backgammon GG Robots

Other Games

Community Standards
Current Rankings
World Champions
Professional Lessons

Daily Free Tournaments
Monthly Free Tournaments
Monthly Money Tournaments

Money Play

Server Status

GG Robots

 What are the GG robots? 
GG robots are neural-net backgammon players, based on GnuBG. They are able to play at a very high level.

 What are the different GG robots? 
You will see the GG robots in several incarnations, which are set to play at different levels. GGraccoon plays with no lookahead. GGotter plays with induced errors, so that it does not always make the play that was evaluated as best. This technique causes it to play at an Intermediate level (approximately 1700). GGweasel plays at a Beginner level (approximately 1500), by introducing more frequent errors. GGchipmunk plays at a Novice level (approximately 1300).

At the various levels, they have lowest, average, and highest GamesGrid ratings of

level    target   lowest   average   highest  
GGforever none 1850 1920 2114 Life Members only
GGraccoon none 1850 1920 2114
GGotter 1700 1543 1701 1827
GGweasel 1500 1410 1516 1652
GGchipmunk 1300 1171 1275 1487

 How were the GG robots developed? 
The GG robots were developed based on the GNU Backgammon open source. In addition, the robots are wired to play multiple-simultaneous matches, in order to make them available to more players.

 Who may play the GG robots? 
GamesGrid members may invite robot players. GGotter accepts invitations only from players rated below 1800. GGweasel accepts invitations only from players rated below 1600. GGchipmunk accepts invitations only from players rated below 1400. The robots play either regular backgammon or Nackgammon, with a maximum match length of 9 points.

 What do 0-ply, 1-ply, and 2-ply mean? 
These terms refer to the level of evaluation used to determine the best play. You can think of a "ply" as a turn for one player. The basic level of play is called 0-ply, meaning that the robot does not look ahead. When playing at this level, the robot evaluates all legal moves, then selects the play that maximizes its match equity.

When playing with 1-ply lookahead, the robot goes further in investigating the situation before selecting its play. For the best candidates moves, it evaluates not only the match equities that result from its current play but also the equities that will result from your next play. It does this by examining the moves that it would make for all possible rolls.

When playing with 2-ply lookahead, the robot goes even further and evaluates the equities from its next play as well. These extended evaluations are the reason that the 2-ply robots play better than the 1-ply or 0-ply.

In many positions, there may be hundreds of 2-ply continuations to evaluate, which can take a lot of computing power (and time). The robot can evaluate faster by looking at an intelligently-selected subset of the possible plays. By concentrating on a fraction of the possible lines, The robot can usually select the best move in significantly less time than it would take to evaluate all possible continuations.

 Why do the robots' ratings fluctuate so much? 
Even though the robots play completely consistently from match to match and day to day, you will see wide variations in their ratings over time. Although unfinished matches and the particular opponents played also have a role, these fluctuations are due almost entirely to the luck inherent in backgammon. Like all of us, the robots experience good and bad streaks, and when they do, their ratings change accordingly.

Because the robots play so many matches (between 2000 and 4000 matches per day), their ratings fluctuate more than those of human players. On the other hand, the robots never get tired or steamed or overconfident. Over an equal number of matches playing against the same player mix, their ratings tend to vary less that would those of a human player against the same mix of opponents.

Because of the luck of the dice, it is fairly common for a robot to be rated 100 points higher or lower than its average, and it sometimes will be rated more than 200 points from its average. At times a 0-ply robot may be rated higher than a 2-ply robot, and two that play at exactly the same level may have significantly different ratings. This is normal in backgammon, and suggests that humans players should be less concerned about such fluctuations in their own ratings (but human nature makes that difficult).

 Do the robots get better dice? 
No. Robot players receive dice in the same way as every other player. The robot programs run on different machines than the game servers. They are connected to the server using the same client program that human players use, modified to communicate with the robot evaluation engine and to send the selected plays automatically. As far as the game server is concerned, the robots are the same as every other player on GamesGrid.

Like the best human players, they arrange the checkers so that more rolls are good for them, and fewer rolls good for their opponents. Because the higher-level robots play better than the great majority of players, they appear luckier to the uncritical eye.

 A GG robot is not accepting invitations. What's wrong? 
The robots are playing multiple-simultaneous matches. When they are available for an invitation, their green available indicator will light up. When they are playing as many matches as they can simultaneously, the available indicator will be off.

 Can I get robots to play slower (or faster)? 
Yes. A robot begins each match at its "medium" speed. Although you cannot affect its calculation speed, a robot's opponent may ask it to change the speed at which it makes its moves, by typing in the chat entry box
/tell robotname slower
/tell robotname faster
where robotname is GGraccoon, or GGotter, or one of the other robot names. The robot will confirm that it is adjusting its playing speed.

At its fastest speed, a robot moves the checkers as soon as the best play is selected. At its slowest speed, it will pause approximately two seconds before it makes each move. This can be useful if you want to visually follow the action in bearoff situations, since the moves can be played very fast indeed.

Only a robot's opponent may request speed adjustments.

Copyright © 2005-2006 GGweasle Ltd. All rights reserved.
The GamesGrid name and GGotter logo are licensed from, and property of, GGraccoon, LLC.

last modified: 2005-04-27