Programming

 Anticomputer positions

 From: Bill Taylor Address: mathwft@math.canterbury.ac.nz Date: 26 June 1998 Subject: Are there "ANTI-computer" positions at BG? Forum: rec.games.backgammon Google: 6mvivi\$m08\$4@cantuc.canterbury.ac.nz

```In chess, there are several well-known "anti-computer" positions;
cunningly constructed to take advantage of known (and currently
inoperable) blind spots that the sandbrains have.

bots do badly on;  e.g. trying to get a 2nd checker back behind the prime.

But does anyone have a definite "test position"?

That is, a position from which a bot will (with high probability) lose;
but from which a human will (ditto) win?

That would be waaay koool.       Can anyone help?

Bill Taylor            W.Taylor@math.canterbury.ac.nz
```

 Gary Wong  writes: ```I would claim that there are essentially no positions that the best bots play terribly badly -- there are certainly positions where the best humans outplay the best bots by a small amount (typically relatively rare and highly tactical positions, like recirculating a chequer to pick up a second blot like you mention) but I wouldn't go as far as to say the bot would lose with a high probability where the human would win. As Chuck Bower mentions in another article, cube errors by the bots tend to be much worse than chequer play errors. One contrived position which I believe Jellyfish misplays springs to mind (bear in mind that it is tremendously unlikely to arise in play and so I don't know if you'd consider it a "test position") -- this is position #127 in Robertie's _Advanced Backgammon_ which he uses to illustrate the Kauder Paradox: +13-14-15-16-17-18-------19-20-21-22-23-24-+ | | | | | | | | | | | | | | | | | | | | O on roll, 1 cube, v| |BAR| | money game. | | | 6 6 | Jacoby rule applies. | | | O O | | X X X | | X X O O | | X X X | | X X O O O | | O X X X | | X X O O O | +12-11-10--9--8--7--------6--5--4--3--2--1-+ It should be fairly clear that O wins a certain gammon if he rolls a 6 to clear X's 5-prime, but is cubed out otherwise. However, Jellyfish does not seem to evaluate X's prime so far from its own board sufficiently valuable (I'm not sure whether it can walk it home from there) and drops the cube here (in reality X has a beaver). This is probably about as bad as you can get Jellyfish to play (X wins 0.083 cubeless equity by playing on as a human would, or loses 1.000 by dropping like Jellyfish). Note that I don't have a Windows machine myself and can't test this on a recent version of Jellyfish -- apologies for bashing it if it has learnt to play the position in the meantime :-) Cheers, Gary. -- Gary Wong, Department of Computer Science, University of Arizona gary@cs.arizona.edu http://www.cs.arizona.edu/~gary/ ```

 David Montgomery  writes: ```I would say that there are anti-computer positions. However, defining them as positions that humans would win with high probability and that the bot would lose with high probability seems a bit much. Also, it doesn't take into account gammons and backgammons. I think it would be better to define them as positions where bots give up some amount of equity relative to good human play. As Gary pointed out, the bots can be far off on their cube actions in some positions, so you can get big errors that way. But it is also true that even just looking at cubeless checker play, there are many positions where the bots play terribly. Gary showed this position: > +13-14-15-16-17-18-------19-20-21-22-23-24-+ > | | | | > | | | | > | | | | > | | | | > | | | | > v| |BAR| O O | X on roll, cubeless > | | | O O | equity? > | | | O O | > | X X X | | X X O O | > | X X X | | X X O O O | > | O X X X | | X X O O O | > +12-11-10--9--8--7--------6--5--4--3--2--1-+ O's home board I've put X on roll and made the position cubeless. With good human play, X is probably better than +.6 cubeless; with SW at the helm, X is *negative* about the same amount. SW still wins the position nearly half the time, so I wouldn't say that it meets Bill Taylor's definition of an anti-computer positions, but giving up ~1.25ppg cubeless, it meets mine. BTW, this position isn't that contrived. If you play for a backgame from the start against Snowie, or JF versions 1 and 2, it isn't that uncommon to end up with positions similar to this. With the cube in play, you can jack it up to 512 and come out ahead even though you lose gammons on most games. > +13-14-15-16-17-18-------19-20-21-22-23-24-+ > | O O | | O X X | > | | | | > | | | | > | | | | > | | | | > v| |BAR| | X on roll, cubeless > | | | | equity? > | | | | > | | | | > | | | O O O O O | > | O O | | O O O O O X | > +12-11-10--9--8--7--------6--5--4--3--2--1-+ O's home board This is a position from the 4/97 Chicago Point. With good human play, X is worth about .25 cubeless. Rolling the position out with JF, you'll get between .4 and .5, depending on the version and level you use. (It looks like version 2 plays it better.) Whether a .2 cubeless error qualifies as an anti-computer position I'll leave up to the reader. A minor point: the bots get these kinds of positions wrong because they are strategic, not tactical. The bots don't have the right long term plan. In tactical positions, the bots are extraordinarily strong. David Montgomery monty@cs.umd.edu monty on FIBS ```

### Programming

Adjusting to a weaker opponent  (Brian Sheppard, July 1997)
Anticomputer positions  (Bill Taylor+, June 1998)
BKG 9.8 vs. Villa  (Raccoon+, Aug 2006)
BKG 9.8 vs. Villa  (Andreas Schneider, June 1992)
BKG beats world champion  (Marty Storer, Sept 1991)
Backgames  (David Montgomery+, June 1998)
Blockading feature  (Sam Pottle+, Feb 1999)
Board encoding for neural network  (Brian Sheppard, Feb 1997)
Bot weaknesses  (Douglas Zare, Mar 2003)
Building and training a neural-net player  (Brian Sheppard, Aug 1998)
How to count plies?  (Chuck Bower+, Jan 2004)
How to count plies?  (tanglebear+, Mar 2003)
Ideas for improving computer play  (David Montgomery, Feb 1994)
Ideas on computer players  (Brian Sheppard, Feb 1997)
Introduction  (Gareth McCaughan, Oct 1994)
Measuring Difficulty  (John Robson+, Feb 2005)
Methods of encoding positions  (Gary Wong, Jan 2001)
N-ply algorithm  (eXtreme Gammon, Jan 2011)
Neural net questions  (Brian Sheppard, Mar 1999)
Pruning the list of moves  (David Montgomery+, Feb 1994)
Search in Trees with Chance Nodes  (Thomas Hauk, Feb 2004)
Source code  (Gary Wong, Dec 1999)
TD-Gammon vs. Robertie  (David Escoffery, June 1992)
Training for different gammon values  (Gerry Tesauro, Feb 1996)
Training neural nets  (Walter Trice, Nov 2000)
Variance reduction in races  (David Montgomery+, Dec 1998)
Variance reduction of rollouts  (Michael J. Zehr+, Aug 1998)
Variance reduction of rollouts  (Jim Williams, June 1997)
What is a "neural net"?  (Gary Wong, Oct 1998)
Writing a backgammon program  (Gary Wong, Jan 1999)