The history of backgammon

Five thousand years of the same idea

Backgammon belongs to the oldest family of board games we know about. Race games with dice — move your pieces round a track, hit your opponent's, get home first — appear in Mesopotamia and Egypt around 3000 BCE, and the line from those boards to the one in your cupboard is continuous.

Two ancestors are firmly documented. The Royal Game of Ur (c. 2600 BCE) is a race with dice and capture on a twenty-square track. Egyptian senet is older still. Neither is backgammon, but both are the same idea, and the family resemblance is obvious once you have played them.

The direct ancestor is Roman. Ludus duodecim scriptorum — "the game of twelve lines" — used three dice on a three-row board; by the first centuries CE it had simplified into tabula, played on two rows of twelve points with two dice. Tabula is recognisably backgammon. There is even a surviving account of a game played by the emperor Zeno in 480 CE, complete with the roll that ruined his position.

Nard, tables and the medieval game

In Persia the game was nard, and a Persian text of the sixth or seventh century presents it as a deliberate invention with symbolic meaning: the board as the year, the twenty-four points as the hours, the thirty checkers as the days of the month. That is literary framing rather than history, but it shows the game was prestigious enough to be worth the framing.

The Islamic world carried nard west; medieval Europe called the whole family tables, and it was for centuries the most popular gambling game on the continent — popular enough to be repeatedly banned. Church and crown issued prohibitions against tables from the twelfth century onwards, which is the surest evidence we have of how widely it was played.

Hoyle, the cube and the 1970s boom

The English name "backgammon" appears in the seventeenth century, and the rules were fixed in print by Edmond Hoyle in 1743, whose treatise settled the arguments that had kept the game fragmented.

The game then stayed essentially unchanged until the 1920s, when somebody in New York — the inventor has never been identified — introduced the doubling cube. It transformed the game. Suddenly the question was not only "what is the best move?" but "is this position worth twice as much to me as to you?", and that second question turned out to be deeper than the first.

The cube also made backgammon fashionable. The 1960s and 1970s produced a genuine boom, driven by club play, jet-set tournaments and a set of books that professionalised the theory — above all Paul Magriel's Backgammon (1976), still the standard introduction to the modern game.

TD-Gammon and what the machines proved

Backgammon has a special place in the history of machine learning. In 1992 Gerald Tesauro's TD-Gammon at IBM learned the game almost entirely by playing itself, using temporal-difference learning with a neural network. It reached a standard close to the best humans — twenty-five years before the same approach conquered Go — and it is one of the first great successes of reinforcement learning.

More interestingly, TD-Gammon changed human opening theory. It preferred moves that expert consensus had rejected — notably splitting the back checkers with certain rolls, and slotting the five point — and analysis confirmed the machine was right. Opening tables were rewritten around a program's preferences, which had never happened in any game before.

Today's bots analyse by rollout: play the position out thousands of times with strong play on both sides and compare the equity of each candidate move. Every serious player studies this way, and the resulting error-rate measurement — millipoints lost per move — has given backgammon a more precise measure of skill than any other classical game has.

Is backgammon luck or skill?

Both, in a way that is precisely measurable. A single game has a large luck component: a better player might win 60% of individual games against a weaker one. Over a long match or a session, luck washes out and the stronger player wins consistently — which is exactly why matches are played to several points rather than as single games.

The doubling cube is what shifts the balance. Dice decide what you are given; the cube decides what you do with it, and cube errors cost far more than checker errors. Analysis of top players' matches consistently shows cube decisions as the main difference between the very good and the best.

Common questions