How good do chess engines play chess?

It is obvious, that if the chess engine is allowed to use his brute force technique optimally, that chess engines beat the best of the GMs all the time. They see combinations people can only dream of.

Their rating is several hundreds of points above the rating of the super GMs. So, on a first glance, chess engines play chess better than the best. But do they really?

What is a good situation to measure the actual strength of a chess engine in comparison to a human chess player? Humans play better chess too when they are allowed to use brute force technique (getting a lot of time to analyze the game - with or without the use of a chess engine). They will play better if they have more time. Just like chess engines.

The best chess is played in Centaur Chess, that is the combination of human and chess engine. Hence, the presence of a human strenghtens the chess played by a chess engine. That is weird. Think about the analogy that I (rating here 1866) would play a game of chess with Carlsen as my second and as a result I actually improve the play of Carlsen?! Yet that is all the time happening in Centaur Chess. So the question is:

Do chess engines really play better chess than humans do and what are the settings of a chess engine to compare its power of playing chess with humans objectively? It will be always comparing different strategies how to play chess, but nevertheless what should be equal circumstances to compare chess engines and human players. And if those circumstances are settled, who will win?

Not too bad.

"Centaur control" helps you win games.

That's not really deep analysis, though.

I appreciate the selfreference in your post. :-)

EDIT: I was referring to post #2.

@Khzyx: upgrading is not where I tried to start a discussion about.

@astronomer999: what do you mean exactly?

LockBergman

Chess engines already have been measured for playing ability and already have a rating.  Yes, they play better than the best humans. Why? Because they can analyze and look at millions of positions during the time a supergrandmaster looks at a line with maybe 20 possible main positions.

I am not sure why you ask if chess engines can play better than humans? Is'nt it obvious they can?  One way to compare chess engines vs humans is to have them play each other.  This most humans will not agree to.

I really do not understand why you are not sure who would win in a competition humans vs chess engines?

Yes, if you were to help Carlsen by playing a game--you would not be a help to him at all as he is a far better player and playing you [and the vast majority of players] would just be an unwelcome distraction. If you were to try and help Carlsen with an opening variation he would be far ahead of you.  However Centaur players can help chess engines with opening variations.

However, Centaur Chess is different.  The very best Centaur players know that the chess engines are very powerful.  But they can "guide" the chess engines.  If a particular variation keeps running up as not so good they can force the chess engine to look at some other variations.  If the human sees a possible sacrifice but it is too hard for the human to evaluate--he can just feed the position to the chess engine.  In other words the human can suggest variations and cut down on variations which do not work.

Thus in Centaur Chess the humans might add 100 -150 rating points to chess engines alone. 

By the way, what I described what humans can do is not the whole of what they can do. There are other ways humans can compliment chess engines including looking at past games and putting in the best opening theory.

Dey playz reeeeeeeel gud.

But by referring to his post, which refers to your own post, you have now referred to your own post.

@ponz111: The point I am trying to make is that when you would look only at difference in rating points you would expect that the one with the vast lower rating would not be a help for the one with the vast higher rating.

Yet, that is happening in Centaur Chess. I can only explain that if the lower rated player had good ideas that the higher rated player was not capable of seeing. That is a situation that would never occur between two humans with a huge rating difference. But it does happen when a human and chess engine play together.

When two humans with huge rating difference play each other, the higher rated will understand all positions better. I have understood that chess engines are not at their best playing in closed positions and should not be followed blindly in the opening. That is a little bit odd, I would say.

That is why I would be interesting to know if it is possible to let chess engines play with settings that make it possible to compare the 'understanding' of chess engines and the 'understanding' of human players. (Both type of players 'understand' chess in a different way.)?

If you let chess engines play with enough ply etc.., yes, then is it clear that they will outplay the best human players. I never doubted that. It is like asking someone if he could go faster then Usain Bolt. Uh, Vettel said yes.

It might give a good measurement of the quality of the program sec instead of the quality of the program when using dope. There is nothing wrong with a chess engine that would have a human rating of 2000 and a full engine rating of 3300. It might inspire the programmers to enhance their program even more.

Man...when you crack subtle jokes about playing with the horsey thing you expect a little titter from the crowd don't you?

@wafflemaster: we agree.

@Khzyx: It is good that you laughed, because that was my intention. :-) Of course I knew that you were referring to my post, but why would you do it then? There is no obligation to react. I guess you will let a lot of other threads unread, hence you have a reason to react. It might be interesting what you have to say about the subject. If you do, please use decent language.

Let an engine play a quiet position. I'm quite sure that an experienced player finds the best move much faster than an engine would.

I am not sure what you mean. Are you referring to the remark of Khzyx? If talking about an upgrade was a subtle joke, then did I completely miss it. Still don't see it. I talked about program settings. That is within a release.

You might say something else. I don't know. Please explain.

You should keep tatoos of Foucalt, Derrida and Badiou handy...

@manfredmann: Nevermind :-)

@ultimateichigo: I believe that too. I remember that at the candidates this year Carlsen made moves that Houdini had not evaluated. I thought it was in his game against Gelfand.

First of all... as has been repeated a billion times... ratings are relative to the pool of players.  Computers don't compete in human events.  The ratings are not directly comparable.  Computer ratings tend to be a bit inflated when compared to FIDE ratings from what I recall.

To answer your original post:

The computer plays better chess overall i.e. it makes less mistakes on average.  In a man vs machine game the machine will win because it makes fewer mistakes.

The method of play between computers and humans is different so you can't compare centaur chess to you+Carlsen.  Centaur chess combines the calculating power of computers (near perfect to perfect play in the short term) with the experience and knowledge of the human (play very strong in the long term).  The computer benefits because it's literally blind to the long term.  The human benefits because it's impossible to check millions of variations.  In the you + Carlsen combo, you can't help Carlsen.

Your final question, how to put humans and computers on equal footing to compare them.  Give a GM the ability to calculate millions of positions per second and give the computer the ability to draw on past experience and knowledge to relate lessons to similar positions.

Alternatively, take away the computer's ability to calculate and take away the human's knowledge.  i.e. a beginner vs a very weak program.

Which is to say, there is no good way to compare them presently.  The methods of play are too different.  This is also why you'll often see the advice to avoid playing the computer as practice.  It doesn't play like a human and your experience is worth much less than if you play people instead.

@spider-waffle: No doubt that chess engines on full power outplay GMs. No doubt that GMs will outplay chess engines when the engines are crippled by program settings. Let them think only for 1 ply for instance and the GM will win.

In the first setting has the chess engine an equipment that is beyond human capacity, in the second setting has the chess engine an equipment that is far below human capacity. There must be some setting in between that is reasonable to let chess engines compare with humans. Or is it not possible and is it impossible to compare the chess capacity of humans versus engines?

To answer that question you'll have to define "chess capacity" for us.

If you simply mean the ability to play chess then it is impossible to directly compare them because the method of play is entirely different.

For each player I bet you could find a setting so that the score will be very close to 50% in man vs machine.  But no matter the restrictions you use for the computer to achieve this, the method of play is too different so you can't make a good comparison.

If "chess capacity" means the ability to understand chess, then computers lose because they don't think in the first place.  They preform sets of instructions quickly.  The human devised the method of play but is too slow to preform the calculations so the computer does the fast calculations and plays from the human instructions.  In this sense when you watch Houdini play chess what you're really witnessing is how the team of programmers have chosen to play a game of chess.  The computer is the tool that preforms the human mandated calculations quickly enough.

@wafflemaster: our posts crossed each other. Thank you for your response.

Until now did I not know that ratings of computers and humans are uncomparable. Well if the ratings are uncomparable, the method how to find moves are uncomparable and there can be no standard setting defined in which chess engines should have a human like capacity, then is it not possible to get a decent answer to my question other then 'they play very good'.

If you think about it... human vs computer is really just human vs human.  The human player at the board is playing chess using his brain.  The human player away from the board is playing chess with a set of very specific instructions.  The computer follows this list and makes the move for the programmer.