Can the top players beat the best computers anymore?

Hi there

As we know, the reason why computers are winning in standard chess is because they have access to two hundred years of opening theory that puts them into a position midgame that is always reasonable. Once they are in the midgame, computers use their limited depth to outplay humans tactically which is a human's weak point. Because the opening has provided the computer with the backbone of the strategy as well, the computer is unlikely to go wrong strategically. Once the computer reaches the endgame, it is unbeatable because of the exhaustive endgame database. In this way, it is mainly the databases that have given computers the advantage they have today, and just a little bit of calculation advantage due to human inaccuracy in that department.

I disagree that it is just a case of software developers concentrating on better software for Chess960. The issues for the computer go far deeper. Firstly it will take decades for a decent Chess960 database to emerge. In the meantime the computer hardware does not have a search horizon big enough to see the required structures that any particular 960 opening requires. Even a super computer like Deep Blue or the like only adds a few ply of search depth onto our standard desktop computes. However in the Chess960 opening, it is more about pattern recognition than raw search depth in any case. The computer tends to build flaws into it's opening play because of this in ability to see patterns. Once those flaws are set in concrete, a trained Chess960 player will take advantage of those flaws without blundering because the human has the initiative. Once the endgame is reached, the computers position will be loosing and so it's end game database is of little use.

Two things will be required before Chess960 computers are at the top level of Chess960. Either the technology will have to change so that computers can recognise patterns, or some fundamental opening rules of play will have to be determined that are generalised for all Chess960 openings. However that is very unlikely, considering the diversity and sheer enormity of the Chess960 possibility set.

If you are seeing Chess960 computers playing well and beating quality players, it is not because the Chess960 computer is playing that well. It is because as yet we humans have not practiced Chess960 sufficiently. We have become reliant on our Chess opening theory for SP518 and this complete line of theory has only limited relevance to the other 959 positions.

In order for humans to play Chess960 well, all they need to do is expose themselves to thousands of games. I am in the process of doing this myself. We condition our mind to see "principals" that we can apply to any of the 960 positions. These principals do not translate to machine code because the principals are pattern principles that we begin to see after thousands of games.

My honest gut feeling is that if there actually was a specialist stream of Chess960 players out there, from what I am seeing of the best 960 engines out there, a 2000+ rated dedicated *Chess960* player would regularly beat any engine on the planet in classical time controls and will do so for the next few years as well. A 1900+ rated Chess960 player would have the advantage, but would loose to blundering. If Carlsen or Kramnik were able to overcome their Chess960 blindness because of their life time of playing only one position, they would absolutely slaughter any 960 engine for the next era of computing until the technology changes. This is irrespective of the software. The problem is that as yet we have not even had one single generation of human players out there that have played Chess960 from their very early years (5 years old). If children were to be exposed like this, I would say that within their first ten years, they would be so rock solid in all Chess960 starting positions, that they would still outplay computers even a decade from now (except if the technology is due to change more rapidly than I am aware of).

Only my opinion. We have a window of opportunity for another five perhaps ten years. Unfortunately we didn't see Bobby Fischer's vision soon enough back in the late nineties for us to already have a generation of 960 players through the system. Most of the problems humans are having with 960 is purely because old habits die hard in us.

Cheers

of course they can... they just need a hammer.

Or pull the plug (and the battery if used).

I agree that there is no ECO of opening for Chess960 (yet), but I'm pretty sure that people have started to establish sound opening theories for it. In time we will see publications about it ala Reuben Fine.

I disagree that endgame tablebases are "useless" for chess960 engines for one simple reason: the pieces and their characteristics are exactly the same for chess960. A KP vs K endgame is the same for chess and chess960.

There is one major requirement for stronger chess960 engines: more monstrous computer processors. Since chess960 is like the middlegame on move 1, the faster the CPU, the faster it will calculate the best move. We both agree on introducing opening theory for the first moves of a chess960 engine, but a faster CPU would still be needed to handle the extra calculation.

If there's money in chess960 already, five years is enough for the chess960 engines to evovle.

i dont thinks so.Computer generally have thousands of moves to choose from and databases of games.it therefore searches for the best move depending on how a good player is playing at that given time, and therefore playing a suitable move at that given time.You remember forexample between deep blue and kasparov,most of the games the deep blue chess computer played were really strong.Because of the thousands of moves it calculates and its choice of the best out of these moves,top players cant beat computers anymore..

Even the best player in the world cant have such a vast memory and may rely on lots of theories..

klamarson, your evidence does not seem to support your thesis. 

Furthermore, computers actually calculate moves after they are out of the installed book openings and before they can access their tablebases. That included Deep Blue in 1996 and 1997.

Computer calculation was not limited to the strength of chess players of the time; they were limited to the cleverness of the programmers AND the speed of the computer processors that they ran on.

Computers aren't so tough -- have them calculate at my speed, about 1 move per second and we'll see who understands the game better.

Computers beat humans at chess but humans can still beat computers at chess-boxing.

I was under the impression that engines are still monsters at that speed, actually. The chess.com computer is far from being the strongest engine around and plays moves almost immediately; yet is still reasonably strong.

There is in general a rather different relationship between the speed of play and the quality of play for computers and humans. The result of this difference is essentially that the slowest rate of play suits humans best and the fastest beats computers best. However, these days, it is generally acknowledged that computers are strongest at all over the board speeds. Correspondence chess gives humans the best chance, as evidenced by the recent draw between the Chess.com Alliance and Deep Shredder 12 (with a 2010 SSDF rating of 3117 at normal over the board tournament speeds -  enough to get a huge plus score against the human world champion).

... and of course, none of the chess.com alliance member used an engine to double check for tactical mistakes

I have no reason to believe so, based on the discussions that took place. With several reasonably strong players analysing, it is not going to be short term tactics that decide a game, but rather long term evaluations.

For most of the game the play of the human team was rather uncomputer-like and rather inaccurate according to the computer evaluations. The direction of the game was influenced by ideas from several players, and there were apparent minor misevaluations at various points (the last one being a rather lengthy one by the computer, it appears). There was very good agreement between the choices and the computer choices only at the end, which was partly a matter of there often being a clearly better move, and partly a matter of luck in the choice from several moves of similar quality, it appears.

Many different strategies for this are discussed here:

http://www.chess.com/forum/view/fun-with-chess/i-beat-deep-blue--finish-the-sentence

notlescu, I am reminded of the multiple times someone who has just been crushed in a blitz game by me cries "computer". It was an explicitly a match between a human team against a computer, with no assistance. There is no need to make (joking) accusations to slight the achievement of those who put in a lot more effort than I did (see the discussion).

The game was tough to draw (in my opinion) after we (including I) were a little over-optimistic about our chances after a pawn sacrifice, until the thematic point where our kingside attack was compromised from computer play on the queenside. I think we were a little fortunate in the final phase (where the computer and we agreed about the best course of play for about 10 moves, but the computer's assessment changed a lot from the start to the end, to its disadvantage).

One thing that is interesting is whether there is some sort of threshold of diminishing returns for calculation speed in chess. Does adding another ply at 20 ply make as much difference as adding another ply at 10 ply? Is there a limit to the standard of play in terms of ratings, or will they continue to increase to 4000+ in the coming decades? I do not know the answers to these questions but there is some evidence in the comparison of performance of the same engines at a range of speeds, over many years. Other evidence would be if games between computers of a similar level get increasing fractions of draws as they get stronger (as is the case for humans), indicating that they approach a level of adequacy to avoid losing against each other.

If there is some sort of threshold, it may be possible with enough (human) effort to reach a level which is adequate to draw against engines much of the time, even as they increase in speed.

My guess is there is no threshold level for an extremely long time (until things like the 50 move rule and the finite number of positions become relevant.

Notlesu would beat an engine by driving it crazy.

 fewer Garry Kasparov has said "at any given time there are only at most 6 truly good chess players in the world" and at the time he was certinly not talking about himself because he had been retired for many years.

Human who achieved the highest ELO rating: Garry Kasparov (2851)

Computer with the highest ELO rating : Rybka 4 (3227)

Computers are stronger BUT humans can do more than chess computers and still compete with them.

We know that humans can do more and compete with computers, but the question is: Can humans still beat the computers?

Exactly! It may be possible, with enough training and effort, that a human will fly like a bird using only the force produced by his body.

Tell Yelena Dembo that we miss her.