Explainable chess analysis • page 1/3 • General Chess Discussion • lichess.org

I've used decodechess dot com but was not happy with its analysis result.

What I expected is, under certain situation, a tool can suggest me what are the best, second best... moves, and why.
But not only that, as a human (not an engine/computer), I wanted to know what are the ideas, the strategies behind the moves.

Are there any tools or services that provide these features?

I've used decodechess dot com but was not happy with its analysis result. What I expected is, under certain situation, a tool can suggest me what are the best, second best... moves, and why. But not only that, as a human (not an engine/computer), I wanted to know what are the ideas, the strategies behind the moves. Are there any tools or services that provide these features?

bufferunderrun

#2

Tools? No. Services? Sure, a coach should be able to explain you the idea, if there is one. Or you can try to figure it out on your own.

vuhung

#3

Hi @bufferunderrun ,

As I mentioned in the OP, to some extent, the tool decodechess dot com does a fair job but the question is that any tools can do that.

I believe, with helps from AI, it is possible to explain engines' moves with human-understandable explanations.

Hi @bufferunderrun , As I mentioned in the OP, to some extent, the tool decodechess dot com does a fair job but the question is that any tools can do that. I believe, with helps from AI, it is possible to explain engines' moves with human-understandable explanations.

bufferunderrun

#4

A LLM, or as you called it "AI", can certainly generate some generalizations, but would any of this be helpful? I doubt.

kindaspongey

#5

A few years ago, I was in a turmoil over 1 d4 e5 etc. I thought that I had settled the matter by choosing 2 dxe5 Nc6 3 Nf3 Qe7 4 Bf4 Qb4+ 5 Bd2 Qxb2 6 Nc3 Bb4 7 Rb1 etc., but then I noticed that, after 6...Nb4, the approved move was 7 Nd4.
“How in the world am I going to remember that?”, I wondered. “Maybe I can get by, just using 7 Rb1 as a reaction to both 6...Bb4 and 6...Nb4 ? How bad can it be after 6...Nb4 7 Rb1 ? Ohhhhhhhhhh!”
Machines are not so good at understanding what a human is thinking, but a little human-guided computer exploration can sometimes clarify things enormously.

A few years ago, I was in a turmoil over 1 d4 e5 etc. I thought that I had settled the matter by choosing 2 dxe5 Nc6 3 Nf3 Qe7 4 Bf4 Qb4+ 5 Bd2 Qxb2 6 Nc3 Bb4 7 Rb1 etc., but then I noticed that, after 6...Nb4, the approved move was 7 Nd4. “How in the world am I going to remember that?”, I wondered. “Maybe I can get by, just using 7 Rb1 as a reaction to both 6...Bb4 and 6...Nb4 ? How bad can it be after 6...Nb4 7 Rb1 ? Ohhhhhhhhhh!” Machines are not so good at understanding what a human is thinking, but a little human-guided computer exploration can sometimes clarify things enormously.

BananaBeaver

edited

#6

@vuhung said in #1:

I've used decodechess dot com but was not happy with its analysis result.

What I expected is, under certain situation, a tool can suggest me what are the best, second best... moves, and why.
But not only that, as a human (not an engine/computer), I wanted to know what are the ideas, the strategies behind the moves.

Are there any tools or services that provide these features?

Use SF analysis as follows:

1/ multi PV = 2, because you want to know if the best move is really the only good move to be found, or just a good move among several

2/ when SF recommends a move you don't understand, play it, and click on the red cross hair (show threat). This is extremely useful, it inserts a null move for the opponent and search as if you could play twice in a row.

Note how 2/ combines with 1/ : you also want to know if there is another threat...

@vuhung said in #1: > I've used decodechess dot com but was not happy with its analysis result. > > What I expected is, under certain situation, a tool can suggest me what are the best, second best... moves, and why. > But not only that, as a human (not an engine/computer), I wanted to know what are the ideas, the strategies behind the moves. > > Are there any tools or services that provide these features? Use SF analysis as follows: 1/ multi PV = 2, because you want to know if the best move is really the only good move to be found, or just a good move among several 2/ when SF recommends a move you don't understand, play it, and click on the red cross hair (show threat). This is extremely useful, it inserts a null move for the opponent and search as if you could play twice in a row. Note how 2/ combines with 1/ : you also want to know if there is another threat...

#7

Professor74

#8

Interesting topic. Engines are generally very good at finding the best continuations. But what about the other valid continuations? In my research on artificial intelligence applied to chess I have found that often the "best" continuation is difficult to justify in simple chess terms, and that the other valid continuations are easier to follow. I sincerely suggest you seek professional chess advice for these analysis.

Brian-E

#9

@Professor74 said in #8:

Interesting topic. Engines are generally very good at finding the best continuations. But what about the other valid continuations?

Maybe sometimes they are actually better than the engine "best". Mainstream top correspondence players playing under the auspices of the ICCF, where engine use is permitted, avoid always playing the top engine choice - they lose if they play that all the time - and judge which of the alternative valid continuations are actually better in the sense of giving them more chances.

In my research on artificial intelligence applied to chess I have found that often the "best" continuation is difficult to justify in simple chess terms, and that the other valid continuations are easier to follow. [...]

That doesn't surprise me. I'm sure it doesn't surprise those who are busy detecting engine cheating too.

I sincerely suggest you seek professional chess advice for these analysis.

Seconded! Software cannot (currently) replace the value of human chess tutoring which explains the game in human terms.

@Professor74 said in #8: > Interesting topic. Engines are generally very good at finding the best continuations. But what about the other valid continuations? Maybe sometimes they are actually better than the engine "best". Mainstream top correspondence players playing under the auspices of the ICCF, where engine use is permitted, avoid always playing the top engine choice - they lose if they play that all the time - and judge which of the alternative valid continuations are actually better in the sense of giving them more chances. > In my research on artificial intelligence applied to chess I have found that often the "best" continuation is difficult to justify in simple chess terms, and that the other valid continuations are easier to follow. [...] That doesn't surprise me. I'm sure it doesn't surprise those who are busy detecting engine cheating too. > I sincerely suggest you seek professional chess advice for these analysis. Seconded! Software cannot (currently) replace the value of human chess tutoring which explains the game in human terms.

kindaspongey

#10

Here is an example that I sometimes mention: 1 e4 e5 2 Nf3 Nh6 3 d4 f6 4 Bxh6 gxh6 5 Nxe5 fxe5 6 Qh5+ Ke7, and, at this point, the machine says that 7 Nc3 is best. Maybe so, but the reason does not seem to be readily perceivable by humans, and alternatives seem to be good enough. (This example comes from discussion of a game, but, in the actual game, if I remember correctly, White did not play 5 Nxe5.)