Predictable Identities: 1 – Guess What’s Coming

This entry is part 1 of 9 in the series Predictable Identities

Can you predict the last word in this sentience? It’s not “sentence”. The last word in sentience research is that most of what our brains do is try and predict the signals they’re about to receive, like the words you read on a page. Prediction shapes our perception, which is why that word appeared as “sentence” the first time you read it.

Our brains implement predictive models at multiple levels, from general worldviews to detailed patterns. When reading a text, your brain starts predicting the language and theme based on its model of the publication. This drives the prediction of sentences based on a model of grammar, prediction of how words should be spelled, and finally a detailed prediction of how characters should display.


  1. This looks weird.
  2. This looks wiedr.
  3. Looks weird this.
  4. Your mom looks weird.

Our brains optimize for predicting incoming signals over our entire lifetime. This is achieved in two ways: doing a good job of predicting inputs right now, and learning new models that will allow us to make great predictions in the future. Does reading the post so far feel unpleasantly confusing? That’s because the content was too unpredictable, and contradicted too many of your existing models of how brains work. Did it feel awesomely mind-blowing? That’s the joy of acquiring a new model that offers a condensed explanation of what you already know, and thus a promise of better predictions to come.

In either case, you should learn more about the predictive processing paradigm of cognition from this series of articles or this book review; this blogchain is mostly done covering the established science. Instead, we’re going to forge forward irresponsibly and use predictive processing to explain political polarization, identity, war, and adjunct professorship.

Do you think you know what’s coming?

Predictable Identities: 2 – Active Inference

This entry is part 2 of 9 in the series Predictable Identities

One sentence recap of Part I: our brains are constantly trying to make true predictions about the world. We do it in two ways:

  1. Assembling good models that make accurate predictions.
  2. Changing the world to match our predictions.

When Apple released a buggy version of Maps, The Onion joked that “Apple is fixing glitches in Maps by rearranging Earth’s geography”. That’s exactly what our brains do.

Actions are driven by predictions propagating across different levels. We shoot a basketball by forecasting the flight of the ball, which leads to predicting that we will lift the ball and push it, culminating in precise anticipations of the required tension in the arm muscles. We are satisfied when the ball flies according to our projection and upset when it doesn’t

That’s why predicting well is so important to our evolved brains – when we predict well we know how to act to achieve our goals. A predictable environment is an exploitable environment.

Of course, basketballs are not a big component of our milieu, and our ability to predict them isn’t crucial. What is vital for us to model above all else are people, from faraway strangers to neighbors and friends. Also ourselves: prediction happens at different parts of the brain simultaneously, and each module has to predict what the others would do, now and in the future.

Predictive processing expert Sun Tzu observed:

If you know the enemy and know yourself, you need not fear the result of a hundred battles.

We observe other minds, interrogate them, and push them to conform to our models of them as best we can – all to maximize our predictive power and capacity to act effectively. This is a powerful lens through which to observe how we interact with others, and how we build our own predictable identities.

Predictable Identities: 3 – Prisoner’s Dilemma

This entry is part 3 of 9 in the series Predictable Identities

Much of human interaction is shaped by the structure of the prisoner’s dilemma. We put in place institutions and norms to enforce cooperation. We tell shared stories to inspire it. We evolved moral emotions to achieve cooperation on an interpersonal level: empathy and gratitude to assure cooperators of our cooperation, anger and vengefulness to punish defectors, tribalism and loyalty to cooperate with those we know well.

But the crux of the prisoner’s dilemma is that defection is always better for the defector. We try to get others to cooperate with us, but we also try to defect as much as we can get away with. We want our peers to pay their taxes, admit mistakes, share credit, and stay faithful. We also fudge our taxes, shift blame, boast, and cheat.

There are many strategies for dealing with PD, and some of them can be formalized in code and entered in competitions with other strategies. The simple strategies are named and studied: tit-for-tat responds to each play in kind, tit-for-two-tats forgives one deception in case it was a mistake, Pavlov changes tacks after being defected against and so on and so forth. Which strategy works best?

It turns out that the success of each strategy depends almost entirely on the strategies played by opponents. Each approach can fail to reach cooperation with others or under-exploit opportunities to defect; even a strategy of always defecting is optimal if enough other players always cooperate. If you only knew what your opponent is playing, you could always choose the best option.

And this brings us back to predicting other humans. If we can model their strategies, if we know who will be forgiving and kind, who will be vengeful and dangerous, we can play optimally in any situation. Predicting well is the unbeatable strategy.

Predictable Identities: 4 – Stereotypes

This entry is part 4 of 9 in the series Predictable Identities

How do we predict strangers?

Humans evolved in an environment where they rarely had to do this. Practically everyone a prehistoric human met was familiar and could be modeled individually based on their past interactions. But today, we deal with ever-stranger strangers on big city streets, in global markets, online…

We need to make quick predictions about people we’ve never met. We do it using stereotypes.

Early research on stereotypes focused on their affective aspect: we dislike strangers, but less so as we get to know them. But new studies have looked at the content of stereotypes, finding that groups are judged independently on the dimensions of warmth/competitiveness and status/competence. For example, Germans see Italians as high-warmth low-competence, i.e. lovable buffoons; Italians see Germans as the exact opposite, i.e. mercenary experts.

These dimensions are primarily about predicting someone’s behavior and capability. In game theory terms: Will that person cooperate? And can I safely defect on them or do I have to play nice?

Contrary to the well-meaning wishes of most stereotype researchers, there is robust empirical evidence on stereotype accuracy. The short of it is that people’s stereotypes are, in fact, quite accurate, especially stereotypes of gender and ethnicity. Whether stereotyping is good or bad, it cannot simply be wished “educated” away. It is universal because it’s very useful for prediction.

A smart person noted that problems arise from having too few stereotypes, not too many. If you have a single stereotype for “Jews” you’re doing a bad job of modeling Jewish people, and are likelier to mistreat them due to prejudice. If you have separate stereotypes for Hasidic Jews, Brooklyn conservative Zionists, liberal Jewish atheists, secular Israelis etc. you’re one step closer to treating (and modeling) unfamiliar Jews as individuals. Studying cultures is about acquiring many useful stereotypes.

Predictable Identities: 5 – Outgroup Homogeneity

This entry is part 5 of 9 in the series Predictable Identities

There are more ways for someone to be different from you than to be similar. But psychologically, it works the other way around. We perceive those like us as uniquely distinct and the outgroup as undifferentiated. It’s called the outgroup homogeneity effect.

This effect extends to physical appearance (e.g. faces of other ethnicities looking the same) and mental traits (e.g. people of the other gender all supposedly wanting the same things). Surprisingly, the effect is unrelated to the number of ingroup and outgroup members one knows; it’s not about mere exposure.

I recently wrote about a remarkable case of the outgroup homogeneity effect: Ezra Klein’s strange attempt to make the case that liberaltarian podcaster (and Klein’s co-ethnic) Dave Rubin is a reactionary.

Klein starts by looking at the network graph of podcast appearances which links Rubin to several unsavory reactionaries. But Klein himself is just two podcasts removed from Richard Spencer, so that’s not great. He then defines “reactionaries” narrowly as those who seek “a return to traditional gender and racial norms”. Of course, most of Rubin’s flagship positions (gay marriage, drug legalization, abortion access, prison reform, abolishing the death penalty) have to do with gender and race norms. Specifically: changing them.

I think what happened is that Ezra Klein picked up intuitively on the one important similarity between Rubin and conservative reactionaries: they both strongly dislike him.

People legislate the distinction between pies and tarts and between plums and nectarines, but only geologists care about telling apart inedible rocks. Same with people: it’s important to keep track individually of potential cooperators, reciprocity relationships, etc. But once you model someone as a defector, you don’t need more detail to predict that they’ll defect. The outgroup is good for writing snarky articles about. For this, you don’t have to tell them apart.

Predictable Identities: 6 – Creeps

This entry is part 6 of 9 in the series Predictable Identities

We’ve looked at predicting people from a distance: employing stereotypes and homogenizing outgroups. Moving a step closer to the self, consider an individual you have just met. What are you looking for in the first few minutes of interaction? Among other things, often first among them, is predictability.

A member of your tribe is highly predictable. If you share a taste in clothes and podcasts you can predict with high confidence how they’ll react in social situations, what their habits and motivations are, etc. That’s why small talk about last night’s episode of Game of Thrones isn’t a waste of time. It communicates: I am like you, you can model me well by looking at yourself, we can cooperate.

Second best is someone who fits well in a group stereotype, even if that’s not your ingroup. You may not befriend many middle-aged bearded men who speak a language you can’t recognize, but you feel comfortable getting in a taxi with one.

Now consider someone who looks like you, watches the same shows, and also eats spiders they catch around the house. Would you get in a taxi with them?

The opposite of predictability is creepiness. In the leading research paper on the nature of creepiness, the author describes it:

The perception of creepiness is a response to the ambiguity of threat. […] While they may not be overtly threatening, individuals who display unusual patterns of nonverbal behavior, odd emotional responses or highly distinctive physical characteristics are outside the norm, and by definition unpredictable.

Eating spiders by itself doesn’t make one dangerous, but it’s a signal that this person can’t be predicted well by pattern matching to stereotypes or by self-modeling (unless you’re an arachnophage yourself). We react to creepiness not with fear but with aversion. We dislike what we cannot predict.

Predictable Identities: 7 – Weirdness Budget

This entry is part 7 of 9 in the series Predictable Identities

Social life needs predictable people who conform to expectations. There are common expectations which apply to every person in the group, and personal expectations based on one’s perceived role and past behavior.

Deviating from common expectations costs idiosyncrasy credits. Dressing differently, eating strange diets, watching documentaries that no one else does and not watching the show everyone talks about – these all exhaust a limited budget of nonconformity. The budget accrues to people who conform or are popular; the two often go hand in hand.

As Homer Simpson noted: “Marge, I can’t wear a pink shirt to work, everybody wears white shirts. I’m not popular enough to be different.”

When someone exceeds their idiosyncrasy budget, their opinions will get dismissed on grounds of absurdity bias and the horn effect. If you want to tell your millennial friends about crazy ideas like unfriendly AI or Ribbonfarm, make sure you are otherwise appropriate.

Personal expectations include playing out roles and simply being the same person over time. Even changes of mind that don’t violate group norms, like hating a movie that you liked last year, can be grating. If you’re not changing in ways that are fun for your friends, you’re just making yourself hard to model and get along with.

Imagine a conversation with just a few people. When deciding whether to tell a joke, you have to consider your friends’ likely reactions to the joke. But you also have to consider their reaction to everyone else’s reaction, and what it implies about everyone’s relationships and status, and so on down many layers of metacognition. If you can’t rely on heuristics of shared group expectations and consistency over time, the computational effort is overwhelming. You’re likelier than not to go look for a more predictable group of friends to joke with.

Predictable Identities: 8 – Roles People Play

This entry is part 8 of 9 in the series Predictable Identities

We fit strangers in stereotypes, and we like strangers who fit our stereotypes and act congruently. Dealing with people we know allows for more personalized modeling, but we still want associates and companions to stick to their roles.

Most everyone in most every office resembles a character on The Office, not just in the broad strokes of the Gervais Principle but down to details of job, dress, and personality. Husbands and wives have played out “if it weren’t for you” patterns long before Games People Play described them.

The main difference from predicting strangers is that dealing with familiar people allows for active inference: enforcing others’ conformity to prediction. Valentine Smith describes this in his excellent essay The Intelligent Social Web:

You move away [from your family] and make new friends and become a new person […] But then you visit your parents, and suddenly you feel and act a lot like you did before you moved away. You might even try to hold onto this “new you” with them… and they might respond to what they see as strange behavior by trying to nudge you into acting “normal”: ignoring surprising things you say, changing the topic to something familiar, starting an old fight, etc.

This nudging is effected by thousands of small actions perpetrated by dozens of people. We receive small negative reinforcements when we do something unpredictable that causes others’ models to momentarily fail, and positive rewards when we conform to our roles. The social life of an office or a family is too complex to compute without stable roles assigned to people, the same way a brain can’t cohere a visual scene without the expectation that visual objects remain stable.

Life in the social web means that growth and change are harder than they appear. Hard, but not impossible.

Predictable Identities: 9 – How to Change

This entry is part 9 of 9 in the series Predictable Identities

You’ve changed your mind and who you want to be. Your social web is having none of it. What to do?

One option is to simply power through. Stick to your (new) guns and admit that your mind is indeed changed. The “zeal of the convert” is important here: it takes extra commitment to convince those who knew the old you to change their story.

A savvier approach is to leverage common tropes and adopt a role that entails transformation. If you want to switch from layabout to responsible professional, or vice versa from workaholic to self-caretaker, it may help to have a romantic partner break up with you. “Area person reassesses life priorities after heartbreak” is a familiar story that people can get behind. “Area rationalist reassesses life after Hamming circleisn’t, even if that’s what actually happened.

But the easiest option may just be to GTFO and change the scene. I’ve reset my social web several times, usually with positive results.

By the time I learned how to be funny in elementary school, everyone had decided that I’m a boring nerd. I moved to a new city for high school and did better as the clever class clown, but these traits were not in high demand when I enlisted in the military. I eventually grokked professionalism, but it was too late to gain the trust of my officers. I emigrated. I did better in my next two stops but blew through my weirdness budgets and couldn’t shake off my reputation as a weirdo.

Finally, I came to New York, where I did not know a single soul. I took my craziest opinions out of the office and onto the internet. I started acting like the person I wanted to be seen as, not who I was before.

It works, for now.