329 | Steven Pinker on Rationality and Common Knowledge
What this covers
Steven Pinker discusses the mechanics and consequences of common knowledge—the recursive state where everyone knows that everyone knows something—in conversation with Sean Carroll on Mindscape. Rather than treating social coordination as a puzzle of shared information alone, Pinker argues that the presence or absence of common knowledge explains a vast range of human phenomena: why paper currency holds value, how language meanings emerge, why bank runs happen, what triggers embarrassment, and how political revolutions overcome entrenched dictatorships. The episode traces how conspicuous public events generate common knowledge more efficiently than private communication, using examples from reef coral spawning to the dynamics of financial bubbles.
The conversation moves across coordination problems—situations where multiple parties must align their expectations—and explores why these differ fundamentally from cooperation. Pinker applies recursive mentalizing to explain phenomena as distinct as indirect speech (which preserves deniability by avoiding common knowledge) and involuntary displays like blushing (which create common knowledge despite our wishes). He discusses Aumann's Agreement Theorem, which proves that rational agents sharing the same priors must converge to identical beliefs once they make their positions common knowledge—a result that helps explain why disagreement persists in practice, since people typically hold different prior assumptions and doubt each other's rationality. The episode also examines how autocrats suppress media and internet access precisely because these tools generate common knowledge of discontent, and how conspiracy theories spread partly through false beliefs about how widely others share them.
Pinker argues that common knowledge—not mere shared knowledge but the recursive state where everyone knows that everyone knows—is the hidden foundation of social coordination, explaining everything from money and language to bank runs, embarrassment, and political revolutions.
- Social relationships and conventions are ratified by common knowledge, not just private universal knowledge
- Conspicuous public events are the primary mechanism for generating common knowledge
- Failures or absence of common knowledge explain coordination problems like resisting dictatorships or financial panics
Common knowledge is the solution to coordination problems where mutual benefit requires parties to know each other's moves and that the other knows their moves.
- Unlike altruistic cooperation, where the scientific puzzle is the danger of being exploited, much human working-together is mutualistic where both parties win; the distinct scientific problem of mutualistic coordination is one of knowledge—how parties end up on the same page—since it isn't enough to know what the other will do, you must know that the other knows what you will do, requiring common knowledge as its solution.
“The problem in coordination is one of knowledge, namely, how do you end up on the same page, given that it isn't enough to know what the other guy's going to do. You have to know the other guy knows what you're going to do”
- Common knowledge—the recursive state where everyone knows that everyone knows something—is the precondition for being social at all, because conventions (driving on a side of the road, valuing paper currency, respecting authority), informal relationships, and collective action all depend on parties being on the same page about what each other knows.
“common knowledge, one could argue, I do argue, is the reason that we can be social in the first place. Namely, that common knowledge is necessary for conventions”
- Pinker argues the notion of group selection as an explanation for altruism is rather confused, defending kin-selection-style accounts in his paper 'The False Allure of Group Selection,' though he concedes he is not the one to judge his own steel-manning of opponents.
“you can Google a paper that I wrote a number of years ago called The False Allure of Group Selection... I think the whole notion is rather confused.”
- Because natural selection all else equal favors selfishness, the evolution of altruism is a weighty scientific puzzle, addressed since Dawkins's Selfish Gene through mechanisms like reciprocity and reputation.
“how could altruism, probably a better term, ever have evolved, given that all things being equal, you'd expect that natural selection would favor selfishness... there's been a lot of discussion on how cooperation can evolve through reciprocity, through reputation”
- Many wars are fought over saving or losing face, honor, and humiliation rather than material interest; Pinker argues the war in Ukraine is really about Russia's desire to undo its humiliation at the hands of the West.
“an awful lot of wars are fought over saving face, losing face, honor, humiliation, including the war in Ukraine right now. What is it about? It's really about Russia's desire to undo their humiliation at the hands of the West.”
Public, conspicuous, self-evident events generate common knowledge most efficiently, enabling simultaneous action that private signals or polls cannot achieve.
- The surest and quickest way to generate common knowledge is a public, conspicuous, self-evident event: when something happens where you see it, see everyone else seeing it, and they see you seeing it, common knowledge is established in a single stroke rather than through laborious recursive reasoning.
“if something is public, conspicuous, self-evident out there, that can generate common knowledge in a stroke. Not always, but that's the surest way to do it.”
- In the spinach-in-teeth puzzle, N people with spinach all clean their teeth on the Nth glass-clink purely by recursive reasoning from a public announcement; this works only because the chair's statement is common knowledge—if she had whispered to each privately, the inferences from others' inaction would not go through.
“But that crucially depends on common knowledge. It wouldn't work if the department chair went over and whispered something in everyone's ear.”
- A public demonstration coordinates resistance better than even a poll showing majority discontent, because the poll—unless itself made common knowledge—leaves each dissident still individually punishable, whereas simultaneous public action only occurs when people know others will act at the same time, which requires common knowledge.
“public demonstrations can generate common knowledge when everyone in a public square can see everyone else. And that can give them the safety in numbers to coordinate resistance”
- Even brainless organisms like coral solve coordination problems through a conspicuous public event rather than common knowledge: lacking the ability to think or move, they synchronize spawning a fixed number of days after the full moon so eggs and sperm meet, using the full moon as a shared signal analogous to a common knowledge generator.
“they use the full moon as the, in a sense, common knowledge generator. A fixed number of days after the full moon differs for different species, they engage in what marine biologists call the Great Barrier Reef Annual Sex Festival.”
Rational agents with equal information should converge to identical posteriors through Aumann's theorem, yet real disagreement persists due to differing priors and doubts about others' rationality.
- Aumann's theorem is an exercise in epistemic humility: there is no justification for privileging your own assessment over that of others you grant are equally rational, nor any guarantee that truth lies halfway between two opinions, so treating argument as a contest to win rather than a path to truth is a counterproductive bad habit.
“there really is no reason to privilege your own assessment on the assumption that other people like you are rational.”
- Real-world disagreement persists despite Aumann's theorem largely because people have different priors and because they aren't convinced the other party is being rational; Cowen and Hanson note people disapprove of others bending evidence in their favor yet do exactly that themselves.
“they suggest that people are... they don't approve of other people bending the evidence in their favor, setting up the straw man, but they do it themselves.”
- Under Aumann's theorem, when two rational agents exchange posteriors, they do not gradually converge to a middle position as we intuitively expect; instead their positions follow random walks that can leapfrog and lurch to extremes before suddenly landing at the same conclusion—more like a dance than a battle.
“they don't gradually converge and meet somewhere in the middle, which is kind of how we expect arguments to go. Their positions are random walks that end up in the same place, but could go every which way. They could leapfrog each other.”
- Aumann's Agreement Theorem proves that two rational Bayesian agents with the same priors who make their posteriors common knowledge must arrive at the same posterior—they cannot agree to disagree—even without sharing the underlying evidence, because each should update on the mere fact of the other rational agent's stated credence.
“Two rational agents with the same priors in the Bayesian sense... who then make their posteriors common knowledge... those posteriors must be the same. That is, it is, they cannot agree to disagree.”
- In imperfect-information strategic standoffs like poker bluffing or a hockey shooter versus goalie, the optimal strategy is to be perfectly random, because any deviation or preference is a tell the opponent can exploit—yet generating true randomness is very hard for humans.
“an ability to be perfectly random is an advantage, because as soon as you deviate from being random, that is something your opponent could use against you... The optimal shooter and the optimal goalie have a mental random number generator.”
Involuntary nonverbal displays and indirect speech function strategically to generate, suppress, or maintain plausible deniability of common knowledge in social interaction.
- Humans take signals evolved in other primates and repurpose them: eye contact, generally a threat signal among primates where a locked gaze precedes a fight, is generalized in humans into a broader common-knowledge signal that can equally mean threat or romantic interest depending on context.
“we humans just kind of take what we evolved with and we repurpose, we repair it. So eye contact, which in other primates is generally a threat signal... in humans... it's a signal that what has so far been private knowledge between us is heretofore common knowledge.”
- Indirect speech like euphemism, innuendo, and hints work by avoiding common knowledge: even when there is no plausible deniability of the message's meaning, there remains plausible deniability of common knowledge of the message, which lets speakers convey intent while preserving the social relationship.
“even though there's no plausible deniability of the message, there's plausible deniability of common knowledge of the message. And with the additional claim that our social relationships are ratified by common knowledge”
- One reason it took so long for AI to understand language is that algorithms decoding grammar and word meaning misjudge speakers' intentions, because people use euphemism, innuendo, and indirect requests whose intended meaning differs from their literal meaning.
“one of the reasons it took so long to have AI understand language is that if you simply give it the algorithms for figuring out who did what to whom based on the rules of grammar and the meanings of words, it will misjudge people's intentions.”
- Involuntary nonverbal displays—eye contact, blushing, laughter, crying—function as common knowledge generators because they are simultaneously felt by the producer and visible/audible to observers, so each party knows the other perceives the signal; we often try to suppress them (stifling a laugh, avoiding eye contact) precisely to avoid generating common knowledge.
“Blushing, you feel the heat inside your cheeks at the same time as you know other people can see the reddening on the surface of your cheeks. And they know that you know that you're blushing... all of these are common knowledge generators that sometimes we try to avoid.”
- Self-conscious emotions like embarrassment and shame are driven not by the mere detection of a faux pas but by the common knowledge that it was detected; experiments showed people felt far more embarrassed during a karaoke performance when they believed the judges knew that they knew they were being observed, confirming common knowledge as the acute trigger.
“what makes someone feel self-conscious is not so much that some faux pas or infraction was detected, but rather that you acknowledge that it was detected. That is, it's the common knowledge that drives the acute embarrassment.”
- Socially avoiding an obvious truth—the elephant in the room—is a form of common pretense: even when something is privately and reciprocally known (a speech impediment, obesity), people commonly pretend the opposite is the case, and Pinker argues this can be dysfunctional, citing Lindy West's view that the burden of pretending distorts things.
“The common knowledge is that we pretend as if something opposite to reality is the case... it shows the tension between what everyone knows and what everyone knows that everyone knows, and this common pretense, this elephant in the room metaphor”
Common knowledge drives financial and consumer phenomena including bank runs, hoarding, bubbles, and crashes through recursive anticipation of others' expectations.
- Bank runs are driven not by an initial conspicuous signal but by recursive mentalizing about common expectation: a rumor leads each depositor to reason that others will withdraw because they fear others will withdraw, so each withdraws first while money remains, causing the run they feared.
“they're gonna withdraw their money because they're afraid that other people will withdraw their money if only out of fear still other people will withdraw their money. I better withdraw my money while there's still money to withdraw”
- Hoarding (like COVID toilet paper buying), financial bubbles, and crashes are phenomena of common expectation: people act on what they imagine others will do, creating self-fulfilling shortages or asset runaways through recursive anticipation, not through any real initial scarcity or fundamental value change.
“people hoarded toilet paper because they thought there'd be a shortage of toilet paper, which they then caused by hoarding the toilet paper, even though there hadn't been a shortage in the first place.”
- Keynes explained speculative bubbles and crashes via a beauty contest analogy: investors don't pick the asset of greatest underlying value but the one they think the most other people will pick, requiring recursive anticipation of others' anticipation, which can produce runaway appreciation (the greater fool strategy) until the supply of greater fools runs out.
“you had to pick the face that the most other people picked as the prettiest, knowing that they were picking a face trying to outguess everyone else picking it. And he said that would often involve the second, third, and fourth order of anticipation of anticipation of anticipation.”
Autocrats and resisters contest control over information and signals precisely because common knowledge enables collective action and coordination against rulers.
- Autocrats suppress freedom of the press and the internet precisely because newspapers, magazines, and social media generate common knowledge of shared discontent; the Arab Spring was kindled by social media until dictators learned to control it, and even blank signs or leaflets are punished for their common-knowledge-generating power.
“That's why autocrats don't allow freedom of the press, why they have censorship and repression. The Arab Spring was kindled by social media, by Facebook and Google, until dictators kind of cottoned on to that danger”
- Overthrowing dictatorships is a coordination problem: even when a populace overwhelmingly wants change, no individual will move first because solo resistance is crushed, so rulers stay in power by controlling one person at a time; public demonstrations, free press, and coordinating signals defeat this by generating the common knowledge and safety in numbers that enable simultaneous resistance.
“public demonstrations can generate common knowledge when everyone in a public square can see everyone else. And that can give them the safety in numbers to coordinate resistance”
Scientific culture's bias toward paradigm-threatening and consensus-overturning findings produces systematic errors by discarding well-founded priors in un-Bayesian fashion.
- A provocative heuristic holds that roughly 90% of what's in scientific journals is false while 90% of what's in textbooks is true, because confirming the consensus rarely earns publication while novel contrarian findings—often the false ones—do.
“He says, 90% of what's in the journals is false. 90% of what's in the textbooks is true.”
- The bias in science journalism and science itself toward paradigm-threatening, consensus-overturning discoveries is responsible for much error because it is un-Bayesian: it discards well-founded priors and treats the latest tidbit of evidence as reason to overturn an entire established understanding, contributing to the replicability crisis.
“the reason that it's a recipe for error is that it's very un-Bayesian. It's throwing out the priors. It's treating the latest little tidbit of evidence as if it was reason to change your entire understanding.”
- Academic freedom is indispensable not because academics are privileged but because converging on truth requires the open process of publishing claims and allowing others to attack and falsify them; canceling or punishing people for their beliefs disables this process and prevents discovering what is true or false.
“without academic freedom, you can't converge on the truth because the process you just described... we'll only know when other people get to attack it, try to falsify it. If you disable that process by canceling or punishing someone for what they believe, you're never going to find out what's true or false.”
Meaning, currency value, word usage, and social conventions are grassroots common-knowledge phenomena without top-down legislation or universal agreement.
- No one legislates the meaning of words; meaning is a grassroots phenomenon consisting of common knowledge of how a word is used, so when people collectively start using a word differently, that becomes its meaning, and children and speakers tacitly assume this shared convention rather than polling everyone.
“No one decides, no one legislates the meaning of words. It's a kind of grassroots phenomenon where if people start interpreting a word or using a word differently, that is the meaning of the word. The meaning of the word is common knowledge of what it means.”
- FDIC deposit insurance and the bank's conspicuous gold seal prevent bank runs not merely by reassuring people their deposits are insured, but by generating the common knowledge that everyone else knows deposits are insured, removing the incentive to withdraw before others do.
“The purpose of that seal is not just to reassure people that their deposits are insured, but to reassure them that other people know that they're insured. So it's less likely that the bank will fail.”
- The value of paper currency derives not from intrinsic worth but from common knowledge: I treat it as valuable because I know others treat it as valuable, which they know only because they in turn know that still others treat it as valuable.
“respecting paper currency, which, you know, what's the value on a green piece of paper? The value is that I know that other people treat it as having value, which they only know because they in turn think, know that other people treat it as having value.”
- Common knowledge is always relative to a particular community or network of knowers; a signal that is common knowledge within one in-group (such as a co-opted hand sign) may carry no such meaning for outsiders, which is how covert in-group signaling and deniability operate.
“common knowledge is relative to a community of knowers. You have common knowledge within some network. And if you're not part of that network, what's common knowledge to all of them may not be common knowledge to you.”