- #1

PAllen

Science Advisor

- 9,046

- 2,281

- TL;DR Summary
- Typically, the converse is claimed - Go is harder to solve than chess for (among other reasons) simple combinatorics. But for a perfect knowledge player to play effectively against "near perfect players" I think the reverse may be true.

It is universally admitted that error free play in Go requires

However, the point I want to discuss is that for Go, given the absence of draws (except for rare exotic cases that exist only for some rule sets), and the fact that losing moves can be ranked within the rules (by final point score with perfect play), error free play leads to effective play against a "near perfect" opponent (by the error free player using the hypothetical Go table base).

In chess, this is not true at all based on current conjectures. Even though none of the following conjectures have been proven, for the sake of this discussion of the difference between the games I want to assume the following are true:

- the starting position in chess is a draw.

- there are a very large number of error free move sequences from the starting position for which the resulting position has many (e.g. more than 5) moves that are draws with mutual error free play.

I believe that most top chess players as well most theorists believe these are true, though unproven.

If they are true, then effective play for chess cannot even be defined within the rules of the game. The only objective advantage to one drawing move over another is the probability of provoking an error by a less perfect opponent. This is dependent on the particular imperfections of the opponent, and is in principle unrelated to the game rules without an algorithmic model of the opponent.

*enormously*more information than error free play in chess. To define error free a bit, let us simply posit a complete table base for each game. Though little has been achieved with table bases for Go, in theory the problem is well defined: for any Go position, given the chosen rule set (Japanese, Chinese, etc. ) and komi, what is the final point result for each possible move, given perfect play on both sides thereafter. Then positing the existence of initial position table bases, any player using them plays without error. It is easy to compute that a complete (32 piece) table base for chess is infinitesimal in size compared to a complete Go table base.However, the point I want to discuss is that for Go, given the absence of draws (except for rare exotic cases that exist only for some rule sets), and the fact that losing moves can be ranked within the rules (by final point score with perfect play), error free play leads to effective play against a "near perfect" opponent (by the error free player using the hypothetical Go table base).

In chess, this is not true at all based on current conjectures. Even though none of the following conjectures have been proven, for the sake of this discussion of the difference between the games I want to assume the following are true:

- the starting position in chess is a draw.

- there are a very large number of error free move sequences from the starting position for which the resulting position has many (e.g. more than 5) moves that are draws with mutual error free play.

I believe that most top chess players as well most theorists believe these are true, though unproven.

If they are true, then effective play for chess cannot even be defined within the rules of the game. The only objective advantage to one drawing move over another is the probability of provoking an error by a less perfect opponent. This is dependent on the particular imperfections of the opponent, and is in principle unrelated to the game rules without an algorithmic model of the opponent.

Last edited: