what is the % of chances of grandmasrer winning chess game against computer?

hi friends!

please give your views and opinions

1) what is the percentage (% ) of chances of grandmasrer winning chess game against computer?

for example:

if there are 10 matches to be played between grandmaster and a computer programmed chess software,

will the GM win, 10 out of 10, or 9/10, or 8/10..... ?

thanks.

Depends on a few factors.  Firstly, how strong is the GM, there is a big difference between a 2500GM and Magnus Carlsen.  Secondly, what is the time control?  In classical time controls no human has any chance, in a correspondence game it's another matter entirely.  

I think Carlsen (and probably many other players) wouldn't have much trouble beating an engine in correspondence chess.

You only have to plug into this formula ( the 'Glicko' formula ) the rating of the GM ( r ), the rating of the engine ( rj ) and the rating deviation ( RD ) of the engine ( RDj ) and you have the expected value of the score E ( s | r , rj , RDj ).

The only problem is that not many games are played between high rated human players and top engines under tournament conditions. This means that rj is very uncertain. [ You have to choose a high value for RDj because of this uncertainty ( possibly the recommended maximum of 350 ), this is for certain. ]

A numerical example:

Take r to be the rating of the highest rated human player ( that of Magnus Carlsen ) 2872. The rating of the engine is uncertain so let RDj be 350.

If the engine is rated rj = 3000, then the expected value of the score:

0.379

This means that a 3000 rated engine would beat Magnus Carlsen 6 - 4 or 6.5 - 3.5 in a match of 10 games.

Also, the ratings of engines are highly speculative.

Some (engine makers) throw out numbers like 3300-3400 ELO, some study I read guesstimated Houdini at slightly above 2900.

Also, the Ivanov dude, who played Houdini's moves against several (2500) GMs did not win all the games. Several draws and maybe even a loss or two.

Good point. 3300-3400 is too much for my taste as well.

Yeah the problem with using that formula is that no engines have "real" elo ratings because they don't play FIDE rated games.

Engine elos come from a separate playing group so its hard to say exactly how strong they are.

This is true. My personal opinion is that on a high end commercially available hardware ( not a supercomputer ) the current engine world champion without tablebases can be around 3000.

Super GM's like Carlsen and Kramnik can draw all the games in a short match like 6 games. 10 games is long and fatigue would wear the humans. My score engines = 6.5 SGM = 3,5

what is correspondence game ?  and why there are chances of GM winning correspondence games?

thanks for reply.

where did u get this formula from ? any other simple formula or method to reach the conclusion ?

seems to be very complicated and confusing.

thanks for reply.

Im not doing this mathematically but I think it depends on the type of grandmaster and the type of computer

A bit like karaspov and deep blue

Correspondence as in you have multiple days for each move.  When you consider how accurate Carlsen is already (he already makes better moves than the engines quite often in classical chess) you can easily imagine that he just isn't going to make mistakes given a week to analyse a position, this is probably true of most 2700+ players.

The Glicko formula is explained by the author Mark Glickman here:

http://www.glicko.net/glicko/glicko.pdf

If you want to translate playing strength into rating points you have to use the either the Elo or the Glicko formula ( this latter is used here on chess.com ) so this won't get much simpler than this.

If you want to use no formula than you can simply say something like 'my feeling is that Komodo will beat Carlsen 6 - 4 in a 10 game match' without referring to any rating calculation.

What would be handy is to get one of the top engines to participate in human tournaments for a year or so (not as an actual competitor for prizes just as an experiment) to get a real rating then it could take this back to computer chess competition and you would get a better idea of their relative strength.

I wonder why this is not happening already.

There must be some fundamental reason since strong engines have been around for quite a long time now.

Probably top players don't really want to play serious games against engines.

It could maybe work if they got cash bonuses for beating/drawing the engines and the engines were not included in the final tournament standings.

They also could have their rating adjusted if this means an increase in their rating ( but leave it unchanged if they would lose points ). This way they would be interested in the result but would have nothing to lose. The rating of the engine would be adjusted always of course.

As long as even a 2200 or so can still specifically study techniques to atleast draw engines very frequently (close the position) ofcourse they are not 3200 Elo or whatever.

Playing engines is a whole different game than playing humans. Ofcourse no top ten player will seriously prepare or adapt his style to do so because most likely it will only hurt his human chess?

But if taken seriously some top ten players (some ofcourse are much better suited to do so than others) could probably win fairly frequently. Perhaps 1-2 games out of ten or maybe even a bit more. Engines still don't "learn" very well. But it will remain theory because no one will do so as it's rather pointless. And the rest of us don't play chess well enough.

Not sure about that because that would artificially inflate players ratings (since they can go up but not down).  I think it should be an unrated game for the human but a rated game for the engine perhaps.