Relying on the Bots

The position below came up in a Chouette (money game) recently, and White had to play a 6-1.  I was White but had a partner.  Before I go on, why don't you see how you would play the 6-1.





If you are like me, and most people, you would certainly hit Red's checker on the 9 point.  You may be expecting me to try to surprise you and tell you that's wrong, but that is right.  The question was what to do with the 1.  I wanted to simply play 6/5 and have the most builders.  A case could be made for 9/8.  But my partner, a very gifted and inventive player, wanted to play 3/2*!!!!

He made a pretty compelling, passionate please for that play, saying that we can destroy Red's timing, maybe get some more checkers back, and even though he admitted it could cost us some games, he felt certain it would win us a lot more gammons.

The play just looked too nutty for me, so as is our custom, I told him I would agree to his play only if he would give me a side bet on the play.  He declined, because he said he wasn't positive enough to bet, but we took a picture of the position and put it in to ExtremeGammon after the game, and I was absolutely shocked to see that his play was only a .020 error according to XGRoller+.   XGRoller+, according to the latest studies, is the most accurate reading you can get anywhere other than a full rollout.  If you look closely (below) at the evaluation, however, you will see that my play is better because you win just a few more games, but you really don't win more gammons by hitting off the 2 point.


XG Roller +
My friend felt reasonably vindicated, but he didn't stop there.  He was quick to point out that when the difference of plays is this close, a rollout could easily change the order and magnitude of the plays and his play might even be best after all.  Of course, I mentioned that it could go the other way.   So, we did a full (1296) rollout of the top 4 plays, and 20 minutes later returned to the screen, and this is what we saw:


Full Rollout

The .020 error of hitting off the 2 point had transformed itself into a major league blunder-- .158!

If you have been a visitor to this web site before, or if you know me personally, you will also know that I helped Xavier along the way in testing and making recommendations to improve ExtremeGammon and that I am an "official promoter" of the program and one of it's 4 "endorsers" on the EX web site (along with Neil Kazaross, Mochy, and Chiva).  Most agree it is the best backgammon program in terms of accuracy, speed, ease of use, and graphics, and I go a step farther and will tell you that if you use ExtremeGammon you will become better looking as well (it worked for me). 

But even with the best program, if you really want to be sure of a checker play or cube decision, YOU MUST ROLL IT OUT.  And that is your lesson of the week!

This particular example is extreme.  Most of the time you will not see changes this big from the XGRoller+ to the rollout.  Most of the time, you can be relatively confident of the answer, but all it takes is a few examples like this to remind us that you should not hang your hat on anything short of a full rollout.   By the way, even then, there are STILL some positions that the bots may be proven to be wrong about in the future as the bots and computers become more sophisticated.
Comments