The Vibe Handbook, Part Two: Understanding the World Through Vibes

Vibe-aware communication

I have vibes in my head. You have different vibes in your head. What now?

Language acquisition

Here’s a bad way to teach someone English: hand her a list of five hundred vocabulary words with definitions and tell her to memorize it. She’ll dutifully make flash cards and be able to regurgitate the definitions when you test her. But then she walks up to an English speaker and freezes. Is “Hello” an acceptable equivalent to “Hi”, or too formal? When the other person says, “How’s it going?”, is she supposed to answer or is it rhetorical? What’s the difference between “going to”, “going for”, “going on”? Is “where’s the bathroom” all right, or should it be “where’s the restroom”?

Even native speakers have this problem. Test a teenager on his SAT words, he’ll get a hundred percent, and then he’ll go around writing sentences like “There are a myriad of possibilities” or “Chapter Six was ephemeral in length”. The problem is that no word can be directly replaced with its definition across a variety of contexts – a word is a complex bundle of connotations and relations. That’s why no one naturally acquires language through memorization. You have to hear a word in context several times to build subtle intuition for the contours of the thing it describes. If I want to teach you the proper use of a word, my best bet is to make up some sentences that include it.

Rhetoric

Here’s a bad way to persuade someone of a position you hold. Get her to explain the reasons she believes what she does, identify the points on which you disagree, and give her some evidence to prove you’re right. Then she should either refute your argument, offer some counterevidence, or give in and change her mind. If she does none of these, conclude that she’s fallen victim to cognitive bias.

Except, outside of hardcore rationalist circles, this basically never happens, because that’s not how opinions are formed, nor should it be. You can simplify an opinion into a logical predicate to communicate its gist to another person, but opinions are actually much deeper than that: they’re broad senses generated by data you’ve largely thrown away, leaving only the resulting changes to the weights of your mental model of the domain. They can be overridden, but only with enough data to shift the weights significantly. If I want to convince you of my opinion, I have to find you some data points that roughly approximate the data my mental model was trained on. (Alternatively, I can build up a reputation in your mind for usually being correct – but that requires me to convince you on several other points, allowing you to generate your own opinion that I’m often right.)

Characterization

Here’s a bad way to describe a person: list a few adjectives. “John is ambitious but also friendly.” You can paint someone with broad strokes like this, but not only do you lose vividness, you lose expressibility. The point of characterizing, whether inventing a fictional person or conveying the nature of a real one, is to flesh out a Type of Guy, and to do this in an interesting way you have to think of such a guy and then select a small set of actions and traits that let the reader unambiguously connect the dots. This is the art of creating a starter pack meme: the best are either untitleable – the title feels more like just another element of the pack – or have titles so specific that one doubts the author will pull it off, and yet once you see the content you realize “oh… that guy.”

The key process in both these cases is reconstruction. You take some complicated input, you learn a compressed essence of it, then you try to communicate to another person what you’ve learned. But the essence is almost always difficult to describe, since you didn’t consciously learn it as a collection of rules – you learned it from examples. You can try to remember some old examples, or make up new ones from scratch, but often the easiest way to come up with them is to consciously go on the lookout for sentences including the word, or anecdotes that align with your opinion, or moments that make you think, “Classic John”. (The plural of anecdote is data – what else could it be?)

Corollary: Black-Hat Vibe Communication

What does it mean to lie with vibes? It means giving bad data – an unrepresentative sample. Even if you’re including statistics, you can transmit an incorrect vibe by cherry-picking stats that give a vibe different from the full vibe of the phenomenon you’re talking about. This is sort of what words like “misrepresentation” and “spin” are getting at, but the popular interpretation of lying is still a dial ranging from True to Mostly True to Mostly False to Pants on Fire. I posit that there are two axes of truth: factuality – how literally accurate are the factual claims – and honesty – how accurate is the vibe. So Pants On Fire isn’t just a more extreme version of Mostly False. Mostly False usually means “at least somewhat factual, but not honest”; Pants On Fire means “not factual, lazily skipping the evaluation of honesty”.

A good novel is honest but not factual. An ill-deserved hit piece in the paper, an out-of-context quote, a resume fluffed up with professionalese, or a strategic cut of a reality-TV heel: factual but not honest. A good documentary is factual and honest. A lie told for the teller’s personal gain is neither factual nor honest.

Of course, it’s much harder to evaluate honesty than factuality. That’s because you have to prove that the sample you’ve been given is unrepresentative – and of course, providing a countersample isn’t enough, since maybe that one is unrepresentative too. Ultimately, you have to either find a way to get a roughly random sample of the thing in question, or form your opinion by deferring to the judgment of someone you trust. Why would you trust their judgment? Either because they having a track record of accurately conveying vibes you’ve confirmed firsthand, or because you already trust someone else who trusts them.

This is why corporate and government “fact-checkers” are useless. Since factually incorrect information is easy to refute, it’s rarely disseminated at scale. The term “misinformation” is used to conflate factual incorrectness with dishonesty: it implies made-up facts, but actual accusations of misinformation almost always allege something closer to dishonesty. So fact-checkers mostly try to refute the vibes of contentions they don’t like. (”My aunt got the Covid vaccine and then had a weird rash for a month.” Facebook: “CDC STUDIES INDICATE THAT THE COVID-19 VACCINE IS SAFE AND EFFECTIVE.” Neither directly challenge’s the other’s factuality – the CDC really has satisfied itself that the vaccine meets its standards, and probably the aunt really did get a rash.) But again, there are only two ways to decide whether to believe in a vibe that has been communicated to you, and being given one new piece of data is not one of them. You have to either trust Meta (or Twitter, or whomever), trust Meta’s report that the CDC approves, and then trust the CDC; or trust the sample you’ve gotten from direct experience.

Vibe-aware governance

So now you can speak vibe. How do you use this to exert influence? It’s easy in the lindiest human environment, which is the small village or tribe – people are purpose-built for transmitting and receiving vibes. All cultures have a layer of vibe structures like religion, tradition, ritual, ceremony, taboo, music, and dance, on top of the base layer of day-to-day social interactions and the norms that govern them. Building social structures at scale and from scratch is much more difficult. Vibes aren’t logical predicates, and they don’t slot neatly into technocracies, bureaucracies, or abstractions.

The trouble with laws

The axiom of a legislative state is “We can write laws that will cover most situations; new laws can be added as needed, but fundamentally, society can be governed by a set of general rules known to everyone.”

People used to think something like this about AI too. For the first thirty years or so of the field, there was significant support for the idea that we could get computers to think by teaching them a sophisticated set of rules. About the best we got with that approach was Deep Blue, the computer that beat Kasparov at chess – it turned out that modern neural-network-based machine learning was the only successful way to achieve more general cognition. And of course, neural networks run on vibes.

All crime is vibecrime. Moral intuition doesn’t come with a rulebook; humans and the world are way too high-dimensional for a succinct set of rules to be enough to make people coexist peacefully. In a society small enough, there’s no need for rules – if your neighbors disapprove of how you act, they’ll sanction you, if they approve, they’ll reward you. But at the scale of civilization, you need something else for cooperation to scale, and generally that’s the rule of law.

But there are problems. For one, laws sometimes lead to outcomes that conflict with most people’s moral sensibilities; that’s the cost of scale. Consider Edward Snowden, for instance: it would probably be a disaster if the punishment for leaking classified military documents were a slap on the wrist, but many people would say Snowden was fully in the right given the circumstances. At the least, he was clearly acting on conscience rather than trying to sabotage the military.

Then there’s Goodhart’s Law: simple measures don’t work if you use them as targets because simple measures are never ends in themselves, only proxies for good vibes. If you make a rule incentivizing the proxy, then you create another force acting on the proxy, which pulls it away from the vibe it was supposed to be a proxy for. Because of this, a law based on a simple measure can’t capture a vibe. But a law based on a more complex measure is far harder to write.

Likewise, if you make a law that imposes a constraint, you invite malicious compliance at worst, loophole exploitation at best.

Financial regulation

Matt Levine writes about an interesting trend going on at the U.S. Securities and Exchange Commission. The context is a settlement between crypto lending startup BlockFi and the SEC, where BlockFi had to pay $100 million (but the SEC also let them release the first officially SEC-compliant crypto lending product) because the regulators decided that BlockFi’s offerings were security-ish enough that they should have been registered as securities.

Here is a dumb model of financial regulation. There are two general ways for authorities to regulate financial activities. Call one “rulemaking.” A regulator thinks about some financial things, decides how they should work, solicits comments from industry participants and the general public, and writes a rule explaining in some detail how things will work. After that, everyone who does those things knows what the rules are and what things they are and are not allowed to do.

Call the other “regulation by enforcement.” The regulator doesn’t spend much time writing specific new rules, but just relies on some general old rules saying things like, you know, “don’t use an artifice to defraud.” What that means can be unclear and, more to the point, if the regulators decide to go after you for using some artifices, your life will be unpleasant even if you ultimately win. Knowing this dynamic, the regulators can use enforcement actions to decide and declare what is and is not allowed. Everyone goes around doing things, and then some of them get sued by the regulator and have to pay a bunch of money to settle the lawsuits. Everyone watches those lawsuits, and each lawsuit tells everyone a bit more about what is and is not allowed.

Rulemaking is hard! What if we tried to do regulation the old-fashioned way, where everyone is mean to you if you’re a bad boy?

One way to think about this is that it is unfair: That person didn’t know X was against the rules, why should she get in trouble? Often though it does not feel that unfair, because X probably does look pretty bad; it probably was at least arguably against some general rule like “don’t use any artifices.” It’s a little random that this person got in trouble for doing X, but it’s not like she was totally innocent either.

They’re going after vibecrime. Instead of optimizing for the letter of the law, now you have to optimize for passing an SEC vibecheck.

In regulation by enforcement, the industry has to put a lot of expensive effort into figuring out what is not allowed, and the risk of missing things is on the industry; the regulator gets to focus its attention on what is actually happening and what it wants to prevent rather than theoretical generalities.

It’s easier to evaluate the vibes of a particular action than to try and condense the vibes of every possible action into a rule.

Companies want to stay in compliance with regulations; the SEC wants to create regulations that align with good vibes. The second thing is harder! Maybe making rule-vibe alignment easier by making less specific laws and using more discretion in enforcing them, and making regulatory compliance harder by making the companies guess how the SEC will use its discretion, distributes the workload more evenly.

Cultural regulation

Another problem with laws is that they can be too weak: there are antisocial behaviors we’d like to sanction, but the cure would be worse than the disease: censoring speech, for instance (risking totalitarian censors), or outlawing being mean to people (too easy, too frequent, too small-scale, hopeless to write laws about, much less enforce them). Maybe we could do extralegal cultural rulemaking instead: culturally disseminate a set of rules for staying in the good graces of polite society. But one, much of culture is decentralized; two, new behaviors get invented every day; and three, every situation has a vast context that would have to be mined for mitigating factors. (Recall the ill-fated proposals for people to sign pre-hookup consent contracts.)

How about cultural regulation by enforcement? Enforcement is unpredictable, but the sanctioned usually understand they were treading in a gray area, sanctions are widely publicized so everyone else gets a new data point about where the vibes lie… wait, I’m describing cancel culture. Instead of holding everybody equally to an official set of rules, there’s an unwritten cultural zeitgeist you have to not fall afoul of, lest you be hit with likely minor sanctions and unlikely major ones. The downside of cancellation is that it enforces a monoculture; the upside is that it does a pretty good job with the enforcement, both in terms of deterrence and not punishing the innocent (at least, innocent of a transgression against the culture, even if you think the culture is wrong).

The nice thing about regulation by enforcement – which is in the same spirit as the paradigms of common law and framers’ intent – is that it gets around Goodhart’s Law. You can’t optimize for the letter of a law that’s unwritten. Vibes don’t have loopholes, because the definition of a loophole is the space between a vibe and its written approximation. And you can’t maliciously comply with a rule that isn’t a rule. Or if you can, maybe that’s what civil disobedience is. I wonder if designing nonviolent protest uses the same muscle as working in compliance at a big bank, but in the opposite direction – flexing instead of extending.

Coming up in Part 3: a complete workout for your vibrational muscle group.