Correlation of FIDE ratings vs chess engines ranking system

[ChessLover]

In 1997 it was confirmed that chess engines became at least as strong as best human players. It is obvious the level of computer chess is now far above classic 2800 FIDE rating. Numerous test runs demonstrate awesome performance for top chess entities (including LC0, Stockfish and Komodo Dragon) of something like 3600-3700 Elo or even more. Checking Elo table states that the difference of 800-900 Elo leaves very small chance for the weaker (human) player to stay alive (should be somewhere near 0.1-0.3% with a chance for a draw at best). I assume, there is no super GM that would play a match from a starting position with no handicap seriously hoping to surviwe. Do you think the difference is indeed that huge? Isn`t there a chance for inflation between human ratings and computer chess engines?

[janus]

In 1997 it was confirmed that chess engines became at least as strong as best human players. It is obvious the level of computer chess is now far above classic 2800 FIDE rating. Numerous test runs demonstrate awesome performance for top chess entities (including LC0, Stockfish and Komodo Dragon) of something like 3600-3700 Elo or even more. Checking Elo table states that the difference of 800-900 Elo leaves very small chance for the weaker (human) player to stay alive (should be somewhere near 0.1-0.3% with a chance for a draw at best). I assume, there is no super GM that would play a match from a starting position with no handicap seriously hoping to surviwe. Do you think the difference is indeed that huge? Isn`t there a chance for inflation between human ratings and computer chess engines?

I think it is - without odds of some sort no human player is now going to beat an engine. In a short while when AI begins to flex its capability its game over and not only for chess.

[massimilianogoi]

To be fair the difference is quite superior. That comparison was made prior to NNUE engines, so now you have to add other 300 points: it's 3900-4000 at its worst... In case of very powerful computers used full force the difference is even greater.

In 1997 it was confirmed that chess engines became at least as strong as best human players. It is obvious the level of computer chess is now far above classic 2800 FIDE rating. Numerous test runs demonstrate awesome performance for top chess entities (including LC0, Stockfish and Komodo Dragon) of something like 3600-3700 Elo or even more. Checking Elo table states that the difference of 800-900 Elo leaves very small chance for the weaker (human) player to stay alive (should be somewhere near 0.1-0.3% with a chance for a draw at best). I assume, there is no super GM that would play a match from a starting position with no handicap seriously hoping to surviwe. Do you think the difference is indeed that huge? Isn`t there a chance for inflation between human ratings and computer chess engines?

[massimilianogoi]

It's mere computations, not proper AI. The mechanic is more similar to conditioned reflexes, none of those programs are self-aware.

I think it is - without odds of some sort no human player is now going to beat an engine. In a short while when AI begins to flex its capability its game over and not only for chess.

[ChessLover]

Totally agree that NNUE made the difference even more perceptible. But lets compare chess with tic-tac-toe. All the computing power in the world will not allow the stronger side to dominate in the game versus more or less competent teenager. That means there exists a top margine of the game`s complexity. Something similar can be observer in simple chess endgames. E.g., FIDE Masters in general case can hold rook endgames with 3 vs 3 pawns on the same side of the board versus GMs (obviously, in some particular case there is a chance to make a blunder in human games). In other words, absence of the practical chances to complicate the game helps more weak players to survice. It means that there is an importance of the openings and once both opponents are following some forced line that ends in a known endgame, there is always a chance for a draw. Do you think that poorly selected debuts (not necessarily losing ones, just with drawing tendencies) is a dominant tendency nowadays?

To be fair the difference is quite superior. That comparison was made prior to NNUE engines, so now you have to add other 300 points: it's 3900-4000 at its worst... In case of very powerful computers used full force the difference is even greater.

[FritzUser]

I think there's a lot that elite chess players could do to unify the ratings. We need data, and they can provide it. Someone with a FIDE rating should play comparable engines. Just like in an engine tournament, where you need to lower the margin of error, the matches wouldn't mean much if the engines weren't beatable.

Granted, if the world were to hand me that data, I'd only be able to run it through Ordo. :-) And I'm pretty sure that even a hundred GMs playing a hundred different engines a thousand times wouldn't provide enough data to connect the MegaDatabase and the CCRL in a way that rating list software would accept.

But that would be so interesting. It would also make for excellent human v. computer databases.

[massimilianogoi]

Nor sure of whar you are referring to... But if you are referring to top humans vs engines, well the chances are very little because nowdays engine analysis are so much sharp that they can exploit even the slightest weakness to get a better position or a gaining combination.

There was an old trick I knew to fight against machines to try to get a draw which I guess it works even now, but one must be very strong anyway: closing the position and exchanging pieces as much as possible.

Totally agree that NNUE made the difference even more perceptible. But lets compare chess with tic-tac-toe. All the computing power in the world will not allow the stronger side to dominate in the game versus more or less competent teenager. That means there exists a top margine of the game`s complexity. Something similar can be observer in simple chess endgames. E.g., FIDE Masters in general case can hold rook endgames with 3 vs 3 pawns on the same side of the board versus GMs (obviously, in some particular case there is a chance to make a blunder in human games). In other words, absence of the practical chances to complicate the game helps more weak players to survice. It means that there is an importance of the openings and once both opponents are following some forced line that ends in a known endgame, there is always a chance for a draw. Do you think that poorly selected debuts (not necessarily losing ones, just with drawing tendencies) is a dominant tendency nowadays?

To be fair the difference is quite superior. That comparison was made prior to NNUE engines, so now you have to add other 300 points: it's 3900-4000 at its worst... In case of very powerful computers used full force the difference is even greater.

[FritzUser]

Nor sure of whar you are referring to... But if you are referring to top humans vs engines, well the chances are very little because nowdays engine analysis are so much sharp that they can exploit even the slightest weakness to get a better position or a gaining combination.

There must be lots of engines that any GM could beat. It wouldn't have to be anything near the top. As long as connections are drawn between the data.