November 16th 2003
Game 3 Kasparov v Deep Fritz New York 1-0
http://en.wikipedia.org/wiki/X3D_Fritz
The match was drawn 2-2.
That is the last human win that I can find.
Adams lost 5.5-0.5 in 2005 and Kramnik lost 2-0 in 2006. Not bad players.
Any advance on that? Going ... going ...
Heh, I remember that one. Those glasses were such a daft gimmick.
A super-GM can still comfortably beat a machine, if he has enough time to check his lines for tactical flaws (say 2 days per move) - quite simply because he does understand chess, while the machine only understands ones and zeroes. But I guess we will never find out, because there is no motivation. If someone offers Magnus several thousand bucks to beat a machine, I do not have the slightest doubt about the result. The problem is that silicon fans will claim that the actual winner was not MC, but a centaur.
pfren, how do you know? This looks like a complete and utter assumption to me.
Oh, I see, you justify it with common sense. Ok, but what you call "common sense" involves assuming something to be true simply because it sounds natural to you. That's good for making a hypothesis, but if you don't test it, you don't have much scope to measure the accuracy of your speculations -- they may be rational, but speculation is very open-ended.
Common sense rejected continental drift. Nearly everyone was absolutely wrong, and the man that was criticized for bringing it up was quite right (most likely). Doesn't that say a lot about the problem of assumption that makes up common sense?
Hello All,
This might be of interest to the group. Back in 1999 Master Stephen Ham went head to head against a computer in correspondence chess. The machine won--even back then. Anyway, the coverage is excellent. I hope you find this helpful.
Cheers,
brotherbear380
Just don't tell him the world is round.
Demands for evidence cuts both ways. I dont think we have a recorded result of a CC 3 day move game between a super GM and Houdini, which is what leaves open the door for kibitzing.
There is no doubt that humans become stronger relative to machines at longer time controls. The question is, is it enough?
Every day that this challenge doesn't happen, means that it is less less likely to occur, at all. Next week brings what? Houdini 3? Rybka 5? each program stronger than the last....
.... and yet, the GM's don't come forward to take up the guantlet, while they still have a glimmer of a chance. One can only ponder the reasons why.
Yes, like lack of pay for devoting several months to one game. I think that would be enough. And who is willing to pony up enough cash to entice someone to devote all that time?
If Houdini is 400-500 points stronger than any GM, why shouldn't it win at any time control? Otherwise, the Glicko system, or the Elo system, makes no sense probabilistically.
Do people really think that old saw about "tactical play versus positional play" has any meaning, when the (free) silicone opponent is 500 points stronger than Vishy?
50 years ago, Herbert Simon, a Noble Prize Winner in Economics, and Professor at Carneige Mellon, predicted that (in 5 years) a young child, using a hand-held-computer, would be able to beat the World Chess Champion.
Simon was wrong by about 45 years.
But that day has (now) arrived, has it not?
Interesting comment pfren... My common sense tells me that a computer rated at 3300 would beat 2800 rated humans every time.
I guess your common sense is a little bit different ;-)
Give the computer 2 days to think per move vs an unaided human with 2 days to think per move and my money is still on the top engines.
But do human ratings vs. computer ratings mean the same thing?
The highest ICCF rating is in the 2700's, but obviously an ICCF centaur would beat an OTB 2800 GM almost every time.
It's not so difficult to beat Houdini in correspondence chess. Computers still lack certain elements of positional understanding, and they can certainly be outplayed by a strong player. On rapid/blitz games though, it is a totally different story.
I'd be kind of surprised if humans do better in correspondence chess against a computer than in a 'normal' timed (but not rapid/blitz) game, because it can spend every second calculating (unless restricted in some way) whereas a human has the matter of life to attend to.
No reason to be surprised. Chess engines lack the ability to play positionally. Top GM's can steer the game into positional waters, and have a chance.
I assume that the rating systems are the same, otherwise bragging that an engine has a rating of 3300 would be pretty pointless.
How could they possibly be the same? Engines get there ratings by playing engines, not humans. I'm having a hard time finding information on this, but I'm not great at researching with the internet.
I assume that the rating systems are the same, otherwise bragging that an engine has a rating of 3300 would be pretty pointless.
How could they possibly be the same? Engines get there ratings by playing engines, not humans. I'm having a hard time finding information on this, but I'm not great at researching with the internet.
Only because humans can't compete at that level. Hence the 3300 Elo rating.
Demands for evidence cuts both ways. I dont think we have a recorded result of a CC 3 day move game between a super GM and Houdini, which is what leaves open the door for kibitzing.
There is no doubt that humans become stronger relative to machines at longer time controls. The question is, is it enough?
Every day that this challenge doesn't happen, means that it is less likely to occur, at all. Next week brings what? Houdini 3? Rybka 5? each program stronger than the last....
.... and yet, the GM's don't come forward to take up the guantlet, while they still have a glimmer of a chance. One can only ponder the reasons why.
Top GMs do not have much motivation to devote large amounts of time and resources to preparing for a match against a computer. This would probably involve significant work developing strategies to beat a specific engine, I would imagine that it is still possible to do this but the fact is no one really cares anymore, you don't hear about mathematicians trying to prove that they can crunch numbers better than a calculator.
***
As far as the question of human understanding goes, people overlook a basic principle, whilst I am not much of a chess player I can see the philosophical issue here. Positional principles are a set of abstractions imposed on the basic logical framework of chess to enhance human ability, they are not inherent in the game as such but they allow humans to make calculations which would otherwise be impossible. A human doesn't 'understand' chess any more than a computer does, they just have a different way of calculating. The machine's method, while much less efficient, is more exact, hence why machine's surpass man's ability quite easily once their efficiency is improved and their calculating ability increased. They still have their weaknesses, but soon these will be too small to exploit and right now there is not much motivation for top players to try.
I'm confused about what you are trying to say? I have highlighted it in red.
How is "positional" chess not an inherent part of the game? Advanced pawns on one side of the board is an inherent part of the game (more space) Just as controlling squares is an inherent part of the game. If i am misunderstanding what you are saying my apologies, but please explain.
That is a long winded answer, that doesnt explain why "positional" chess isnt an inherent part of the game. Positional chess is as much a part of chess as tactical play is. And this is where chess engines destroy human opponents.
The irony of this discussion is each position only has one game theoretic
value (W,L, or D) but has a myriad of values due to the limitations of search
and the strengths and styles of the participants.
Engines still have evaluation functions though, which includes positional elements. When they plonk a Knight on e5, they don't have calculated up to a material gain just yet, but their evaluation function takes into account that a centralized Knight is worth slightly more than a Knight on a rim for example (all other things being more or less equal at the end of the tree).
But of course, the calculation power is what makes the difference.
Interesting comment pfren... My common sense tells me that a computer rated at 3300 would beat 2800 rated humans every time.
Or at least 24 times out of 25 (according to the ELO rating vs win-probability table).
The irony of this discussion is each position only has one game theoretic
value (W,L, or D) but has a myriad of values due to the limitations of search
and the strengths and styles of the participants.
Nice summary of the essence of chess.
Interesting comment pfren... My common sense tells me that a computer rated at 3300 would beat 2800 rated humans every time.
Or at least 24 times out of 25 (according to the ELO rating vs win-probability table).
I'm standing by my claim that Engine ELO's can't be effectively compared to human ones. That conclusion comes from only my common sense so I'm trying to find more proffesional opinions.
Demands for evidence cuts both ways. I dont think we have a recorded result of a CC 3 day move game between a super GM and Houdini, which is what leaves open the door for kibitzing.
There is no doubt that humans become stronger relative to machines at longer time controls. The question is, is it enough?
Every day that this challenge doesn't happen, means that it is less likely to occur, at all. Next week brings what? Houdini 3? Rybka 5? each program stronger than the last....
.... and yet, the GM's don't come forward to take up the guantlet, while they still have a glimmer of a chance. One can only ponder the reasons why.