The next major revolution in backgammon bots
Posted By: Timothy Chow
Date: Wednesday, 9 May 2012, at 12:02 a.m.
For bots to get significantly stronger than they are now, I think they will have to analyze their opponents' weaknesses in real time and adjust accordingly.
This looks like a difficult problem, and I'm not sure if it will be solved any time soon. However, if it is, then my expectation is that the new bots will perform uncannily well against all but the very best human players.
Should you doubt this, I invite you to consider the results of rock-paper-scissors programming contests. If any game is 'solved,' rock-paper-scissors is. However, if you organize a round-robin tournament that includes a sizable pool of weak players (some extremely weak), then it is possible for a rock-paper-scissor bot to perform extremely well. In the contest that I linked to above, each of the 64 contestants played 1000 games with each of the other contestants, so the theoretical maximum score was +63000. The winning program, Greenberg, scored +9268. The "optimal" strategy of playing randomly scored -7. Greenberg never scored worse than -23 against any opponent, and managed to score an amazing +281 against Russrocker5b4, itself no slouch at +4971 overall.
These kinds of results might not be replicated in backgammon, of course. Still, I strongly suspect that a bot that knows your weaknesses and whose 3-ply or 4-ply analysis accurately takes into account how likely you are to blunder during the next couple of moves (as well as during the rest of the match) will be a much more formidable opponent than today's bots are.
Messages In This Thread
BGonline.org Forums is maintained by Stick with WebBBS 5.12.