The Internet and Backgammon: How Bots Changed Theory

For centuries, backgammon knowledge rested on experience and argument. Strong players had opinions about the best opening moves and the right time to double, and those opinions were passed down in books and clubs. Over the last few decades, computer programs and the internet changed that. Today, many questions that used to be matters of taste can be checked with numbers.

Neural networks learn the game

Early programs struggled because backgammon positions are hard to score with simple rules. The breakthrough came with programs based on neural networks. Instead of following hand-written advice, such a program plays a huge number of games against itself and slowly adjusts its own judgment of which positions tend to win. The result is an evaluation of a position that is often more accurate than any human's. The story of how this developed is told in our page on computer backgammon.

Rollouts: settling arguments by playing on

The most powerful tool these programs offer is the rollout. To compare two candidate moves, the program plays each resulting position to the end many times, with different dice each time, and records the results. Because the dice are varied systematically, the luck tends to cancel out, and what remains is a reliable estimate of how good each position is.

Rollouts settled many old debates. Some opening moves that had been popular for generations turned out to be weaker than alternatives. Some plays that experts once considered reckless proved correct. Cube decisions in particular were measured far more precisely than before, which changed when strong players double, take and pass.

The tools players use today

• GNU Backgammon is free, open-source software. It can play, analyze finished matches, run rollouts and estimate the cost of each error.
• eXtreme Gammon is a widely used analysis program among serious players, valued for its strong evaluations and detailed reports.

Both are used as study tools: you play a match, save it, then let the program mark the moves and cube actions where you lost the most equity. A summary error rate gives a simple measure of how well you played overall. Our page on luck and skill analysis explains how these reports separate good play from good dice.

How the internet spread the new ideas

Software alone would have changed only the habits of a few experts. The internet carried the results to everyone. Players posted rollout results, argued about them in public, and built shared reference material on openings, replies and common cube positions. A new idea discovered by analysis could reach players around the world within weeks instead of years.

This also changed what it means to be a strong player. Top competitors now study with software every day, and the general level of play has risen. Positions that once separated experts from club players are now common knowledge.

Using bots without losing your own judgment

• Ask why. A program tells you which move is better but not always the reason. Try to put the idea into words.
• Focus on big errors. Fix the mistakes that cost the most before polishing small ones.
• Look for patterns. If the same type of error keeps appearing, study that type of position directly.
• Keep playing people. Analysis improves your understanding, but practice against real opponents builds the concentration you need in a match.

For a full study routine built around these tools, see becoming a better player.