Computer ELO ratings compared to human ELO ratings ?

[Henk]

If it is possible that humans can win from computers in chess games
than I severely doubt if these computer ELO ratings are correct.

If a human player has 500 or less ELO than a computer program than winning chances are practically zero.

I found somewhere:

computer-428 elo=human elo
or some say computer-200 ~= human elo


Of course if you know the weaknesses of your opponent your winning chances are better.

Or is it the undo or takeback button in the user interface that makes humans win from the best computer programs ?!

[hgm]

According to the logistic Elo model you should score 5% against someone who is 500 Elo stronger. So an occasional win would be normal.

[Sven]

I found somewhere:

computer-428 elo=human elo
or some say computer-200 ~= human elo

If you found that "somewhere" then please show the link. However, I don't believe it since it is utter nonsense. You can easily figure out why by also looking up "somewhere" how the ELO rating system works.

[Henk]

http://www.ascotti.org/programming/chess/elo.htm

Yes you're right if the difference in ELO rating is greater than 700 ELO than winning chances are practically zero.

[Henk]

http://www.ascotti.org/programming/chess/elo.htm

Yes you're right if the difference in ELO rating is greater than 700 ELO than winning chances are practically zero.

It can never happen that the best human player wins from the best chess program 6 times in a row or more the chances will be less than 0.000001

[Uri Blass]

http://www.ascotti.org/programming/chess/elo.htm

Yes you're right if the difference in ELO rating is greater than 700 ELO than winning chances are practically zero.

It can never happen that the best human player wins from the best chess program 6 times in a row or more the chances will be less than 0.000001

I disagree.
The best chess program may be deteministic or almost deterministic and in this case the best human player may memorize some comp-comp games and simply repeat the games from his memory.

He may need more than one game because the program has some opening book but memorizing 1000 games against a copy of the program may give good chances to win 6-0.

Uri

[Henk]

http://www.ascotti.org/programming/chess/elo.htm

Yes you're right if the difference in ELO rating is greater than 700 ELO than winning chances are practically zero.

It can never happen that the best human player wins from the best chess program 6 times in a row or more the chances will be less than 0.000001

I disagree.
The best chess program may be deteministic or almost deterministic and in this case the best human player may memorize some comp-comp games and simply repeat the games from his memory.

He may need more than one game because the program has some opening book but memorizing 1000 games against a copy of the program may give good chances to win 6-0.

Uri

I don't know how variation is implemented in these chess programs.
If you add a small random number to the eval() and select randomly if more than one 'best move' candidates have almost identical values.

Maybe if you implement more randomness performance of the chess program drops?

[Vinvin]

http://www.ascotti.org/programming/chess/elo.htm

Yes you're right if the difference in ELO rating is greater than 700 ELO than winning chances are practically zero.

Don't forget the opening preparation that could have a big impact on the results.

[Sven]

http://www.ascotti.org/programming/chess/elo.htm

Yes you're right if the difference in ELO rating is greater than 700 ELO than winning chances are practically zero.

It can never happen that the best human player wins from the best chess program 6 times in a row or more the chances will be less than 0.000001

We simply don't know the approximate ELO rating of the best chess program that it would have in a human ELO rating list, we can only guess. All ratings within a rating list are relative to all other players in that rating pool itself but are unrelated to players outside the pool. This is one of several reasons why today human ratings and computer chess ratings are not directly comparable.

Apart from that, I certainly agree that a 6-0 result of the best human player against the best engine is very unlikely. But the same applies to any other two strong chess players, especially if we consider today's draw rates, so why is that so interesting for you?

[overlord]

It would be interesting to see some of top chess engines participatng in top human events.Anyway,default engines ELO ratings are rather overestimated.