How good do chess engines play chess?

Score 6.5 does not mean "white is going to win" - that's just your interpretation. I'd rather interpret it as "if black doesn't play the only best move every time he's going to lose"

I think your interpretation is wrong. Computer evaluation is entirely objective, and doesn't take into consideration at all if "the only best move every time" is required or not.

Edit: To be clear, if the computer finds a line that's a series of 45 only-moves for one side, where all other moves lose horribly but these only-moves draw, then the evaluation for this line is 0.0.

It's strange that stockfish prunes this so agressively. The developer should take note and re-tune I think!

Fantastic move btw, a real beauty, or a real joke - I don't know which :)

Qd4 would be a pretty good move after Rf2, unless your opponent catches you :)

I still don't see how from score 6.5 you conclude that "white is going to win". It just means Stockfish thinks white has an advantage which is in a way true because if you count all the possible variations there are more of them that lead to win for white. Stockfish doesn't see that in the worst case scenario the draw can be forced via 50-move rule because it will mean calculating to 100 ply deep.

Hence we agree, because the rook on the fourth line is moved to the second when playing f2, making it impossible for white to capture the queen on that square. :-)

Yes, the pruning is too intense. Fritz 6 finds the winning move very fast, also.

uri65,

you have little understanding about computer chess.

If any engine shows 6.5 or more than 1.00 that does mean one side is going to win.

Suppose,

if there any position, where white has extra rook but black gets perpetual check , computer will show 0.00 . That does mean 'it is draw' and it will never show 5.00 for material surplus.

This is a draw position and computer will show 0.00.The score does not care about so called material difference.

 But you said , '' 6.5 is showing material difference and it does not mean win".

Do u get it ?

You say: "If any engine shows 6.5 or more than 1.00 that does mean one side is going to win." From where do you get this idea? Please provide some sources for your claim.

In which post did I talk about "material difference"??? May be try to read and understand what others are saying before you answer.

see my given example, everything will be clear to you.

What is advantage ? computer does not know ' what is advantage?'.

Computer evaluation is based on material occurence. Without material recognition , coding is blind.

In your last example black forces 3-fold repetition - that requires just 6 ply depth to see. Contrary to this your first example draws via 50 move rule - that's 100 ply. Apparently Stockfish doesn't see it's a forced draw.

I didn't quite understand your question about advantage. Chess engine uses evaluation function to assign a numerical value to position. The rest is done via pruning and minimax algorithms.

"Without material recognition , coding is blind" - I don't have any idea what you mean here.

You didn't answer 2 of my questions:

1) please provide some backup to your claim that evaluation of 6.5 means white is going to win

2) please show in which post did I talk about "material difference". Or just acknowledge that you did misquote me

@earth64: wow, that is unexpected that the pawn blocks the queen to escape. I saw the idea of the rook capturing the queen and had the idea that the a- and b-pawn were unstoppable by the king, but I would never see that the rook sacrifice would effectively block the queen.

Computer evaluation is like the standard deviation of a set. It's not just material different. It means something, but we don't quite know what it means. We know it's important, but we don't know why it's important nor can we rigorously define how to interpret an evaluation.

Rembmer that in any chess position, there are only 3 possibilities:

- With perfect play, white will force a win.

- With perfect play, black will force a win.

- With perfect play, it will be a draw.

If a computer was perfect, it would abandon evaluations altogether and simply use one of these three states. There is no difference between "+6.5" and "+1.1" if both advantages are enough for white to force a win.

So as other people have said, giving an engine a forced draw and having it spit out "+6.5" does not mean its stupid. Just because the engine says that one side has an advantage doesn't mean that it thinks that side is going to win. In fact, it has no opininion on what side will win unless it sees a forced check mate. It is simply uses its formulas to make an evaluation of the position.

So in the previous example when somebody claimed stockfish was inferior because it claimed "white is going to win", that is a leap in logic. Stockfish did not make any assumptions or share any opinions about whether or not white was going to win. It just said that white has a +6.5 advantage.

Here is an intuitive way to understand this:

If stockfish evaluates +0.1, does that mean it believes white will win?

Everybody who understands how engines work would say no.

What about +0.2, is that big enough?

Still no.

+0.3?

+0.5?

+1.0?

+1.5?

At what point does the evaluation turn from "advantage" to "I think this side can force a win".

The answer to the paradox is that it never does. Unless stockfish is displaying forced mate, it has no opinion about who can win the position or who will win the position. This is often why people critisize computers in the opening too- because they take the evaluations more seriously than stockfish takes them.

Stockfish uses the evaluations not to predict the future, but to find the best move. In the example where stockfish displays +6.5 in a drawn position, you mean critisize it for giving an "incorrect" evaluation... but does it make the best move? Does it preserve the draw? If it makes moves that preserve the draw, then you can say that it "understands" the position as a draw as much as a human does. It is the moves it makes and not the numbers it outputs which determine how well it understands the position.

The only way you can claim computers don't understand positions is to give them a positional draw which is easy for humans to see, and then show that the computer makes the wrong move which does not maintain the draw. Reading outputs will never prove anything about the playing strength or creativity of computers.

So can anybody show an example of a position which is trivially, or even not so trivially drawn, and then is misunderstood by stockfish, making it decide on a move which will give up the draw?

I don't think this is possible. Read the moves, not just the evaluations.

EDIT: My entire conclusion is incorrect, such positions do exist, and they are amazing! Eventually computers will get better at pruning, but some lines look so ridiculous even to computers that they get pruned prematurely. Thanks for helping me learn something today.

However, the rest of my post is still useful to read for those who do not understand how chess engines work. I simply stated what I would need to capitaluate my point and then that evidence was provided, so I fully believe that computers can in fact lack positional understanding.

It was a game when Rybka gave-up a 50 moves draw, trying to win, but it was defeated.

xman720, thank you so much for explaining the meaning of computer evaluation in such clear way!

Good post xman720.

It's true that engines are tuned with the goal of finding the strongest possible move in every position, and not necessarily to give an evaluation in form of a numeric value that makes sense to a human.

Xman and uri,

Tell me please.

a.What does it mean  by advantage?
b.Why do engines provide evaluation , though they go for pruning ?

c.What is the significance of 6.5 scored advantage? [Even with 6.5 scored advantage, it is a draw position.]do you consider it as advantage?

d. What is the significance of 0.00 score for the 2nd position [scored by stockfish], though it is winning position for white ?

An engine can force a draw regardless, and does NOT see any way for white to win, so there is really no difference there from programming point of view. However, it might be possible to guide an engine towards more human-like play, by selecting a move that leads to faster capture of a piece.

I suppose the engine authors did not see much need for that.