Back to "The Zare Formula" written by Douglas Zare

New Backgammon Positions

by Douglas Zare
1 February 2009

Douglas Zare

Backgammon remains exciting to experienced players and comprehensible to beginners because most games have both easy and tough decisions. Several moves in a row may present simple decisions, but a few decisions per game are difficult even for experts, whether we realize they are difficult or not.

It is likely that the person with the most backgammon experience ever is still alive, and indeed, may be under 40. While the checker play rules of (international) backgammon have not changed much for hundreds of years (see Hoyle's 1745 book on backgammon), computer backgammon allows us to play much more, both against other people on servers such as FIBS, and against bots. I've been playing backgammon seriously for just under 10 years, but I've played tens of thousands of games. It would have taken several decades of live play to get as much experience before computer backgammon. And yet, I often encounter decisions I recognize as new, or even of new types, and I believe this is true even for those who practice more intensely than I do.

The rest of this article (15.10 K) is premium content. Please subscribe below.

GammonVillage Subscriptions:

You must be signed in to rate articles.

Feedback about this article:

Subject: Re: New Backgammon Positions

From: zare

Date: 04 Feb 2009 03:06 EST

Stick Rice pointed out by e-mail that after an opening 2-1 slot, 24/20* might not be the right play with a 1-1. It's an interesting decision, and Snowie and Gnu Backgammon evaluations disagree significantly. Whatever the right play is, the (third) bot I was using for the experiment was hitting with 1-1 there on the level I used for the rollouts, leading to more repeated positions in early rolls.

Douglas Zare

Subject: Re: New Backgammon Positions

From: mochy Member has a photo available

Date: 04 Feb 2009 22:15 EST

You mean opening 41slot ? I think both program agreed to make both points with 11 after opening 21 slot

Subject: Re: New Backgammon Positions

From: zare

Date: 05 Feb 2009 18:13 EST

It could be that I'm not using the latest nets for Gnu backgammon. The one I'm using (0.15) hits with a 1-1 response to a 2-1 slot on 2-ply. The bot I was using for the experiment also hit.

My own rollouts didn't separate the plays, putting them dead even after 3600 trials, with 6/5(2) 24/22 back by 0.010. Stick Rice provided me with the results of a couple of rollouts: one favored 8/7(2) 6/5(2) by 0.031, and one favored it by 0.012. The 0.031 was obtained using a method which is biased toward overestimating the difference between plays.

Douglas Zare

You must be signed in to post comments.