in Times Square in case anyone has it.
Learning to understand English is altogether different -- Language has a very complex set of very loosely defined rules which change over time, sometimes quickly, sometimes slowly. Understanding English is very much an AI task.
The problem is knowedge, and how it should be represented -- with Chess you just need a big calculator and present as much of the game (projected) as possible. There is no such way to do this with language... a much more complex representation with much more hueristic knowledge is required, and this is where AI starts coming in. Natural language processing is a very tricky field, one which I won't even pretend I understand, and in my opinion nobody quite does... Chomsky probably coming closest, but then again I'd disagree with him on many points!!
D.
It's been said before, but before we talk about computers becoming superior to the human mind, how about creating an AI that's *equal* to the human mind?
In other words, there's no point in talking about the future where computers rule supreme etc. if we still have no way for a computer to recognize, say, a table from a picture of a table if it does not comply with a series of previously-specified standards. I know it's a horrible analogy but jeez, it's 3:18 AM.
...Which reminds me. Why am I still up? *sighs* Damn you, caffeine.
"The power of accurate observation is frequently called cynicism by those who don't have it." - G.B. Shaw
We judge what the situation most resembles from our experience, and we react accordingly. We act like a case-based learning AI program. We use heuristics to weight our decisions...we just call our heuristics "common sense."
Computers act more like humans, and humans act more like computers, than many people are comfortable to admit. Computers just don't have the mechanisms to experience as wide a variety of stimuli as us.
Take a look at the work of Douglas Hofsteader (sp?). His book, "Fluid Concepts and Creative Analogies," shows relatively simple programs demonstrating surprisingly human-like behaviors.
Language has a very complex set of very loosely defined rules
Shhhhh. Don't let the objectivists hear you say that.
On the problem of knowledge, representation should be easy considering the obvious fact that All of man's knowledge and all of his concepts have a hierarchical structure.
So, knowledge is hierarchial only insofar that simple addition and subtraction pave the way for more complex algebra... knowing algebra without knowing addition and subtraction would give algebra no meaning.
However, this doesn't account for leaps of thought where entire planes of thinking are bypassed... and nor does it account for mad people (always a tricky one in AI) , especially mad geniuses!
How about a distributed chess programm, just like seti..???
Of course against distributed human thinking power, everybody can suggest the next move. The move that is suggested most often within a certain period of time, will be the next move.
10.000 computers vs. 10.000 human minds. Could be interesting.
What does "learning from mistakes" imply? Well, what is a mistake? It's when our plan of action failed to achieve its goal. A computer can easily simulate this, given a goal (that doesn't even have to be very well-defined).
Say I want to drive to work. I have a choice of roads to go down. At first, they all seem equal to me, but eventually I learn which ones are heavily trafficked and which ones run smoothly. I then bias my trips towards the roads with less traffic. I have learned from both my mistakes and my successes.
Comprably, a computer is in state A, and wants to get to state B. The computer tries all its available methods to get from state A to state B, and weights them according to the (utility per resources) that they provide. In the future, it uses this information to choose the best path.
It's the same process, effectively. AI Planning is all about this stuff, especially Reinforcement Learning and Iterative Repair.
The game of Chess is not a measure of intelligence. It's a measure of mathematics and memory.
But will intelligence for a computer EVER be anything else than mathemetics and memory?
Will our brain EVER work in another fashion than sending chemical signals to our synapses?
Beware: In C++, your friends can see your privates!
This article gives an introduction to the problems involved in getting computers to play Go:
http://www.ishipress.com/times-go.htm
The interactive way to Go -- http://www.playgo.to/iwtg/en/
Just think of the world's first conscious, intelligent computer claiming that we can't possibly be conscious because we're merely the products of neurons firing.
Can't say I agree with that.
Addition and subtraction derive their meaning (in a strict mathematical sense, which when talking about pure knowledge, why not use a strictly mathematical sense?) from algebra. I mean, their meanings change depending on what algebra you are using.
However, this doesn't account for leaps of thought where entire planes of thinking are bypassed... and nor does it account for mad people (always a tricky one in AI) , especially mad geniuses!
Exactly!
If knowledge were hierarchical, one would be able to present said hierarchy. Since you can't, it's not.
The problem is, where do we draw the line?
1) Make a computer with true free will. Let's see AI do something it wasn't originally designed to do because it wants to.
2) "True intelligence", at least on par with us, will happen when a computer does everything we do mentally, while having full articulate motor skills, and then takes it upon itself to create an AI that crunches numbers better than it does, beats itself at chess, etc.
The full-circle of AI doing everything we do will be "true intelligence".
You are indeed correct, I can't
I would argue that there is a hierarchical base for knowledge, it just breaks off the further up the tree you get. Incremental learning must take place before independant thought can take place, so in that sense there is certainly a degree of hierarchy, I think.
Nobody got outraged when that new-fangled mechanical auto-mobile contraption started to outpace the world's fastest human runners.
Allow me to recommend to you the legend of John Henry. About the time period you mention, too. I always mention this story in the Intro AI class I teach.
Baloney. A man is allowed to memorize as many openings as he wants, just as the computer has "memorized" them.
Again, so? Humans are allowed to memorize as much endgame stuff as they want. Why should computers be disallowed this?
Awwww... Why the hell not? Human brains aren't single processor; why should computer opponents have to be?
The same fallacy, repeated over and over again. The human doesn't have none, he has as many as he cares to remember.
And if I were on the computer team, I'd let you. Knock yourself out! Go ahead and fiddle with your chessboard when you could be considering countless more positions in your head.
So, the humans are cheaters then, because they capitalized on computer blunders?
I object to that article, and to the next reply.
Consider Searle's Chinese Room problem. You feed someone (written) Chinese under the door, and they have an extremely complete book of rules for "translating" one set of Chinese characters into another. The person then feeds a written "reply" in Chinese back under the door. Do the people in the room know Chinese? To the people outside, it appears that they do. What if the person memorizes all of the rules. Can that person now be considered to "know" Chinese?
You can't just blanketly designated computer chess-playing as unintelligent, because we don't really have effective ways of designating whether something is "intelligent" or "human-like" other than gut-feeling or statistical analysis of "performance" (i.e. outward appearence). Turing had the right idea when he gave his version of the Test, in that the true test for intelligence is just the appearence of it.
Such irE
Kasparov takes the NYT log postings into account in his recent post. He cites Elo (chess rating) numbers by Ken Thompson (an old school computer chess guy) derived by extrapolating numbers generated by setting a computer program against itself with differing search depths, "world championship"[1] level performance would require 1 billion nodes (moves) per second. Interestingly, "one billion nodes/sec on a single chip" is possible with todays 0.13 micron process, while "a trillion nodes/sec machine is actually possible today" according to one of Feng-Hsiung Hsu (Deep Blue hardware designer).[2]
0 .html?804278037510812]
_____________
[1]Kasparov notes also that the chess performance ranking numbers that Ken Thompson derived were asymptotic(?); "which flattens at the top end" . From Garry Kasparov on Chess Computers (22.01.2003) [ONLINE][http://www.worldchessrating.com/52162987
[2](Note: The "one of Deep Blue's two programmers." citation is incorrect... the followup post clarifies the error.)
"A computer would never be able to compete with competent chess player, and could certainly never compete with a Grandmaster. Chess requires true human intelligence."
Douglas Hofstadter put forth basically this point of view in his Pulitzer winner "Godel, Escher, Bach" in 1979. He predicted that computers wouldn't be able to beat grandmasters until they'd achieved human-level intelligence in general.