Signals, Indices and the Empty Fort: Robert Jervis's 'The Logic of Images in International Relations' — Epoche C2
The open gates at Xicheng, and what needs explaining about them In chapter ninety-five of Luo Guanzhong's fourteenth-century novel Romance of the Three Kingdoms — read here in Moss Roberts's complete translation of 1991 — Zhuge Liang is holding the small town of Xicheng with a garrison of a few thousand, the campaign having collapsed with the loss of Jieting, when he learns that Sima Yi is approaching with an army the novel numbers in the hundreds of thousands. He orders the banners struck, the four gates thrown open and a handful of soldiers dressed as townspeople to sweep the roadway; then he sits on the gate tower with two attendants and plays the qin. Sima Yi arrives, looks, concludes that a commander so conspicuously undefended must be inviting him into an ambush, and withdraws. The episode is the standing example — in the Chinese strategic tradition and, increasingly, in Anglophone management writing — of deception as an instrument of command. What needs explaining is not that the trick worked but how it could have worked, given that Sima Yi knew perfectly well that Zhuge Liang wanted him to withdraw. Any account that stops at "he projected confidence" has not touched the problem. The projection was free; both men knew it was free; and a free display emitted by an adversary whose interests are exactly opposed to one's own ought, on the face of it, to carry no information whatever. The book under review here is the one that first stated that problem properly, and it is not about the Three Kingdoms. Robert Jervis's The Logic of Images in International Relations (1970) asks how states go about controlling the beliefs other states hold about their capabilities and intentions, and what limits that enterprise runs into. Its vocabulary is still the vocabulary in which the question is discussed. The Empty Fort turns out to be an unusually clean test of it — clean enough to show both where the framework holds and where it has since been superseded. The apparatus: signals, indices, and the cost of counterfeiting Jervis's organising distinction is between two kinds of thing an actor can emit. A signal , in his usage, is a statement or an action whose meaning is fixed by convention — by tacit or explicit understandings shared among the actors. A diplomatic note, the recall of an ambassador, a military exercise staged where it will be noticed: these mean what the parties have come to agree that they mean, and their meaning would evaporate if the convention lapsed. An index is a statement or action believed to be tied inextricably to the actor's real capabilities or intentions, and which therefore carries evidence of its own for the image it projects. The line is not the line between words and deeds. It falls between emissions that can be produced whatever the underlying reality and emissions that cannot. Everything in the book follows from one consequence of that difference. Because a signal can be issued by an actor of any type, its informational content is not fixed by the signal itself but by the receiver's estimate of who would trouble to send it. A receiver who believes that a weak state has every reason to announce that it is strong learns nothing from the announcement. Deception, on this account, is not the manufacture of a false statement — that part is trivial — but the manufacture of something the receiver will file under indices when it belongs under signals. The deceiver's difficulty is that the receiver is running the same analysis in reverse and knows what the deceiver's incentives are. This is why Jervis's book is not a manual: the more valuable a given deception would be, the more obviously it is worth attempting, and the more heavily the receiver discounts it. Here the book's age shows, in a way worth being precise about, because what it lacks was supplied within three years and from another discipline. Jervis has no criterion for when an act belongs in the index column. He offers cases, and the cases are well chosen, but the boundary is drawn by judgement. Michael Spence's 1973 treatment of education as a signal in the labour market supplied the missing criterion: an emission separates types in equilibrium when its marginal cost is lower for the type that wants to be identified than for the type that would like to imitate him — the condition later named single-crossing. Education functions as a signal of productivity, on Spence's argument, not because it makes anyone productive but because it is cheaper in effort for the more able; the less able find the imitation not worth its price and decline to attempt it, and the credential is believed because only one type buys it. Transposed back to Jervis, the lesson is that "index" is not a property of an action at all. It is a property of the schedule of costs that the action imposes across the types who might perform it. Any index degrades into a mere signal the moment counterfeiting becomes cheap, and any signal hardens into an index the moment it becomes expensive in a way that bites unevenly. The opposite limiting case is worth having in hand as well, because it fixes what the Empty Fort cannot have been. Vincent Crawford and Joel Sobel's 1982 analysis of strategic information transmission takes messages that are costless and irrelevant to payoffs in themselves — cheap talk — and asks how much information can survive in equilibrium. Their answer is that credible communication is partitional: the sender can convey only which of a finite number of intervals the true state lies in; the number of intervals falls as the divergence between the sender's and the receiver's preferred actions grows; and past a threshold of divergence the only equilibrium remaining is the babbling one, in which the message carries nothing and the receiver rationally ignores it. Zhuge Liang and Sima Yi were not in partial conflict. They were commanders of opposed armies, with preferences over the receiver's action that were exactly reversed. That is past the