Coordination, Common Knowledge, and the Retreat of Liberal Enlightenment

Steven Pinker

It would tank. It would be almost nonexistent just because all of our economy depends on coordination, on people doing things that are in everyone's interest if everyone does them the same way, beginning with money itself, with currency. Why do we accept a piece of paper in exchange for something of value? We know other people will accept it and give us something of value. Why do they do that? They know that still other people will. Everyone knows that everyone knows that money has value. That's what gives it value, as long as they do.

That can unravel when you have hyperinflation, or a bank run, or a currency attack. Political power works the same way. No government can have snipers on every rooftop and telescreens in every room surveilling you 24/7. A lot of our recognition of authority — who's the boss — depends on just common knowledge: she gets her way because she knows I'll obey; I'll obey because I know she gets her way. For as long as that common knowledge lasts, you have structures of authority and power.

Since it depends on nothing but common expectations, it can unravel — as when you have a big public demonstration where people see each other seeing each other opposing the regime, and then the authority of the regime can evaporate.

Pinker

Common knowledge, in the technical sense, does involve this infinite number of arbitrarily embedded beliefs about beliefs about beliefs. The dots in my title signal that common knowledge is not just when everyone knows that everyone knows, but when everyone knows that everyone knows that… and so on.

You're right that we couldn't think through an infinite number of "I know that he knows" thoughts — our mind starts to spin when we do three or four. Instead, common knowledge can be generated by something that is self-evident, conspicuous, salient — that you can witness at the same time that you witness other people witnessing it. That can grant common knowledge in a stroke. You have reason to believe that he knows that I know that he knows, et cetera, even if you don't literally think that thought consciously.

Pinker

Eye contact is an instant common knowledge generator. You're looking at the part of the person looking at the part of you, looking at the part of them. You've got instant common knowledge by the mere fact of making eye contact, which is why it's so potent in human interaction and often in other species as well.

There are even species that can coordinate without literally having common knowledge. Coral have a coordination problem. They're stuck to the ocean floor. Their sperm have to meet another coral's eggs. They can't spew eggs and sperm into the water 24/7 — it would be too metabolically expensive. What they do is coordinate on the full moon. On the full moon, that's the day where they all release their gametes into the water. Of course, they don't have common knowledge in the recursive sense — it's implicit in the logic of their solution to a coordination problem, namely the public signal of the full moon, which evolution has guaranteed each of them can sense simultaneously.

In the case of humans, because we can have recursive mentalising, there's no limit to the number of things that we can coordinate in evolutionarily novel ways by setting up new conventions.

Pinker

There could be a direct link between a public conspicuous signal and the desire to coordinate — and that's probably what happens most of the time. In cases where there's doubt as to whether the signal has been received or how it's interpreted, that's when recursive mentalising does come into play. The choice of words in a fraught situation would be an example — a veiled bribe, a sexual come-on, a solicitation for a big favour — where you put mental effort into how to frame your overture so that your hearer catches your drift without knowing that I know that they know, which could change the basis of a relationship.

I suggest that human relationships — friendship, authority, romance, transaction — are all coordination games solved by common knowledge. Sometimes, when we want not to change the relationship in force, we actually take steps to avoid common knowledge, as in innuendo and euphemism, where the message gets through but there's enough uncertainty that the relationship can survive.

Pinker

A Schelling point is more like common salience. It's closely related. A Schelling point is a choice that people make to coordinate that has no intrinsic value other than the fact that everyone else knows that it exists. Schelling's original example is a couple separated in New York before the era of cellphones. How do they coordinate? It can't just be that each tries to guess where the other would go, because if he knows she likes to go to the bookstore, and she knows he likes to go to the camera store, and he knows that she knows — you get an infinite regress.

If there's some point in the landscape that commends itself as sufficiently salient — like in Schelling's example, meeting at the clock at Grand Central Station at noon — that's a Schelling point. Something salient enough that people alight on it as a solution to a coordination problem in the absence of common knowledge.

Schelling himself pointed out this is very fragile and this is where game theory, rational analysis, logic leave off, and human whimsy, caprice, even poetry, take over. Anything that pops into mind as a possible solution is a solution — because it popped into your mind, and it's likely to pop into the other person's mind too. There doesn't have to be any rhyme or reason other than human psychology.

Pinker

I think a somewhat different case from an outright lie would be a genteel, benign hypocrisy — a common pretence. Robin Hanson gives examples in his book with Kevin Simler: if someone drinks from a whiskey bottle in a paper bag on a park bench, the paper bag isn't really fooling anyone, but the police may choose not to arrest him under the pretence that he's not drinking in public. Why? If they try to enforce the letter of the law, it would be too much like a police state. On the other hand, if they fail to arrest someone who is overtly drinking, they'd be surrendering the authority of the rule of law in general. The paper bag allows them to say the rule of law still holds even if we choose, as we say, to look the other way.

The interesting idiom: "look the other way." There are other cases in law enforcement that work like that — escort services, tobacco paraphernalia. It's a question of where you draw the line so that the law doesn't forfeit all of its authority — which depends on the common knowledge that the rule of law applies.

Pinker

Of course it is, but the official policy of the United States is that it doesn't exist, and of pretty much every other country. Taiwan included, in a funny way.

It's not exactly a noble lie, but it's a common pretence. If you spell out the reason, that undermines the reason. The reason is that formally recognising Taiwan would be a challenge to China's sense of authority and prestige, because China stakes a claim to all of historic China. If nations act as if Taiwan was a sovereign state without saying it's a sovereign state, China can maintain the pretence that its claim is valid.

Yet another example: does Israel possess nuclear weapons? Their official policy is "we're not saying yes, we're not saying no." Everyone knows that Israel has nuclear weapons — if there wasn't some suspicion, the weapons would be useless because they wouldn't be a deterrent. On the other hand, if it was part of official policy that Israel was a nuclear state, its Arab rivals would feel they had to acquire them as well so as not to lose face.

Pinker

One could say that determinism is necessary for the incentive structure that holding people responsible imposes. If I condemn you, punish you, shame you, hold you responsible for bad choices, you're less likely to make those bad choices. Paradoxically, you need some degree of determinism for the idea of holding people responsible for their choices to have any effect.

Pinker

Yes. Of course, it's dependent on the assumptions that go into the theorem. This is the famous theorem that rational agents cannot agree to disagree. More accurately: rational agents who have the same priors, and who share their posteriors without sharing the evidence, must have the same posteriors. The assumptions are seldom satisfied, but if they are, yes.

The corollary of Aumann's theorem, which sounds totally counterintuitive, is that rational agents should not engage in speculative trade — buying a stock in the expectation that you know that its value will rise and you can sell it at a profit. This is exactly what investment analysts tell us: don't try to time the market, just buy an index fund. That's a corollary based on the similar intuition that since prices are common knowledge in a large public market, any hint you have that some company is coming out with an insanely great gadget is almost certainly already reflected in the price.

Pinker

I'm one of those freaks who just does not enjoy the taste of chocolate. This is a case where epistemic peers don't share priors. When it comes to just the brute-force gut feeling — I just don't like chocolate — that's a case where it isn't irrational to not share those priors. That's the way I'm wired.

Pinker

I don't agree with you on that.

Pinker

We aren't. Crucially, there can be no element of just raw subjective preference — it has to be whether the film meets criteria that we both agree are reasonable. Then I'd be open. When it comes to just preferences, priors don't apply or are even meaningful.

Pinker

I sure hope not. I'll say: not gone forever, but certainly in retreat. Why not gone forever? It's got some built-in advantages. One is a reverence for science — the truth is what refuses to go away when you stop believing in it. Any method optimised for finding the truth has just that built-in advantage. And universalism: liberal enlightenment depends on universal human rights, the maximisation of human interests. There is something logically incoherent in saying "I deserve rights because I'm me and you don't because you're not me." I can impose that by brute force, but to the extent that we all have to agree on something, we're not going to agree on me being imperial leader.

I don't believe in historical momentum — I think that's mystical. But some things do have an awful lot of historical momentum. Like human sacrifice. All the ancient civilisations did it. Now it's gone. I don't think it's coming back. Why? A combination of knowledge about cause and effect — throwing a virgin into the volcano will not bring good weather — and some degree of empathy. Likewise, legal slavery no longer exists.

Liberal democracy is on its heels. There is some retreat. But we're still better off than we were in the '90s, the '80s, the '70s in terms of liberal democracy. I don't think it's gone forever.

Pinker

The invasion of Ukraine, rewinding the clock a hundred years, could have been like the assassination of the archduke in Sarajevo. It could have been that NATO decided to go all in, and something like World War I or even worse could have happened. As awful as Ukraine is, it's not as bad as World War I. If what you say is correct, things could suck, but they could suck much worse if we had a world war. I think it is less likely than it was a hundred years ago.

Pinker

The most basic answer is that liberal enlightenment is not particularly intuitive. It doesn't really fit our human nature in the sense of how we conceive of our societies ought to be organised. What's much more natural is tribalism, deference to authority, conformity. Liberal enlightenment is a great idea — it's like a lot of counterintuitive discoveries of organised rationality: markets, science. All are commendable, even if not intuitive.

There's a natural backsliding. Unless it's constantly propped up, people are going to say the solution to our problem is we need a strong leader. The idea that a leader just temporarily presides and is bound by fiduciary duties laid out with checks and balances — it's a fabulous idea. It's not particularly intuitive, in the same way that markets are not particularly intuitive, in the same way that science is not particularly intuitive, but they're still good things.

Pinker

Yes, I am. I don't know whether it's that linguists themselves have not made themselves useful enough, or the cynical view that linguistics has just been on the wrong track and has nothing to contribute. I tend to think more of the former. There are features in large language models that actually implicitly build in some of the insights of linguistics. The fact that you've got all of these hidden layers is an acknowledgment that the regularities in language are abstractions — not triggered by individual words, but patterns of patterns. Attention heads are basically processors that look at temporally separated words in search of regularities. They're trained on tokens that correspond roughly to morphemes, and on embeddings, which are an indirect representation of meaning.

There are a lot of discoveries of linguistics that are implicitly built in, but linguistic theory in the sense of what you see in Linguistic Inquiry doesn't have direct counterparts. Maybe they wouldn't need all of those gigawatts and GPUs if they were able to cut to the chase quicker by building in linguistic insights. My hunch is yes, but the models are so complex that I can't say for sure.

Pinker

It has changed my intuitions — these models are capable of more than I would have guessed, given how they're built. But that may show that my intuitions break down at the scale of hundreds of billions of parameters and ninety hidden layers, training equivalent to thirty thousand years if a child was listening to it. My intuitions may be useless at those scales.

The fact that you have to train these models with the equivalent of thirty thousand years, while a human child can speak in two years, suggests that they're doing something human-like in a very different manner than the human brain. The fact that they are so vulnerable to confabulation and hallucination, as my former student Gary Marcus insistently points out, is a hint that they're doing things differently from how the human brain does them.

Pinker

I really like that song. I think "I've Just Seen a Face."

Pinker

One possibility is to revisit all the indicators of progress. What's the report card on the world ten years after I published Enlightenment Now? What progress has continued? Where have we backtracked? In a sense, answer your question: what are the prospects for liberal enlightenment?

Pinker

Thanks so much, Tyler. Always a pleasure.