Upgrade to Chess.com Premium!

Computers in chess... Good or Evil? Part Three.

  • GM Gserper
  • | Feb 27, 2013
  • | 10562 views
  • | 37 comments

The first two parts of this article generated many comments from our readers so let me address a couple of popular misconceptions.

First of all, the goal of this article is not to find out who plays chess better, computers or human players. This question was definitely answered years ago and the difference in the playing strength will be getting bigger and bigger as the computers become faster and their programs get more sophisticated. So when I analyzed the game where Nakamura managed to take advantage of the 'horizon effect' and beat the silicon beast, it was a good example of human being smarter than a computer, not stronger. (Of course this statement doesn't even need a proof since computers and their programs are created by humans and strictly speaking computer is just a piece of metal that executes human commands quickly and without mistakes). So to me Nakamura's game is a celebration of the human mind. 

The following point is more important in my opinion. Chess.com is a very nice virtual place where people from all over the world can share their views and ideas about our beautiful game. And my article is just that - my personal opinion. So, before you write in your comment that being a Grandmaster for over 20 years doesn't make me an expert in computer chess, please remember that I never claimed that my article is the ultimate truth about computer chess. It is just my personal view based on my personal experience.

After above mentioned disclaimers, let's focus on our subject.  Many chess players use chess engines as their guides. Indeed it is very convenient to ask Houdini or Fritz why Anand played that move and why his opponent didn't capture a pawn for free.  But remember, this approach has its limitations.  If the position is sharp and it is all about calculations there, then your chess engine is as close to the ultimate truth as it is possible.  But if the position is relatively quiet, then the computer's suggestions could be misleading. Here is a simple example:

I call this position "Capablanca's position" because I saw a similar position many years ago in one of Capablanca's books.  What's the evaluation of this position? Capablanca claimed that White is strategically winning due to his superior pawn structure. I am sure that most of the modern GMs would share the opinion of the great Cuban.  Meanwhile many computer engines think that White has a significant but not a winning advantage. Here we again see sort of 'horizon effect'.  Capablanca knows that in the long run a weak backward c6 pawn and a strong , potentially passed e5 pawn should bring White a win, meanwhile a chess engine cannot see that far (White will probably need 20-30 moves to convert his advantage into a win). The next game of Capablanca is a good example of this concept.  Please note that in his game Capablanca's e5  pawn wasn't potentially passed pawn and Black's c6 pawn wasn't backward or weak, and yet he was able to slowly grind down his opponent.
This is a good opportunity to answer a question from one of the readers about my game vs. Kramnik that I analyzed in this article: http://www.chess.com/article/view/what-is-a-proper-moment-to-resign-a-game?page=2
In the position after Kramnik's move 18. f4! I was about to resign and the reader asked why.  According to a chess engine White's advantage wasn't that big (less than +.5)
I wasn't afraid of Kramnik (we played five games and this game was my only loss) but I simply didn't see how to stop his coming attack on the King's Side while I had no counter play whatsoever. Therefore, in my opinion, the computer's evaluation is totally wrong. It simply doesn't see the coming attack due to the same 'horizon effect'.
Finally, let me show you my game vs. Nakamura. But please don't ask me what was going on between moves 17 and 25.  I am planning to discuss it in one of my future articles. The game reached the position on the diagram and there is only one move that saves White:
I was told that when I played my only saving move, many spectators who watched the game live on the ICC screamed that Serper had lost his mind because their chess engines showed evaluation of -4 or even more.  Of course  as I mentioned in the first part of this article, the computers simply don't understand the concept of a fortress and therefore they wouldn't be able to save the game. So, whenever you use your trusted engine for chess explorations, please remember that computers are a wonderful learning tool, but they have their own limitations!
to be continued...

Comments


  • 17 months ago

    EdJohnson26

    The author again makes the mistake of suggesting that computer chess is pure, mechanical calculation, rather than human-type positional understanding. While this was once the case long ago, it's very far from the truth today.

    This is not 15 years ago when deep blue beat Kasparov with pure calculation.

    All of the modern chess programs have human positional understanding and ideas built in. Now, it's true that humans are still able to envision and execute some of these ideas better than computers. But again, it's only a question of when the computer will be taught the idea on an equal level and beyond that of humans.

    Indeed, as computers get stronger, THEY will begin to teach us positional ideas and concepts. In other words, humans will develop deeper positional ideas and understanding simply studying computer chess played on an astronomically strong level.

    Thus, this obsession over the so-called "horizon effect" is pointless. And will only become moreso over time.

  • 17 months ago

    Ranger_Squad

    Horizon effect only a problem with engines using alpha beta seach algorithm and not with using quiescence search algorithm

    Modern engines   begin to apply their knowledge of the position, the pattern recognition, their intuition, their ability to perceive immediate tactical nuances , their strategical knowledge and their positional insight. They are able to use a sort of goal-oriented thinking to take advantage of the assets or to determine that the opponent’s threats are more dangerous than his own and consequently he has to define a plan of defence.just like humans do .

    and if chess can be solved like checkers than engine can predict the result of game in the beginning  with computational speed only.which could be possible??may be in 10 years or 100 years than..than no heuristics will be required and no horizon effect.

  • 17 months ago

    Gustaran

    In the shown endgame,  Houdini 3 finds 82. Nxe4! after a few seconds and evaluates the positon as a draw. The problem is you don't check your examples with modern engines and hardware. Of course, almost 10 years ago, the computer might have shown -4 and probably everybody thought Ne4 was a huge blunder and your assessment of a computer's capabilities might have been correct. But we don't have 2004 anymore. It's 2013 and programmers obviously have come up with new ways of selecting moves.

    That being said, I wonder why you try to show that humans are smarter. People still watch cycling races, even though every motorcycle is faster. But artificially downplaying a computer's abilities is not getting you anywhere. With things like cloud engines and the chessbase livebook we can expect even greater computer strength in the coming years.

  • 17 months ago

    salowolf

    Actually my computer (stockfish 2.1.1 on a 4-core AMD 64 bit) can see White's advantage in Kramnik-Serper including Kramnik's exact plan providing you give it enough time to reach about 30 ply, which is about 10 minutes.  Not that bad at all, and I would be interested to know how much time Kramnik took on that move Smile

    So it's also about patience.  You cannot expect instant oracle-like answers from the computer but given a reasonable amount of time they can look surprisingly human.

  • 17 months ago

    tripathi7

    please continue, and also put your other games with other great opponents.

  • 17 months ago

    Martin0

    @EricWang, while I guess your statement with "24 extra pieces give a few hunded billion possibilities" was made to make it sound like a big number, I think there are far many more possibilities. One hundred billion is 1011 while we need about 1043 (number taken from this site). For those that likes to write out the zeroes we are comparing the difference between

    100000000000 with 10000000000000000000000000000000000000000000

  • 17 months ago

    kcsmith169

    Enjoyable articles; unfortunate that some people continue to opine that you are making statements about computers or people are better at the game, vice your real point that computers have a different set of limitations.

    Thanks as always for the well-written prose and the excellent discussion of the moves and their intent.

  • 17 months ago

    Ranger_Squad

    Interesting article but it is absurd to compare computers with human.The difference lies in the playing algorithm..The humans rely on Heuristics and intution while searching a move played in similar conditions and than finds  different lines giving him the best position.while computer which relies heavily on the wider search tree and computer can evaluate these tree moves really fast bcoz of processing speed makes them superior on this aspect..but humans usually cut down these trees and have narrow search.which in most cases equates the best move finding method of chess engines. but chess engine performance does not depend only on this computational speed.Now they have been equiped with heuristics like double pawns are bad .knights on edges are dim.and some engines are developing themselves with number of games played by them just like humans do. a chess engine with features of both artificial intelligence and computational speed are now there and are so powerfull .i dont know who is better at chess humans or computer?but i know if chess engine wins than person is to be credited who designed it.

  • 17 months ago

    EricWang2000

    Martin, the 24 "extra" pieces give a few hundred billion possibilities.

  • 17 months ago

    Martin0

    While I agree that computers is worse in some areas like fortresses, I still think it's not something a computer will never understand. Being aware of the current computers (or computers from a couple of years ago) mistakes is important and not always believe a computer is 100% accurate in it's evaluation. Chess is only solved with 6 pieces or less and last time I checked we are playing a game starting with 32 pieces.

  • 17 months ago

    Sutirtha11

    Enjoying the series very much! Please continue....

Back to Top

Post your reply: