The Empty Fort Problem: Why Reading Another Mind Has No Floor

Romance of the Three Kingdoms states plainly what Sima Yi was thinking when he found the gates of Zhuge Liang’s town standing open, the streets swept, and Zhuge Liang himself seated on the wall playing the guqin. Sima Yi’s soldiers see an empty town, theirs for the taking. Sima Yi sees a trap. Zhuge Liang, he reasons, is famously cautious and would never invite an army in so openly unless something worse waited inside. He orders a retreat.

The story is far older than the fourteenth-century novel, and far shakier as history. Its earliest version is an anecdote told by one Guo Chong and preserved in Pei Songzhi’s fifth-century commentary on the Records of the Three Kingdoms, and Pei preserved it only to reject it. At the time, he pointed out, Sima Yi was stationed far away at Wan, commanding in Jing province, not facing Zhuge Liang at all; and a general who suspected an ambush could simply have held his ground and waited. The episode almost certainly never happened. That is, if anything, the more interesting fact. What a thousand years of retelling kept alive was a story about how far one mind can follow another mind’s expectations before following them stops making sense.

In the story, Zhuge Liang is counting on a belief Sima Yi already holds. Sima Yi’s whole model of his rival rests on the premise that Zhuge Liang never takes needless risks. An open gate ought to be the one piece of evidence that model cannot absorb. Instead it confirms the model, because a reliably cautious man acting recklessly looks, to a mind expecting caution, like bait. Sima Yi does not take the gate as a gate. He takes it as a message from someone he already has a theory of. Nothing in the town causes the retreat. A belief about a belief does: Sima Yi’s model of what Zhuge Liang wants him to conclude from Sima Yi’s model of Zhuge Liang.

Treating this as proof of Sima Yi’s insight runs into a difficulty beyond the story’s doubtful history. Reasoning about another mind’s reasoning has no natural stopping point. If Sima Yi can think “he wants me to believe the town is empty, so it must be an ambush,” nothing prevents Zhuge Liang from anticipating that thought and leaving the gate open as a true invitation, counting on Sima Yi’s cleverness to talk him out of the obvious move. Nothing prevents Sima Yi from anticipating that anticipation in turn. Logic never resolves the regress. It ends wherever a mind runs out of nerve, time or patience for the next level, and the story, and the general who survives it, later treat that stopping point as though it had been the correct depth all along.

The difficulty is not confined to strategists of the Three Kingdoms. Poker players meet it constantly. A bluff works only if the opponent stops calculating at the right level, and an opponent who reasons one level further can take a strong hand for a bluff and call. In international politics the same structure appears as the security dilemma, a term usually credited to John Herz in 1950 and developed by Robert Jervis. Each state, unsure of the other’s intentions, reasons about how its own precautions will be read, and how that reading will be read, with nothing underneath but judgment, fatigue or luck deciding where the reasoning stops. What looks like superior insight into another mind is often a correct guess about how many levels down the other side will stop digging.

It reaches into ordinary life as well. Rereading a terse message and wondering whether the silence before it means anger or merely a full inbox is the same descent, with the same missing floor: a guess about what someone wants one to conclude, built on a guess about what they feel, halted at whatever level felt sufficient to act on. Most people never learn, any more than Sima Yi would have, whether they read the situation correctly or merely stopped at a lucky depth.

Seen this way, the legend preserves something else. It is not the record of a general who read minds so well that he turned back an army with psychology alone. It is a less comfortable fact presented as strategy: that modeling another mind has no floor, and that every strategist, every reader, every mind trying to outthink another is choosing where to stop and calling the choice a conclusion. Centuries of retelling turned a guess that happened to land into something that reads as insight, as any guess comes to look once enough time separates it from the moment when it could still have gone the other way.

Editor’s note: you can try this regress yourself. Guess Two-Thirds of the Average asks you to pick the number that will be two-thirds of what everyone else picks, then shows how many steps of “they will think that I think” your guess took, and how far other readers went.