THE COMPUTER CHESS HAS REACHED ITS CRITICAL POINT

[ZeroOne]

The goal, before solving the game (even for 7, 8, 9 or 10 pieces -- remember that chess has already been solved for 6 pieces and less), should be to beat human players in correspondence games. Currently humans are still stronger in correspondence chess with very long time controls. Just think of Arno Nickel beating Hydra 2.5-0.5.

[Terry McCracken]

I think he means to play very deep positional chess with Real planning like Kasparov or Karpov, without the human weaknesses, such as a tactical oversight.
Terry

Unfortunately we are now at a point in computer chess where the human approach is inadequate. Going down that route would most likely make todays best engines weaker.

In terms of tactics, I believe most folk make the mistake that it is defined by some combination of moves that leads to a spectacular mating attack or big material gain. Tactics concentrating on the promotion or strengths of a pawn may not provide the same spectacle but are just as effective and in all probability more pertinent to the higher level of criteria needed to produce game winning tactics for chess playing engines.

AK

The Human Approach is superior, when no tactical errors are made.

Look how foolish Kasparov made Fritz look when he took the gloves off and played chess at the highest possible level!

http://www.chessgames.com/perl/chessgame?gid=1269891

[ozziejoe]

Excellent points. I would pay decent money for an engine that could imitate different human styles (i have ordered cm11 but i am not confident it will play like different humans)

[Kirill Kryukov]

I personally think that 1-st place in the rating lists is still a worthy goal and a good challenge. This goal is more practical currently than "Perfect chess" of "Beat human in correspondance chess", but eventually it leads to those goals just as well, I think.

The good goal is one that allows for measurable progress. With those abstract goals lile "Perfect chess" it is difficult to measure the progress, or to know where we are or if we are improving. (All we know is that we are very far from goal ). With manageable goals like "#1 in CCRL blitz list" you have goal that is 1. Possible to reach (As long as #1 does not play Perfect chess, which they don't ) 2. You can see if you are progressing and how far you are from the goal.

All IMHO.. Best, Kirill.

[Mike S.]

Strong human players in correspondence chess are actually multi-centaurs, human+engines. An engine alone can hardly win against a corr. GM who has the same hardware and time available, and up-to-date engines (except he makes a series of bad decisions or one very bad one, which is very unlikely to happen).

While the improvements of the "inside" competition, engine vs. engine, can always be measured, it gets more and more difficult to measure improvements against humans. We have only very few games against GMs, some of which were experimental. But this would be important to know and to have (many) good examples for, because the human strengths/weaknesses profile is so much different.

[ZeroOne]

I personally think that 1-st place in the rating lists is still a worthy goal and a good challenge.

Being number one in the rating lists only proves that the engine is better than any other engine but it still doesn't prove anything about the strenght of the engine against human players. Engine vs. engine games are usually distinguishable from human vs. human games. Both humans and engines have their strengths and weaknesses, neither ones are perfect chess players.

[gerold]

I personally think that 1-st place in the rating lists is still a worthy goal and a good challenge.

Being number one in the rating lists only proves that the engine is better than any other engine but it still doesn't prove anything about the strenght of the engine against human players. Engine vs. engine games are usually distinguishable from human vs. human games. Both humans and engines have their strengths and weaknesses, neither ones are perfect chess players.

I agree with you. Computes are about equal to top gms now
It depends on the gms top players to keep improving to keep
up with the computer.Some anti-computer players can still draw and
beat the computer as was done in past years. As the computer
improves the human players improve.
Do you think that top gms that can memorize good positions in top
good games in diff.opening are better players vs. the computer.

[Anabolic Karpov]

The Human Approach is superior, when no tactical errors are made.

Look how foolish Kasparov made Fritz look when he took the gloves off and played chess at the highest possible level!

http://www.chessgames.com/perl/chessgame?gid=1269891

That was four years ago. Computer chess and today's hardware has changed significantly since then and the discussion as I understood it was for the future so no point looking back at what was.

More recently, GM results have not been so good and after Kramnik's embarassment against DF10 where it turned out it was only running on single core not the four cores that we all thought it was then it is likely that the top GM's would struggle against today's top engines and hardware combinations.

However, your point about "when no tactical errors are made" is very valid and perhaps that is what the goal should be?
But isn't the fact that today's engines can still make errors the very thing that still gives them a human like nature? It's just that the errors are usually beyond most of our comprehension.

AK

[Marc Lacrosse]

Strong human players in correspondence chess are actually multi-centaurs, human+engines. An engine alone can hardly win against a corr. GM who has the same hardware and time available, and up-to-date engines (except he makes a series of bad decisions or one very bad one, which is very unlikely to happen).

While the improvements of the "inside" competition, engine vs. engine, can always be measured, it gets more and more difficult to measure improvements against humans. We have only very few games against GMs, some of which were experimental. But this would be important to know and to have (many) good examples for, because the human strengths/weaknesses profile is so much different.

There is still a considerable gap between top level "humans" (OK : "centaurs" is more appropriated) and best engines in correspondence chess in favor of humans.

I play correspondence chess in the two main organisations where computer use is allowed : ICCF and FICGS. I would like to stress some points that computer chess afficionados do not clearly understand.

1. The human touch is still decisive and worth several hundred elo points in these games. All my opponents use computers and most of them have pretty up-to-date engines and computers. This includes opponents who are in the 2100-2300 range. My recent performances were in the 2400-2500 range with a 70-80% performance against people who all owned and used computers. In the same time I am pretty sure that I could not match those grandmasters whose Elo is 300 points higher than mine although I have good hardware, best programs and (I feel) some experience in the field.

2. The frontier is correspondence chess. I think that top engine makers should have lots of things to learn with their babies playing anonymously in organisations like FICGS. But they do not have the patience for 6-12 months games. I am ready to bet that I am able to beat any anonymous engine playing alone on any hardware at a "rapid" correspondence timing like 30 days + 1 day per move fischer timing. My feeling is that I should be able to achieve a 60-65% score playing as a centaur with my present equipment against any engine at this kind of timing and that a true correspondence grandmaster should end with at least 70-75% points in this kind of situation.

In recent correspondence games the human touch is more than ever decisive. In fact this is really what makes difference when everybody is equipped with all sorts of rybka's and friends. Therefore this is the place where programmers should look for possible improvements.

...my two cents.

Marc

[ZeroOne]

In recent correspondence games the human touch is more than ever decisive. In fact this is really what makes difference when everybody is equipped with all sorts of rybka's and friends. Therefore this is the place where programmers should look for possible improvements.

I agree with you, Marc. Human touch is needed especially when the top engines all suggest a different move, which is certainly not uncommon. They may even suggest completely different ratings for a given position, one thinking the game favors white and the other thinking the game favors black.

Related to correspondence chess, I've found The Chessgames.com Challenges very interesting and educating. The World team consists of regular Chessgames.com users co-operating and voting for the next move. The first game had the World team playing as white against the correspondence GM Arno Nickel, yielding a victory for the World team. The second game was played against Grandmaster Yury Shulman, yielding another victory for the World team. The third game is played against World Correspondence Champion Gert Jan Timmerman. It is currently ongoing so you can still join the team.