The Moment Serious Analysis Became Accessible
In the late 1990s, serious computer analysis came at a price. Programs such as Snowie and Jellyfish gave players access to extraordinarily strong backgammon analysis, but they were commercial products that many casual players simply could not justify buying.
Then a programmer started building a backgammon bot for fun.
Within a few years, one of the strongest analysis engines in the game was available to everyone, free of charge and with its source code open for anyone to inspect.
That engine became GNU Backgammon, usually shortened to gnubg.
It is still important to GamesGrid today. The backgammon engine inside our own platform is built on GNU Backgammon, which means the story of how it came about is also part of the story of GamesGrid.
It started with a bot called Abbott
Gary Wong returned to backgammon after seeing a colleague play on FIBS, the First Internet Backgammon Server.
His first experiment was Abbott, a simple FIBS bot based on Gerald Tesauro's pubeval evaluation function. It was completed in early 1998.
By October of that year, Wong had already built a prototype neural network of his own, named Costello, influenced by the earlier work of Hans Berliner, Gerald Tesauro and Eric Groleau.
In 1999, Wong offered the project to the GNU Project. GNU Backgammon became free software, and development quickly expanded beyond its original author.
The motivation was straightforward: expert-level backgammon software existed, but there was no comparable program available as free software.
GNU Backgammon set out to change that.
A project built by people around the world
GNU Backgammon was never built as a commercial product.
The first major contributors to join Wong included Jørn Thyssen and Joseph Heled in late 1999. Thyssen worked on cubeful evaluation, allowing the engine to evaluate positions with the doubling cube properly included. Heled focused on neural-network training and the evaluation system behind the player.
By 2000, Øystein Johansen had brought GNU Backgammon to Windows.
From there the project expanded into an international volunteer effort, with developers, translators, testers and strong backgammon players contributing to different parts of the software.
That collaborative model became one of GNU Backgammon's defining strengths. The engine was not developed behind closed doors. Players could inspect it, challenge it, test positions and contribute improvements.
Backgammon engines did not start with GNU
GNU Backgammon belongs to a much longer history of computers learning to play the game.
| Era | Approach | Example |
|---|
| 1970s | Hand-written evaluation rules | BKG |
| Early 1990s | Neural networks trained through self-play | TD-Gammon |
| Mid to late 1990s | Commercial neural-network programs | Jellyfish, Snowie |
| From 1999 | Open-source neural-network analysis | GNU Backgammon |
| Modern era | Deeper networks and faster analysis | Modern commercial and open-source engines |
The important breakthrough came before GNU Backgammon.
BKG showed that computers could compete
In 1979, Hans Berliner's BKG 9.8 defeated reigning World Backgammon Champion Luigi Villa in a seven-point match in Monte Carlo.
The result was helped considerably by favourable dice, and Berliner himself did not claim the program was the stronger player over a long match series.
But it was an important moment in computer backgammon. Machines could no longer be dismissed as incapable of competing with elite human players.
TD-Gammon changed the way the game was understood
The larger breakthrough came from Gerald Tesauro at IBM Research.
His TD-Gammon system used temporal-difference reinforcement learning and improved through enormous numbers of games against itself.
It did more than become strong.
Its analysis challenged accepted human backgammon theory. Computer rollouts began overturning long-held assumptions about opening moves and positional play, helping reshape the way serious players understood the game.
The neural-network approach behind TD-Gammon influenced the generation of engines that followed, including Jellyfish, Snowie and ultimately GNU Backgammon.
Commercial engines proved what was possible
Jellyfish, released in 1994, and Snowie, released later in the decade, brought neural-network backgammon analysis to ordinary players in polished commercial software.
For serious competitors, they changed the way matches could be studied.
Instead of arguing endlessly about whether a move was right, players could increasingly test the position.
The problem was access.
Strong analysis existed, but it still belonged to software you had to buy.
GNU Backgammon entered that world with a very different proposition:
make serious backgammon analysis available to everybody.
How a free engine became so strong
GNU Backgammon inherited the neural-network ideas that had already transformed computer backgammon, but continued developing them through specialised networks, databases, rollouts and years of public testing.
Different kinds of positions require different kinds of evaluation. Contact positions, crashed positions and pure races do not behave the same way, so GNU Backgammon uses specialised networks rather than trying to solve every position with one general model.
In bear-offs, where sufficiently small endgame positions can be solved in advance, the engine can rely on databases containing known results rather than estimating from scratch.
Rollouts add another layer. A position can be played forward many times to compare alternatives statistically and determine which choice performs better over a large sample.
And because the project is open source, those decisions and results can be examined by the community rather than accepted purely on trust.
The GamesGrid connection goes back more than twenty years
GNU Backgammon and GamesGrid are not newly connected.
One of the original GamesGrid bots, GGRaccoon, used the GNU Backgammon engine.
GNU Backgammon's own documentation recorded GGRaccoon running an earlier GNU engine on GamesGrid, deliberately configured for fast response, and reaching a server rating as high as 2165 (an October 2002 article recorded 2086).
That makes the relationship particularly relevant to the new GamesGrid.
The original GamesGrid used GNU technology to create readily available computer opposition for its members.
More than twenty years later, we are using GNU Backgammon again, but taking the idea much further. Instead of a small family of bots, the new club contains more than 500 virtual players with different playing strengths, personalities and styles.
The engine evaluates the backgammon.
GamesGrid builds the players around it.
Our Bots & AI page explains how.
GamesGrid even became part of GNU Backgammon
The connection worked in both directions.
GNU Backgammon could import GamesGrid's own .sgg save-game format, allowing GamesGrid matches to be opened and analysed outside the original client.
The importer remained part of GNU Backgammon long after the original GamesGrid closed, later appearing under the GridGammon name after that service inherited parts of the old GamesGrid environment.
For an online server's match format to remain supported in an open-source analysis engine decades later is a small but meaningful piece of GamesGrid history.
Still relevant today
GNU Backgammon is not simply an historical engine.
Its open-source code, analysis tools and playing engine continue to be used and developed. Its importance extends beyond raw playing strength because anyone can download it, inspect it and use it without buying a licence.
That principle is particularly important to us.
GamesGrid uses GNU Backgammon because it gives us a strong and transparent foundation for position evaluation, opponent play and post-match analysis.
Within GamesGrid, it performs two distinct roles:
During play, it evaluates positions for our virtual club members. Their individual playing models determine what they do with those evaluations.
After play, it powers the analysis used to review your decisions.
The opponent can be deliberately weaker. The analyst is not.
That separation is explained in detail on our Bots & AI and Backgammon Analysis pages.
What GNU Backgammon changed
Perhaps GNU Backgammon's biggest contribution was not simply that it played well.
It helped make serious analysis ordinary.
Before strong free engines became available, analysing a match at high quality required access to specialised commercial software. GNU Backgammon gave anybody with a computer the ability to examine positions, measure errors, run rollouts and question long-established backgammon ideas.
That helped create the modern culture of measured play.
Players increasingly stopped asking only:
Did I win?
They could also ask:
Did I play well?
That distinction lies behind Performance Rating, equity loss, blunder analysis and much of the way modern competitive backgammon is studied today.
Downloading GNU Backgammon
GNU Backgammon is free software and can be downloaded from the official GNU project website.
Versions are available for the major desktop operating systems, although installation details can vary between releases and platforms.
For the latest packages, installation instructions and source code, we recommend using the official GNU Backgammon website rather than third-party download sites.
You do not need to install GNU Backgammon to play GamesGrid. GamesGrid already incorporates the engine into its browser-based platform.
Installing it separately is useful if you want your own desktop analysis environment or want to examine exported matches independently.
Using GNU Backgammon with your GamesGrid matches
Every GamesGrid match can be exported in the standard .mat format.
That gives you the option to take a match outside GamesGrid and analyse it independently.
The process is straightforward:
-
Install GNU Backgammon from its official source.
-
Export your GamesGrid match as a .mat file.
-
Import the match into GNU Backgammon.
-
Analyse the match using the settings you prefer.
You can then inspect checker-play errors, cube decisions, equity loss and overall performance independently of GamesGrid's own review.
That is intentional.
We want GamesGrid analysis to be useful, but we also want the match itself to remain yours.
Our Backgammon Analysis page explains the GamesGrid review. Our AI & Bots Performance page explains PR, ratings and the strength levels used by our virtual players.
See Also
Sources
- Albert Silver, "GNU Backgammon", GammOnLine, October 2002 — source of the GamesGrid server rating. Archived at bkgm.com.
- "Interview with Gary Wong", Der Backgammonfreund, undated — source of the Abbott and Costello origin, the 1999 GNU adoption and the arrival of Jørn Thyssen and Joseph Heled.
- GNU Backgammon project credits (credits.sh in the source tree) — authors, code contributors, translators and the general contributors list.
- Albert Silver and Christian Anthon, All About GNU Backgammon — the project's official tutorial, published on gnu.org. Source of the 2165 figure and the GGRaccoon passage.
- GNU Backgammon Manual V0.16, import-formats section — "(.sgg) GamesGrid Save Game".
- GNU Backgammon Manual V1.00.0 — "GridGammon Save Game (.sgg)". On GridGammon's relationship to GamesGrid: Daniel Murphy, BGonline forum, April 2012.
