Man vs Machine

[User923005]

Pure guesses on my part:

I guess that Houdini on a cluster would be at least equal to the strongest human correspondence player. Most of them at least use computers to check their moves now anyway. I suspect that once in a while the human would get the computer in a positional bind, and just as often at least, the computer will see some tiny tactical advantage 45 plies away.

As far as FRC, programs like Stockfish on very high end hardware can search 40 plies in an hour even on opening positions (Stockfish is the best program for analysis of openings, better even than Houdini[*], but most people use a book there anyway. By the time you are out of the book, Houdini is better unless it is a really quiet position). At a depth of 40 plies, the computers almost always guess the right move.
So, if you take a database of FRC opening positions and analyze the 10,000 most probable positions for an hour each, the computer would be truly formidable, because I do not think a human could prepare as well without the slightest defect.

Ten thousand hours sounds like a lot, but 10,000 computers on the internet could do it in one hour. And a million computers could make a book so formidable that no human would stand a chance, especially after mini-maxing it.

Rather than improving the human's chances, I think computers will be much better to adapt to FRC, since the internet holds a lot of computers and those computers can be put to use analyzing chess positions.

[*] based on my statistical analysis which considers correctly choosing the position that leads to actual victories[2] in real games against very high end opponents.

[2] well actually victories + draws margin as a function of games played.

[mschribr]

Don, I think you are right. However, humans can still be ahead in chess if we make one change to the rules. If a player captures a piece then that player can drop that piece back anywhere on the board and use it as his own. I am talking about Crazyhouse. For starters, it would eliminate the computer advantage of endgame databases. It would also eliminate draws. The game becomes much more complex which I think is to the advantage of humans. This would put man back into the fight of computer versus man in chess.

[User923005]

Don, I think you are right. However, humans can still be ahead in chess if we make one change to the rules. If a player captures a piece then that player can drop that piece back anywhere on the board and use it as his own. I am talking about Crazyhouse. For starters, it would eliminate the computer advantage of endgame databases. It would also eliminate draws. The game becomes much more complex which I think is to the advantage of humans. This would put man back into the fight of computer versus man in chess.

Anything that increases the branching factor will favor man.
For that reason, computers are not as strong at Go as humans.
But that tower will fall also, due to the exponential march of Moore's law.
Ten years ago we were asking if computers had any chance against a GM. Now we are asking if a GM has any chance against a computer.

[mschribr]

Yes, computers will win in the end. But not from Moore's law. The computer superiority in chess comes more from smarter programs than from faster hardware. If today's programs ran on 20 year old hardware, then if will beat 20 year old programs running on today's hardware. Another example, Deep blue is still faster than today's laptop computers but Deep blue would lose to today's top program running on a laptop computer.

[User923005]

As I recall, a thread in CCC's programming forum calculated that the increase due to software was about equal to the increase due to hardware.
At any rate c1*exp(x) + c2*exp(x) is still O(exp(x)) so the computer {eventually} wins no matter how you look at it.

[Don]

Yes, computers will win in the end. But not from Moore's law. The computer superiority in chess comes more from smarter programs than from faster hardware. If today's programs ran on 20 year old hardware, then if will beat 20 year old programs running on today's hardware. Another example, Deep blue is still faster than today's laptop computers but Deep blue would lose to today's top program running on a laptop computer.

There was an extensive debate on talkchess about where the most improvement came from. Although there is some disagreement I think most of us believe it was approximately 50/50 - 50% software and 50% hardware. That is pretty staggering. You can see this in recent times simply by looking at the older programs that still run on todays hardware such as Fruit, which was dramatically groundbreaking when it came out - and yet now it's seems like a pretty weak program compared to a number of programs.

[User923005]

I think that almost all of the strength comes from better and better branching factors. A number of programs now have practical branching factors under 2.

[lucasart]

I haven't been keeping up with chess that much but I do realize that computers have come to the point of dominating even the highest rated human players. Bobby Fischer was concerned that chess would be played out in the top levels of competition so he started promoting Fischer Random Chess in hopes of preventing that possibility. Out of curiosity, do computers dominate against Grandmasters in FRC to the same degree as classical chess? Since there is no opening database, do the computers still have such an overwhelming advantage? What about correspondence chess? I read one time that a correspondence GM was actually rated much higher than their over the board counterparts.

I don't have an answer, but my educated guess is that computer would dominate even more in FRC.

Don

Totally agree with that. Without the help of memorizing lots of opening lines, humans are completely lost against a computer. When it comes to solving unclear problems over the board, a lot of tactics are involved, and today's programs are infinitely superior to GM...

[Alexander Schmidt]

I don't have an answer, but my educated guess is that computer would dominate even more in FRC.

Don

Totally agree with that. Without the help of memorizing lots of opening lines, humans are completely lost against a computer. When it comes to solving unclear problems over the board, a lot of tactics are involved, and today's programs are infinitely superior to GM...

I wouldn't be sure about that. In the openings the tactics are not as important as the positional understanding. So maybe the human can get even a little advantage there if he doesn't run into a tactical trap. My personal experience aganst engines of my strength is, that it is easier in FRC though I have not much experience in this variant. When playing against a stronger engine it is more dificult in FRC because I am not saved by some theory moves and run earlier in tactical problems. But the result would be the same. So below the line I guess that engines are relatively a little bit weaker against humans in FRC.

[peter64]

I do not think the proposed match between Carlsen or any other top master against a top computer is interesting. At this
point it is commonly accepted that computers are stronger than humans.

To me the interesting match is Carlsen or any top master plus the top computer against the same top computer, in a match
of several games, the more the better. If Carlsen aided by the computer makes a clear win we will know that human intelligence
still makes a difference. If no side makes a clear win then we will know that the computer not only beats
any human player in the tactical aspect, by calculating a huge number of variations, but we will also
know that the computer makes human intuition and positional evaluation irrelevant compared with
the results of the computer. In other words, the top programs will be playing at a level so high that
even the best human intuition and strategic evaluation given to one of the sides will not alter the
result.