# Are there positions which cannot be properly analyzed by any computer?

Are there positions in chess which every engines is unable to analyze properly ?

Maybe calm positions with a positional advantage the engines do not recognize ? Or positions won in the long term but with no concrete winning moves ?

``````8/3k4/2p5/p2p2p1/P2P2P1/2KB4/8/8 w - - 0 1
``````

In this position the engine gives advantage of around +2.5, but as you can see, this position is a draw as there is no way for white king to enter into Black's side of the board to penetrate any of the black pawns as the black pawns control all the accessible squares which are b4, c4, e4, f4 and h4. Even though white has an extra bishop it can only attack 2 of black pawns which are on c6 and d5. Taking the d5 pawn is recaptured by c6 pawn or taking the c6 first can be defended by the black king.

This is just an example of a position where an engine can give unfair advantage to a particular side though it isn't such. Here I'm not supposed to say the engine is making a mistake because it is programmed in such a way that it sees an extra bishop and with limited depth which is below 50 it says white has the advantage of extra bishop with no strings attached, hence gives you +2.5, but if you let it think up to the depth of 50 from this particular position (for which engine might take long time), it then shows it's a draw (0.00) because of the 50 move rule which is programmed into the engines.

With many examples as such, a simple concept of long term strategy can differentiate human brains from engines. As said by @jknappan in the first answer- openings are very hard for engines to evaluate, because the engines are purely based on mathematical calculations and it does not have it's own opening repertoire and humans do have because of the long term strategy. A queens gambit or Sicilian is studied for a long time by great chess players and they understand what kind of middlegame it will turn into and how then it can turn into a playable advantage in the endgame ( unlike the above position though by having an advantage of 2.5 and yet not able to achieve a win in the endgame). These kind of long term strategies is not thinkable by chess engines. One more concept of bishop pair can be taken into consideration to explain the difference between human thinking and chess engines. A good chess player would not wanna give up his bishop pair in the opening unless it gives him a very good advantage. But a computer would suggest you to play and give up the bishop pair even when it's calculations gives the slightest of 0.01 worth more of marginal advantage from its second line! These are some of the positional strategies chess engines might miss to analyze but human studies do not.

I just saw this example composed by Jim Plaskett:

``````[FEN "8/3P3k/n2K3p/2p3n1/1b4N1/2p1p1P1/8/3B4 w - - 0 1"]
``````

It's a famous endgame problem.

I left Stockfish running for a few minutes and so far the best move it has found is 1.d8Q with an evaluation of -1.5. However this is not the best move.

The best move is 1.Nf6, which after a few minutes has an evaluation of -4.5, but is actually winning for white.

See here for a full analysis.

• This endgame problem is also featured here. Dec 18 '18 at 13:50

In the following position, most engines give the move 1... Rxc4? an evaluation of +2 or more. But a human analysis shows that this move results in a draw.

Black to move

``````[FEN "6k1/5p2/1Pp3pb/2q4p/1rNpPP2/3Q2PP/8/5RK1 b - - 0 1"]
``````

The correct move is 1...Qxc4! and it is typically given a lower evaluation by engines.

Human Analysis

What went wrong?

The problem with 1... Rxc4 is that it gets an immediate material gain, but white can force a draw due to the dangerous passed pawn. The correct solution gains less material but results in a won end-game position. Engines can easily see short term won material, but it is difficult (time-consuming) for them to calculate a long sequence of moves that lead to a won-end game.

Could engines ever solve this position?

Sure, engines are improving all the time. But there are trillions of positions that engines get wrong, and this is just one example. Every year computers and chess engines get better. Each year they can solve a few more positions correctly that previously they got wrong. But until chess engines are perfect (i.e. never), it will always be possible to find some positions where the engines get it wrong.

• As of April 27th, 2020, Stockfish 11 on lichess doesn't even put Rxc4 in its top 5 choices for moves for black in this position. Qxc4 is in the top 5, but Stockfish 11 suggests that Qb5 is the best move for black. It's probably right :) Apr 28 '20 at 2:40
• It finds the forced draw for white after Rxc4, as soon as the depth reaches 17. It does not find the forced draw with a depth of 16 or lower, but it takes about 3 seconds to reach a depth of 17+, and I'm just using a laptop, not some supercomputer. Apr 28 '20 at 2:42

There is a trivial looking answer to your question: The starting position cannot be properly analysed by any chess engine. To play decently, all extant chess engines rely on opening books prepared from games among human players.

• I would prefer a position where the computer is clearly wrong. In other words, a position, where the computer valuation is far from the correct one. The value of the initial position is not known but considered to be equal, and most programs show about level, so the computer is not wrong in the way I mean. Nevertheless, I upvoted the answer. Mar 28 '15 at 18:14
• I disagree. If you let an engine play the opening without a book you will find that it is still very strong. Apr 12 '19 at 18:52
• @qudit Things have changed since 2015 when I wrote that answer. Alpha Zero is able to build an opening book from self-play now. Apr 13 '19 at 18:10
• @jknappen No doubt Alpha Zero would do better, but even an ordinary modern engine can play a respectable opening these days. Apr 13 '19 at 19:18

As a general rule, unbalanced position where common sense tells you that it's a draw, are hard for chess engines:

``````[FEN "3b4/1k6/8/p1p1p1p1/P1P1P1P1/8/6K1/8 w KQkq - 0 1"]
``````

Any engine will probably give a huge advantage to black in this position.