# How many moves ahead does Anand calculate in his mind?

I always wondered about this! How many moves ahead can strong players like Anand actually 'see' during the game? How does the calculation of a supercomputer like Deep Blue differ from that of strong human players?

• I reworded your question a bit, hopefully without altering your intended meaning.
– ETD
Oct 10 '12 at 19:33
• Apparently not enough moves after that huge blunder he made in game 9. Ouch. Nov 21 '13 at 14:23
• Post google's alpha-zero, the answers to this question would likely have to be rewritten, as the approach there is quite different from anything that has been done so far.
– ldog
Dec 19 '19 at 20:45

first of all see here. here is a quote

Asked how many moves ahead he can think, Kasparov replied that it depended on the positions of the pieces. "Normally, I would calculate three to five moves," he said. "You don't need more.... But I can go much deeper if it is required." For example, in a position involving forced moves, it's possible to look ahead as many as 12 or 14 moves, he noted.

for a supercomputer it obviously depends on how well-written it is, there is a concept called alpha-beta, which is used to limit thinking into unnecessary moves, like the computer usually wont consider the opponents (white) `c3`,`b5`,`d6`,`f7`, combo, on the first move. however after `c3`, `b5` it should consider `d6`, and have a reply ready. from a page called Creating a chess engine from scratch on chess.com:

A computer can easily evaluate a few million positions per second, a human probably 1-2 positions per second!! Usually speed is measured in MNodes/sekund which means million positions (Nodes in computer science jargon) per second. Fritz running on my old laptop does about 2.5 MNodes while Deep Blue did about 200 MNodes per second. Raw power is not all - the evaluation function is also very important. Pretty much all engines use the same algorithm for searching through the search tree of possible moves to find the next move. This algorithm is known as alfa-beta search or it is some variant of this.

an important difference is that a computer has to consider things a human takes for granted, such as the concepts behind a known opening, even though a good program can remember which combos work (and rate them, then play based on its ratings), its extremely difficult for it to recognize concepts, such as a fork independent of the result, this can be a advantage because a human may fork, and not see another better future move.

hope that helps!

This is an ill defined question, similar to: what is my girlfriend thinking about when we ... ?

But to stab at an answer, it would be completely dependent on the position. If the position has many tactical variations possible, the answer will probably be very far, 5, 6 or more moves ahead.

If the position is very closed, and positional strategy matters, the answer is probably that he is largely relying on his previous experience, understanding of positional strategy, and analysis of openings/other tournament games in guiding his moves. This usually means he has memorized the possible variations and can move with confidence knowing he is not blundering.

If the position is considered an endgame position, once again, he should be able to see very far ahead since the number of tactics and variations is greatly reduced.

This is mostly true of all chess players, the difference being that master and grand master players can perform all of these things to a greater degree than lesser players.

In terms of how this applies to the development of a Chess Engine such as (Deep) Blue: it doesn't. Computers are largely computational beasts and other than large databases of known openings/endgame position and transposition tables they don't largely rely on previous experience. They simply search for the best move as defined to be the move such that the opponent's best (following) move is weakest (this is the principle behind the min-max search algorithm which is usually used in Chess Engines.) It is a well known trick that when a human is playing a strong computer opponent they should aim to create a largely positional game with few tactics in order to increase their chances of winning. Computers make far fewer mistakes than humans in calculating tactics and, in general, have poor positional play.

• +1 for the first sentence, although a more accurate simile is "how many moves in advance does my..." Sep 3 '13 at 4:11

Just thought to add the famous story (likely apocryphal):

During a tournament in the 1920s a newspaper reporter asked Richard Reti how many moves ahead he could read. Reti replied "I only see one move ahead: The right one."

• `+1` i should find a opponent like that, lol. Sep 25 '13 at 21:08

Players such as Anand and Carlsen can play top-level blindfold chess.  I assume this means that the number of moves ahead such players can "see" is essentially unlimited: at the board they can presumably visualize a continuation of the game to its conclusion.  But a single very deep search down a single branch of an enormous game tree, though very important in certain situations, is not in itself enough to produce the best moves (and also may not be the best form of time management).

• +1 While the answer may sound like nitpicking, I think many non-chessplayers simply don't understand how a grandmaster can see many moves ahead, and the question often means "how many moves can a player follow a line without moving the pieces before he starts forgetting the position" and not for example "how many moves ahead are they calculating while thinking about a move for 20 minutes".
– JiK
Nov 20 '13 at 23:19
• I only answered the first question, since an answer to the second could run to book length. Regarding the latter, the OP might get some insight from Feng-hsiung Hsu's book "Behind Deep Blue: Building the Computer that Defeated the World Chess Champion." Nov 21 '13 at 0:28

How far ahead a strong player like Anand sees depends on the position. On move 1 he doesn't see any moves ahead because he doesn't know which of several perfectly good replies his opponent will play. In the endgame there might be a forced line which is 15 or 20 moves long which he will see and so, by the way, will many weaker players.

The two significant differences between a strong human and a supercomputer are:

1) A supercomputer can calculate much, much faster than any human and so can consider many more moves (orders of magnitude more) than any human in the same time.

2) Very strong players are much better at evaluating a position than supercomputers. This means that they can quickly reject unpromising lines and so dramatically reduce the number of variations they need to consider and do this much better than the supercomputer. Effectively they can often see as far, sometimes even further than the supercomputer.

Incidentally, this superior evaluation ability separates strong grandmasters from not only supercomputers but also ordinary players like you and me, or at least me ;-).

Weaker players are often not that much different in their calculating abilities from stronger players. They just don't recognize whether the resulting position is good or bad anything like as well as the stronger players do. Their poor evaluation abilities mean that they waste time, just like the supercomputer, analyzing unpromising variations.

• Good answer, but I wonder how much of their ability to evaluate is based upon analysis that they have previously done off the clock.
– ldog
Apr 13 '15 at 21:38