Book Review: The Strategy of Conflict
And God said, Let there be a firmament in the midst of the waters, and let it divide the waters from the waters.
And God made the firmament, and divided the waters which were under the firmament from the waters which were above the firmament: and it was so.
And God called the firmament Heaven.
(Genesis 1:6-8)
Legend has it that a given OKCupid user’s answer to the infamous question “In a certain light, wouldn’t nuclear war be exciting?” is the second-best predictor of their likelihood of being willing to have sex on a first date. (The first is “Do you like beer?”) So: if you’re angling to get freaky in a time sensitive way, I recommend sending your date Thomas Schelling’s 1960 classic The Strategy of Conflict ahead of time.
Just kidding! I am being a silly goose! Actually, this book is what I wish AI alignment monographs were, namely, seven chapters of delicious game theory mixed with some pondering on the Nature of Things before even making serious mention of the looming spectre that motivated its writing. It starts out with a chapter called “The Retarded Science of International Strategy” and gets better from there. The second half of Strategy of Conflict shaped the course of deterrence policy in the grandest cold war in history, but I like the first half even better. That’s because, well, you know the thrill people get out of counterintuitive situations where the incentives do a little twist, like that cobra bounty in Delhi? This book is, among other things, a compendium of them. It’s suffused with trickster fable energy.
The big idea underpinning the book is Schelling’s formulation of a novel variety of game theory. Some games are purely cooperative: the players may not value the various outcomes exactly the same, but they each rank all the outcomes in the same order. Other games are purely competitive: the better I make out, the worse you do. Schelling introduces a secret third thing: mixed-motive game theory. This encompasses games that entail elements of both cooperation and competition: one player's preferences are neither directly nor inversely related to the other player's preferences. (Schelling notes that it's not clear whether to call the players partners or opponents.)
It turns out that almost all bargaining and negotiation can be modeled as mixed-motive games, and so this is a surprisingly useful lens on a wide range of situations. At its core, The Strategy of Conflict is a pattern language for negotiations. There are basically two categories of patterns: tacit bargaining and commitments.
Category One: Tacit Bargaining
We start out with a simpler case than the mixed-motive world: a pure cooperation game where communication is restricted or impossible.
Schelling ran a bunch of little experimental games in his own lab where people would be asked to play partnered games. They’d get tasks like “name either ‘heads’ or ‘tails’,” or “write down a positive number,” or “choose a place and time to hypothetically meet your partner in New York”; both won a prize if they wrote down the same answer. The catch was they couldn’t see or talk to each other; no one even knew who their partner was, Schelling just randomly paired up answer sheets. It turned out that people could do pretty well at these games, just by recognizing a shared sense of salience: typically they’d pick heads, one, and Penn Station at noon. There was no right answer – if you and your partner somehow both decided to meet at the same Staten Island gas station at 2:18 AM, you’d win the prize nonetheless – and yet people’s answers tended to converge.
Finding the key, or rather finding a key—any key that is mutually recognized as the key becomes the key—may depend on imagination more than on logic; it may depend on analogy, precedent, accidental arrangement, symmetry, aesthetic or geometric configuration, casuistic reasoning, and who the parties are and what they know about each other.
(Schelling likes the word “casuistic.”)
The power of tacit coordination is strong. Even if the outcome is clearly skewed toward one player’s benefit, the one with the short end of the stick can be forced to capitulate. If you have to meet a stranger in Paris without communicating, and they're at the Louvre while you're somewhere out in the suburbs, there's no meeting in the middle, because what middle could you possibly coordinate on? You still have to schlep all the way over to the Eiffel Tower while they enjoy a leisurely stroll. "Beggars cannot be choosers about the source of their signal,” Schelling writes, “or about its attractiveness compared with others that they can only wish were as conspicuous."
The object level’s time to shine
One quirk of cooperative and mixed-motive games, compared to pure-competition games, is that a certain kind of abstraction becomes impossible. In a purely competitive game, if you give each of two players a list of options, it doesn't matter which order you present the list in. But in a game with an element of coordination, it does! For instance, consider this payoff matrix, where the first player picks a row, the second player picks a column, and they each get the payout at the cell where their choices intersect.
The theory of competitive games would treat the five ones as indistinguishable. But if you're playing a cooperative (or mixed-motive) game, the matrix can't escape you grabbing onto its layout as an element of mutually understood structure, and perhaps picking the first or middle row or column.
(in Soviet-motivated game theory, matrix can’t escape you)
But in the real world, can't people usually communicate? Sure. It turns out, though, that tacit bargaining is still surprisingly relevant.
Swarms
Many kinds of coordination involve large groups of people who don’t even necessarily know who the fellow members of their coalition are. Because attention is scarce, it’s difficult to propagate a signal to everyone, or to know whose signal to defer to. This is especially true if leading a movement marks you as a target; for instance, in the case of riots or protests by an opposition group.
[T]he mob’s problem is to act in unison without overt leadership, to find some common signal that makes everyone confident that, if he acts on it, he will not be acting alone. The role of “incidents” can thus be seen as a coordinating role; it is a substitute for overt leadership and communication. Without something like an incident, it may be difficult to get action at all, since immunity requires that all know when to act together. Similarly, the city that provides no “obvious” central point or dramatic site may be one in which mobs find it difficult to congregate spontaneously; there is no place so “obvious’ that it is evident to everyone that it is obvious to everyone else.
During the summer of George Floyd, when nobody had anything to do but think about protests, I remember thinking, huh, something about this whole endeavor is weird. If police brutality happens all the time, then why only do big protests in this one case? And if it doesn’t, if this kind of killing is rare enough to be egregious, then why protest it as if it’s a ubiquitous problem?
I got a much better understanding of that dynamic when I joined the social network Bluesky. My profile there is characterized by spurts of several posts in a row followed by weeks or months of nothing; if you check the dates, they form a pretty faithful timeline of “when Elon did something out of pocket.” Some instances of his tantrum-driven development actually annoyed me, like the Substack censorship. Other times, though, he just tweeted something dumb that rustled the jimmies of people who weren’t me. Still, I would think “well, everyone seems to expect everyone else to consider this new antic especially bad, and I’m still mad about the Substack thing, so I’ll take this chance to give the Bluesky momentum a little push.”
What you can’t say
There are also, of course, many cases where you’re only coordinating with a single individual. In situations like war or sales, there’s one clear counterparty. But because their incentives are only partly aligned with yours, it’s easy to communicate explicitly but hard to communicate credibly.
How can the prisoner being tortured for secrets that he really does not know persuade his captors that he does not know them? How could the Chinese, if they were really determined to take Formosa at the cost of an all-out war, persuade us that they could not be deterred in any fashion and that any threat on our part would only commit us both to all-out war?
If the truth is more favorable to your bargaining position than an alternative is, you need a credible signal to convince an opponent that it is the truth, rather than an attempt to grab more value for yourself.
Also, sometimes making your exact values publicly known is itself a bad outcome:
[M]y awareness that my neighbor does not like me may cause me small discomfort, as does his awareness of my awareness, but if we are forced to accredit the fact overtly, the pain may be acute. “Social etiquette,” remarks Erving Goffman, “warns men against asking for New Year’s Eve dates too early in the season, lest the girl find it difficult to provide a gentle excuse for refusing.”
And you don't always know whether your opponent has received your message; in fact, under some circumstances it's advantage to make yourself difficult to contact, so that your opponent is forced to take your offer or nothing.
(Well-meaning people are always saying "if only people could communicate better, they could cooperate" and then going off to try and bridge a cultural gap or build a tool for collective sensemaking. This never seems to work, and I think The Strategy of Conflict implies a pretty good case for why: often, it’s not that the parties can’t communicate, it’s that one or both has an incentive not to. And in many more cases, explicit communication isn’t the kind of communication that actually matters. I’d postulate that doing a lot of communication and doing a lot of cooperation are both caused by a third thing: people who are similar can communicate, because they share context, and also cooperate, because they share values and hence incentives.)
Credible compromises
How do you convince an opponent that you want to make a concession, but only a limited one?
There is perhaps a little more to this need for a mutually identifiable resting place. If one is about to make a concession, he needs to control his adversary’s expectations; he needs a recognizable limit to his own retreat. If one is to make a finite concession that is not to be interpreted as capitulation, he needs an obvious place to stop. A mediator’s1 suggestion may provide it; or any other element that qualitatively distinguishes the new position from surrounding positions. If one has been demanding 60 per cent and recedes to 50 per cent, he can get his heels in; if he recedes to 49 per cent, the other will assume that he has hit the skids and will keep sliding.
For example:
The physical configuration of Korea must have helped in defining the limits to war and in making geographical limits possible. The area was surrounded by water, and the principal northern political boundary was marked dramatically and unmistakably by a river. The thirty-eighth parallel seems to have been a powerful focus for a stalemate; and the main alternative, the “waist,” was a strong candidate not just because it provided a shorter defense line but because it would have been clear to both sides that an advance to the waist did not necessarily signal a determination to advance farther and that a retreat to the waist did not telegraph any intention to retreat farther.
Quality over degree
One major principle of tacit bargaining is that numerical degree doesn’t make a good basis for coordination, unless the number is especially round or salient. For instance, gas wasn't used at all in World War II, even though there was no formal agreement not to. “'Some gas' raises complicated questions of how much, where, under what circumstances,” writes Schelling, while “'no gas' is simple and unambiguous.” Zero is a good Schelling point. And the simpler the quality, the better.
(A modern example: a new effective altruist diet called "ameliatarianism" aims to ameliorate the worst harms of factory farming. An ameliatarian eats everything but eggs, poultry, and farmed seafood. I don't think it'll work because there's no bright line, in contrast to "no animal flesh" or "no animal products" or even "no land animals.” But maybe there's hope if they can emulate the success of kosher or halal diets and get a critical mass to agree on a particular interpretation of God's will as a rallying point.)
Framing
When natural boundaries don’t exist, you can try to create them ex nihilo. Schelling cites “fact-finding” reports as a way to establish some baseline suggestion that everyone acknowledges as a sort of default, with no particular relation to what the facts are or how they were decided on or even whether there’s real consensus.
Beggars can't be choosers when it comes time to begin a negotiation – but there's lots of room for artful framing beforehand.
The “obvious” outcome depends greatly on how the problem is formulated, on what analogies or precedents the definition of the bargaining issue calls to mind, on the kinds of data that may be available to bear on the question in dispute. When the committee begins to argue over how to divide the costs, it is already constrained by whether the terms of reference refer to the “dues” to be shared or the “taxes” to be paid, by whether a servicing committee is preparing national-income figures or balance-of-payments figures for their use, by whether the personnel of the committee brings certain precedents into prominence by having participated personally in earlier negotiations, by whether the inclusion of two separate issues on the same agenda will give special prominence and relevance to those particular features that they have in common. Much of the skill has already been applied when the formal negotiations begin.
If a country wants to use atomic weapons without triggering all-out war, Schelling thinks it should develop some pretense to use nukes for something else, even if that something else is totally unrelated, like training troops in allied countries on how to survive nuclear explosions using actual nuclear detonations in the training, or using nuclear dynamite for big construction projects. Alternatively, if a country wants to make it more likely that atomic weapons will be seen as cause for all-out war, it should lobby for an international agreement to suspend all nuclear testing – which, notably, can generate an effective Schelling point regardless of whether it's even ratified if the country can get everybody talking about it a lot.
A modern example: pickup artists talk about “breaking the touch barrier.” If you've been hanging out with somebody for two hours with zero physical contact, and then you touch their arm, that's an escalation of intimacy on the literal level but also an explicit Move on the symbolic level, recognized as such by both parties. Whereas if you've established from the start that you're a hand-on-arm-to-emphasize-a-joke person, or you’ve guided them by the shoulder through a crowd, or even just accidentally-on-purpose brushed their hand, then an actual maneuver toward ramping up intimacy doesn't get caught in the air as something that has to be acknowledged and responded to. This might also be the mechanism behind the concept of "normalizing" something, especially in the bad sense, like “normalizing violence.”
Schelling doesn’t talk about it, but I suspect advertising might work this way, too, insofar as it establishes a visible consensus point of “x brand represents y vibe.” You could look at it as brands jockeying for semiotic position. Mass media, where the company can unilaterally talk at the public but the public can’t talk back at the company without coordinating, calls to mind again the beggars-can’t-be-choosers-about-signals principle.
We get a little weird with it
The Strategy of Conflict falls into a book niche you don’t find much anymore: serious enough to be citing specific Nature papers, but accessible enough for a layperson. This means that there are entertaining parts, but rather than the standard pop-econ “…and that’s how supply and demand combine to bring you your daily coffee. Wow – the dismal science sure is relatable!”, they’re of the genre “please enjoy a multiparagraph footnote about space aliens, hallucinations, and God.”
Speaking of Gestalt psychology:
One wants to know how everyone else is going to act and whether everyone else will do what he knows he ought to. A test vote in a legislature or some particular simultaneous action among the group, like a mass protest, is often a means of “ratifying” the existence of the coalition and of demonstrating that everybody expects everybody else to act in concert. But even in a two-person game, as typified by the dare, the phenomenon of psychological dominance or submissiveness may prove to be psychologically identical with the resolution of a bargaining game.
Schelling proposes a wacky twist on his original set of experiments: play a bargaining game with both players hooked up to some kind of attention monitor, similar to a lie detector. Maybe it goes beep beep beep when your heart rate goes up or your brain waves start waving faster. Then show the players the possible choices, one by one, and let them look at each other's monitors as well as their own the whole time. What does that do to the bargaining outcome, he wonders? But he has not, in fact, run this experiment. Maybe you can!
Category Two: Commitments
To get your way in the world of tacit bargaining, you can manipulate the shadowy psychic web of concepts and symbols, stoke a new Current Thing, run a Super Bowl ad, loudly shout “PICK OPTION THREE” and run away. But there’s a second, more direct strategy: commitments, which allow you to manipulate incentives.
In purely competitive games, there’s no cooperative element at stake, so there’s nothing to hold hostage. In mixed-motive games, on the other hand, you can tie your cooperation to conditions, including threats, promises, and combinations thereof.
The fundamental nature of commitments is judo. There are all sorts of moves, but they’re linked by a common theme of cutting off some of your own options, making your position strictly weaker in a sense. Yet because of the way you rearrange the incentives, your position actually ends up stronger.
As with tacit bargaining, the best way to illustrate this is with a gazillion examples.
Industrial-strength haggling
When one wishes to persuade someone that he would not pay more than $16,000 for a house that is really worth $20,000 to him, what can he do to take advantage of the usually superior credibility of the truth over a false assertion? Answer: make it true. How can a buyer make it true? If he likes the house because it is near his business, he might move his business, persuading the seller that the house is really now worth only $16,000 to him. This would be unprofitable; he is no better off than if he had paid the higher price.
But suppose the buyer could make an irrevocable and enforceable bet with some third party, duly recorded and certified, according to which he would pay for the house no more than $16,000, or forfeit $5,000. The seller has lost; the buyer need simply present the truth.
Public infrastructure for commitments
If both men live in a culture where “cross my heart” is universally accepted as potent, all the buyer has to do is allege that he will pay no more than $16,000, using this invocation of penalty, and he wins—or at least he wins if the seller does not beat him to it by shouting “$19,000, cross my heart.”
Generating trust by proving that you have something to lose
Among the legal privileges of corporations, two that are mentioned in textbooks are the right to sue and the “right” to be sued. Who wants to be sued! But the right to be sued is the power to make a promise: to borrow money, to enter a contract, to do business with someone who might be damaged. If suit does arise, the “right” seems a liability in retrospect; beforehand it was a prerequisite to doing business.
Giving your opponent a way out
If one can demonstrate to an opponent that the latter is not committed, or that he has miscalculated his commitment, one may in fact undo or revise the opponent’s commitment. Or if one can confuse the opponent’s commitment, so that his constituents or principals or audience cannot exactly identify compliance with the commitment—show that “productivity” is ambiguous, or that “proportionate contributions” has several meanings—one may undo it or lower its value.
On the other hand, if it's actually to your advantage to concede, you have to give yourself a way out:
Concession not only may be construed as capitulation, it may mark a prior commitment as a fraud, and make the adversary skeptical of any new pretense at commitment. One, therefore, needs an “excuse” for accommodating his opponent, preferably a rationalized reinterpretation of the original commitment, one that is persuasive to the adversary himself.
Build a reputation, then credibly threaten to use it as collateral
If a union is simultaneously engaged, or will shortly be engaged, in many negotiations while the management has no other plants and deals with no other unions, the management cannot convincingly stake its bargaining reputation while the union can. The advantage goes to the party that can persuasively point to an array of other negotiations in which its own position would be prejudiced if it made a concession in this one.
Divide your commitments into smaller pieces, creating an iterated game
In order that one be able to pledge his reputation behind a threat, there must be continuity between the present and subsequent issues that will arise. This need for continuity suggests a means of making the original threat more effective; if it can be decomposed into a series of consecutive smaller threats, there is an opportunity to demonstrate on the first few transgressions that the threat will be carried out on the rest. Even the first few become more plausible, since there is a more obvious incentive to fulfill them as a “lesson.”
Two can play at that game, though!
A piecemeal approach may also be used by the threatened person. If he cannot obviate the threat by hastening the entire act, he may hasten some initial stage that clearly commits him to eventual completion. Or, if his act is divisible while the threatener’s retaliation comes only in the large economy size, performing it as a series of increments may deny the threatener the dramatic overt act that would trigger his response.
Flip the principal-agent problem to your advantage
If you separate the executive and legislative powers, then the executive branch can drive a harder bargain in its international dealings if it can credibly claim "sorry, I'm legally barred from making this concession, and Congress won't reconvene in time to change the law." Or you can interpose some agent with limited flexibility, for instance by sending a messenger to carry out a timeboxed negotiation while you're off vacationing on a desert isle. Or you can strategically pick an agent with different incentives from yours, like an insurance company that will sue on your behalf and won't be tempted to settle outside of court, like you might be, because it has a threat-carrying-out reputation to uphold.
Lump separate matters into the same negotiation
If there are two projects, each with a cost of three, and each with a value of two to A and a value of four to B, and each is inherently a “one-man” project in its execution, and if compensation is institutionally impossible, B will be forced to pay the entire cost of each as long as the two projects are kept separate. He cannot usefully threaten nonperformance, since A has no incentive to carry out either project by himself. But if B can link the projects together, offering to carry out one while A carries out the other, and can effectively threaten to abandon both unless A carries out one of them, A is left an option with a gain of four and a cost of three, which he takes, and B cuts his cost in half.
Force your opponent to act first and look bad
This tactic of shifting responsibility to the other player was nicely accomplished by Lieutenant Colonel (then Major) Stevenson B. Canyon, U.S.A.F., in using his aircraft to protect a Chinese Nationalist surface vessel about to be captured by Communist surface forces in his comic strip. Unwilling and unauthorized to initiate hostilities and knowing that no threat to do so would be credited, he directed his planes to jettison gasoline in a burning ring about the aggressor forces, leaving to them the last clear chance of reversing their engines to avoid the flames. He could neither drop gasoline on the enemy ships nor threaten to; so he dropped the initiative instead.
Offer up, uh, a generation of children as voluntary hostages, just kidding, haha, unless
We could probably guarantee the Russians against an American surprise attack by having the equivalent of “junior year abroad” at the kindergarten level: if every American five-year-old went to kindergarten in Russia—in American establishments constructed for the purpose, designed solely for “hostage” purposes and not for cultural interchange—and if each year’s incoming group arrived before the graduating class left, there would not seem to be the slightest chance that America would ever initiate atomic destruction in Russia.
Stick your fingers in your ears and go “la-la-la I can’t hear you”
It is important, of course, whether or not the threatener knows that his threat cannot be received; for if he thinks it can, and it cannot, he may make the threat and fail in his objective, being obliged to carry out his threat to the subsequent disadvantage of both himself and the one threatened. So the soldiers in quelling the riot should not only be strangers and not only keep moving sufficiently to avoid “acquaintance” with particular portions of the mob; they should behave with an impassivity to demonstrate that no messages are getting through. They must catch no one’s eye; they must not blush at the jeers; they must act as if they cannot tell one rioter from another, even if one has been making himself conspicuous. Figuratively, if not literally, they should wear masks; even the uniform contributes to the suppression of identification and so itself makes reciprocal communication difficult.
You can even just convince your opponent that you have a temper or revenge instinct that will cause you to act irrationally, or that you are simply dimwitted. Perhaps the easiest way to convince them is for these things to be true. "Best of all may be genuine ignorance, obstinacy, or simple disbelief."
*hits blunt*
(…but hear me out.)
Beyond the fun trickster fable stuff, there were three reasons I really vibed with this book.
Reason one is that mixed-motive game theory is delightful insofar as, because it contains signaling, it ranks with the theory of zero-sum games and the theory of evolutionary psychology in the pantheon of “frameworks that let you generate endless unfalsifiable hypotheses which are nevertheless tantalizing because you know the framework itself is sound.” If you read Seeing Like A State and started going around talking about legibility, or got into Taleb and went through a phase of callling everything antifragile, you will enjoy The Strategy of Conflict.
Reason two is that – well, uh, I don’t know if this happens to you, but sometimes I’ll be buying eggs or raking leaves and the thought will pop into my head: “Hey remember how Earth contains thousands of ICBMs that can theoretically be launched at any point on the planet at any time?” In a sense, walking around with guns to our heads has been the human condition since the Cold War. And this book has sort of a beautiful perspective on that – yes, this is the game of life, this is the fundamental condition to which all things lead, this game of mixed motives and cooperation and betrayal is everywhere and always has been.
The final reason is that I read The Strategy of Conflict as a grimoire.
Tacit bargaining is internal magic. On Reddit they say “people aren’t mind-readers”; au contraire, people are nothing if not mind-readers. The ether is in fact a textured geography of concepts; chock full of affordances, archetypes, allusions; ready to be terraformed by the skilled spinner. The hills are alive, the social constructs are tangible, under everything lurks a phantasmagoria of double yellow lines and Formosa straits. This is yin: surfing the material world’s symbolic overlay and giving it the occasional strategic nudge.
Commitment-crafting is external magic – spellcasting. You can tangibly affect reality just by saying things. There’s a terrain here too, but it’s the terrain of incentives and is much less ethereal. This is yang: terraform the ground you’re dueling on.
Happy New Year. Go get chugging on your resolutions at the single most auspicious time of year to do it.
An unrelated but interesting role for mediators: revealing partial information to both parties.
A mediator can consummate certain communications while blocking off certain facilities for memory. (In this regard he serves a function that can be reproduced by a computing machine.) He can, for example, compare two parties’ offers to each other, declaring whether or not the offers are compatible without revealing the actual offers. He is a scanning device that can suppress part of the information put into it. He makes possible certain limited comparisons that are beyond the mental powers of the participants, since no player can persuasively commit himself to forget something.
The computing machines have gotten even better at reproducing this function; The Strategy of Conflict was written 22 years before the Millionaires’ Problem was formulated and solved. Modern zero-knowledge cryptography and fully homomorphic encryption widen the range of possible computations well beyond “certain limited comparisons.”