Counting Rods in the Temple

Sun Tzu ends the first chapter of The Art of War with a prediction. In Lionel Giles’s 1910 translation: “The general who wins a battle makes many calculations in his temple ere the battle is fought. The general who loses a battle makes but few calculations beforehand.” Sun Tzu adds that, by paying attention to this, he “can foresee who is likely to win or lose.”

This practice is called 庙算, temple calculation. Before an army went to war, its leaders made their calculations in the ancestral temple.

One modern way to explain this is the idea of a mental model. In 1943, the psychologist Kenneth Craik wrote a book called The Nature of Explanation. He said that if a creature “carries a ‘small-scale model’ of external reality and of its own possible actions within its head, it is able to try out various alternatives, conclude which is the best of them, react to future situations before they arise.” On this reading, the temple is where a general fights the war in his head before he fights it on the ground.

Craik’s idea describes planning well. But the text itself points to something simpler. The character 算 means “to count.” It is closely related to 筭, the name for the bamboo rods people used for arithmetic in early China. And the passage is about amounts. The side that wins in the temple is the one that 得算多, “gets many 算.”

A commentator from the Tang dynasty, Li Quan, read this as keeping score. He borrowed a counting rule from an old book of divination called the Taiyi Dunjia. Sixty or more counted as “many calculations.” Fewer than sixty counted as “few.” Not every commentator agrees. Du Mu, for example, only says that the calculating was done in the temple hall.

The chapter also gives us the things to count. Just before this passage, Sun Tzu lists seven questions to ask about both sides. Which ruler has the Moral Law? Which general is more able? Which side has the advantages of Heaven and Earth? Whose discipline is better? Whose army is stronger? Whose soldiers are better trained? Whose rewards and punishments are more consistent?

That gives seven questions, each asked about both sides. Sun Tzu does not say how to combine the answers. But if you add them up, as Li Quan’s reading suggests, you get something familiar. In the machine learning classes I teach, it has a name: a unit-weighted linear model. Every factor counts the same, and you add them together. My students usually think of it as the simple model you try first, before you build something better.

Research says it is harder to beat than it looks. In 1979, the psychologist Robyn Dawes published a paper called “The Robust Beauty of Improper Linear Models in Decision Making” in American Psychologist. He looked at prediction problems, such as academic performance and psychiatric diagnosis. In his examples, even linear models with rough or random weights made better predictions than the human experts. The models were not clever. They simply asked the same questions in the same way every time. A tired expert, or one impressed by a single vivid detail, does not.

Read this way, “many calculations lead to victory” is about discipline more than about a richer simulation in the mind. Ask every question about both sides, and count. That is a more modest idea than Craik’s mental model, and it has strong evidence behind it.

But Sun Tzu’s own chapter gives a warning. Dawes studied predictions about a world that does not fight back. A war does fight back. The same chapter that praises counting also says “all warfare is based on deception.” And its first methods of deception are about seeming unable to attack, farther away, or less active than you really are. That means giving false answers to the very questions the enemy will count in his own temple.

A score is only as good as its answers. An enemy who has read the same chapter is working to make your answers wrong. So both parts of the chapter are true at once: counting is better than guessing, and the enemy is busy deciding what you will count.