GM Nakamura said that he would destroyed by a computer......

[Laskos]

It's all semantics - you can spell out the conditions in advance and that is not really a handicap. Is playing on a quad a handicap when there are 8 core machines available? Is playing on a small cluster a handicap because you didn't use a big cluster? It's just semantics.

It's almost meaningless unless you focus on the chess part - such as giving pawn odds, etc. Time odds with pondering off seems like a handicap because it's artificial to forbid the opponent to "think."

There is one kind of handicap that I don't if it's been experimented with much. The idea is to give the human the right to N retractions. The rule might be that after the computer responds, the human has the right to retract only his last played move and play a different one. I think even the right to one retraction would be a serious boost in playing strength for a human. It would help address the issue of whether the computer is really consistently outplaying the human or just taking advantage of one or two errors during the course of the game (but I'm not sure there is a difference.)

Well, it's semantics, but the conditions should be clear. For example, to avoid some exploits, the engine without a book, at least in the opening should be a bit randomized. With a book, define the book well, as to not have other exploits. Otherwise, it's a hidden and unwanted handicap, then peolpe will exaggerate their results against Komodo (in general, as an engine). I would imagine you being right that a strong human could beat Komodo in a match only by using these kinds of hidden handicaps.

The book has always been a bit problematic and difficult to deal with. Do you let the human practice and get comfortable with the computers book or do you surprise him with a strong book at the moment the match begins? There is really no such thing as a fair match because there are a million little things that can sway the results in one direction or the other. The age of the players, the time of day the match is played (for older people early is usually better), the location and so many other things.

It's common however that in a serious match the players have knowledge of their opponents playing style and opening preferences. The Deep Blue match took all of that away from Kasparov for example. Giving the human player the full playing system in advance is an advantage for the human.

A good solution might be to develop a strong high variety opening book. For variety you can actually calculate the amount of guaranteed variety that an opponent can force, as least while still in book. You essentially calculate the probability that the opponent can get to any given book position if he plays the right way. You want the worst case to be as low as possible. For example if the computer has 2 choice 5 times along the pathway leading to that position and the computer choice either with 50% probability, the chances are 0.5 ** 5 (0.5 raised to the 5th power.) That is just over 3%. You want to make any final book position to be as low as possible if variety is the goal. (I'm ignoring the possibility of transpositions here, which complicate the calculation but can be done.)

If you can get the guaranteed variety really high you can give the master the program with the book that will be used - he cannot easily prepare killer lines against this book that he is likely to be able to get to. You have to avoid the situation where the opponent can simply prepare half a dozen opening lines against what you are likely to play. Of course you have to take care that the opponent cannot get you out of book with a reasonable position he can prepare for - that might be easier said than done.

I don't think it's hard to prepare comp's "fair" (some common sense fairness) conditions. Basically, what you wrote could be put into practice in a matter of hours. For example, take Fritz or some huge book to move 8 on "maximum variety", avoid certain openings (which lead to blocked positions), during the match check every book move with the engine for some amount of time for more than 10 centipawns drop in score, then, out of the book, randomize a bit the engine, say a window of 5 centipawns, up to the move 15, then play the best moves always. The master can do with the engine + book whatever he wants, say play against it for 2 months. I don't think a master will be capable of beating Komodo in a match in these conditions even on a single modern core.

Kai

ps Forgot about the contempt to avoid the draws, for example, 30cp against a strong GM, 50cp against an IM, 200cp against non-rated.