Odds for man vs machine

[Onno Garms]

There has been much discussion if the match Kramnik vs Deep Fritz settled the question if mankind or computers are stronger at chess. I don't want to restart that. It is a fact people, esp. Rybka supporters, have started to play at odds. Some odds seem strange, others suggest themselves. Which odds can you imagine? We have heard of:

- Time odds: suggest themselves, but soon will be not sufficient.

- Book and Tablebase limitations. Cheating on the computer side is difficult to prove. Also, this is disallowing certain techniques for the computer and seems artificial.

- Pawn odds. Funny, but won't help to make man vs. machine more attractive to watch. No world Champion would accept a match at pawn odds IMO.

More ideas? I have some.

- Some years ago, the magazine "Computerschach und Spiele" reported in the April edition, that Fisher invented a new chess variant that is more taylored to human playing style. It was a longish rule enhancement. The joke about that (for April fool's day, Fisher did not invent this rule) was that after reading carefully you noticed that the rule enhancement amounted to allow to take a move back after the opponent's reply. What's so foolish about that? This would help men a lot even if we give the computer the same right.

We even could allow to take back any number of moves. To limit the use of this rule, we could devise that the opponent gets his time refunded but the player who takes back moves does not

Over all, still equal conditions for human and computer, but in spite of this much more useful for humans.


- When playing move n, the computer must enumerate all possible replies and its following move for each of them. On move n+1, the computer is forced to follow this line. Thus the time is only used to compute the replies to all the player's next moves. By this rule the computer loses two plies in depth. That is a fairly heavy burden.

The human opponent should be able to see the list. That makes it easier for him to spot the weaknesses in the computer's thoughts.

This rule is scalable. For weaker humans the computer might be forced to show trees of depth n in advance. This obviously requires the human to have access to a tree browser, but that shouldn't be a problem. A tree of 5 moves fits easily into the memory of a currenct computer.


Do you think matches with the rules above might be interesting? Have you other ideas for interesting odds?

[Ron Murawski]

When I play against a program -- to make things more equal -- I work out of an opening book: either NCO or a database. Once the engine leaves book, so do I. Despite my "preparation" I lose anyway!

Ron