Higher education as class commitment

Bridging game theory and common sense

In a previous post, I argued that Bryan Caplan’s signaling theory isn’t a good explanation for why college graduates get higher-paying jobs. Instead, I claimed, understanding the role of higher education in the modern West requires sociological explanations. In this post I argue more specifically that getting an undergraduate degree serves as an initiation into a class of cultural elites, variously called the bourgeois bohemian (bobo) class, the professional-managerial class (PMC), the Blue Tribe, globalists, symbolic analysts, class X, or simply the ruling class. I think of each of these labels as grasping one part of the elephant, but I haven’t yet pinned down a unified description; I’ll mainly use the “PMC” terminology in this post, for reasons I’ll explain in the next section.

Under this explanation, college is the same kind of thing as a fraternity hazing process, or a military boot camp: it demarcates members of a group, via a process which reorients new members’ motivational systems to favor the group they’re joining. College graduates therefore benefit from the nepotism of existing members of their class, which they perpetuate when they gain the ability to make hiring decisions. This lines up well with Bourdieu’s hypothesis that the primary purpose of modern education is to reproduce existing class structures via “consecration” of graduates by academic authorities. However, today’s class dynamics are very different from those of the mid-1900s. The modern PMC is very politically skewed, and implements class nepotism on a visceral emotional level: for example, via deep contempt for populists (e.g. Brexit voters or MAGA supporters); disgust for challenges to modern feminism (e.g. from Scott Aaronson or Robin Hanson); and fear of falling off the elite career ladder.

There’s a way in which my hypothesis is just common sense: of course attending Harvard is a path into the American (and indeed global) elite, and of course employers who strongly prefer Harvard students are tracking some kind of status hierarchy. Meanwhile it’s obvious (e.g. in voting patterns) that there’s a significant cultural divide between those with and without college educations. Unfortunately, no major intellectual traditions have explored these dynamics very deeply. Academic sociologists are loath to characterize themselves as central members of an elite class (and have anyway abandoned the standards of rigor required for sustained intellectual progress).

Other intellectual clusters (like the rationalist cluster I describe here) have developed a kind of “immune response” to vague sociological thinking, and therefore fall back on the closest hypotheses they can formulate in broadly economic language. However, as my previous post argued, this misses a lot. So the main goal of this post is to take a step towards more rigorous sociological explanations of higher education and many other phenomena, by characterizing class dynamics in terms of game-theoretic interactions between rational-ish agents.

I start by drawing on Scott Alexander’s concept of a backscratcher’s club to explain the core benefit of class membership. However, it’s difficult to explain how long-term loyalty to a backscratcher’s club could be incentive-compatible for people who don’t terminally value it. So in the second section, I argue that college serves as a commitment mechanism by changing students’ identities. This is distinct from Caplan’s signaling explanation because signaling is a process of conveying information about existing traits, rather than changing those traits. (Note that I’m specifically avoiding talking about changes in students’ values, because a big part of what it means to be committed to a class is that you’ll defer to the class consensus over your own values. This makes students’ values something of a red herring.)

In the final section, I characterize the resulting class as a “social equilibrium” which is able to propagate and defend itself. (I’ve previously discussed a related concept using the term “distributed agent”, e.g. in this talk, this post, and these slides.) I also give a case study of a (more extreme) historical example of this kind of social equilibrium—namely Chinese foot binding—and its analogies to higher education.

College alumni as a backscratchers club

I’ll start by discussing reciprocal favoritism at smaller scales, before moving on to class-level nepotism. One common-sense benefit of college attendance that I didn’t discuss in my previous post is forming a professional network. This need not be via traditional “networking”—simply making friends is sufficient. A friendship is (amongst other things) a reciprocal arrangement where you expect to favor each other. And the friends you make at university (especially a top university) are likely to be unusually successful.

On an individual level, this can be very advantageous. However, it doesn’t explain why employers so strongly prefer to hire college graduates. When you’re a new grad, your network largely also consists of new grads, who aren’t very helpful to know yet. And while in the longer term having a better network might let you do your job better, it also gives you more opportunities to jump ship to other jobs.

So to explain the college wage premium we need to generalize the concept of reciprocal favoritism. Here I’ll borrow Scott Alexander’s concept of a “backscratchers club”: a group with the rule “club members must favor other club members over outsiders”. (Though this seems like a straightforward concept, I wasn’t able to find any term with the same meaning in the academic sociology literature; pointers welcome.)

Scott notes that an important aspect of such clubs is having some barrier to entry—otherwise everyone could just join it, making favoritism towards other members meaningless. He also argues that such clubs work best when they have plausible “fig leaves” for what the club is about, with favoritism as an implicit norm rather than a legible rule. Being part of a well-functioning backscratchers club should increase your access to desirable jobs, because members of the club will preferentially hire other members (and punish other club members who aren’t nepotistic) in order to remain in good standing within the club.

As a first pass, then, we can explain college attendance as the process of joining a set of overlapping backscratchers clubs. There’s your core friend group; your student groups (including fraternities and sororities); the alumni of your particular university; and finally the group of all university graduates. Picture a hiring manager who sees a resume mentioning their old fraternity or alma mater, and puts it at the top of the pile out of a sense of camaraderie or loyalty. My hypothesis is that the wage premium for college students is explained by a similar kind of loyalty amongst college graduates as a whole, plus a sense that they’ll face social punishments for being less nepotistic (e.g. disproportionate blowback if they hire an outsider who performs badly).

I had intended, at this point, to give a more thorough explanation of Bourdieu’s theory of education and class. However, this led me to procrastinate on finishing this post for six months. So for now I’ll focus on a different tradition of class analysis, centered on Barbara and John Ehrenreich’s concept of the professional-managerial class (PMC). This class was originally understood primarily in economic terms—e.g. James Burnham’s The Managerial Revolution offered an alternative to both the capitalist and the Marxist worldviews. However, over time it became more sociopolitically-oriented—in part because “the values of that class completely took over left-liberal political spaces” (as per Catherine Liu, who has written extensively on this new conception of the PMC).

Unfortunately, most of these authors are still working within the PMC’s value system even as they critique the class itself. From a more independent (but still broadly PMC-aligned) perspective, Scott Alexander’s characterization of the “Blue Tribe” does a much better job at capturing the depth of its animus against its class enemies. (I also recommend Sarah Constantin’s concept of “Ra” as a depiction of one closely-related memeplex.) Scott’s Blue Tribe terminology helps to convey that membership isn’t primarily about one’s job title, but rather one’s culture and identity.

There’s some tension between a cultural conception of the PMC, and my characterization of the PMC as being gatekept by formal credentials. I don’t mean to imply that an undergraduate degree is strictly necessary for PMC membership; such illegible coordination at such large scales will always involve somewhat fuzzy boundaries. However, it still seems surprisingly important, in a way which (I’ve argued) can’t be explained by either the standard human capital model or Caplan’s signaling model—for example, undergraduate credentials are often considered even when hiring people with decades of work experience. In the next section I’ll discuss why I think college degrees aren’t just strongly correlated with PMC membership, but are actually playing a load-bearing role in how it operates and propagates.

Initiation rituals as commitment mechanisms

For a backscratcher’s club to remain stable, club members need to reliably follow its rules. However, there are at least two reasons why it’s hard to incentivize them to do so. Firstly, there’s a strong time asymmetry: people primarily benefit from their status as university graduates early in life, but primarily have the ability to favor other graduates later in life. So there’s a clear incentive to accept the favor of other club members when you’re young, but then fail to reciprocate when you’re older—e.g. by hiring the best person for the job rather than another club member.

This might be prevented if club members also punish defectors. However, without any central authority capable of doing so, enforcing group norms is another kind of work that needs to be incentivized somehow. You need people who will punish others for breaking norms, and people who will punish others who don’t punish others for breaking norms, and people who will punish others who don’t punish others who don’t punish others for breaking norms, and… Such equilibria are technically possible—the game theory folk theorems show that a very wide range of equilibria can be stable. But in any realistic setting, a group would fall apart if everyone were only doing the bare minimum they could get away with. (Kevin Simler hypothesises that a “minimum viable superorganism” can avoid this problem, but doesn’t explain how it could avoid internal nepotism.)

So a formal model of backscratchers clubs needs some mechanism by which joining a club actually commits you to following its rules. Credible commitments are in general hard to make, because they require adversarially constraining your future self. I’ll talk about three potential mechanisms for making such commitments—though note that they’re all closely related. The first is simply sunk costs. If you’ve invested a lot of work in climbing one particular social hierarchy, then leaving it to climb another hierarchy becomes less attractive. You may need to find new friends, learn new skills, and incur additional costs to demonstrate your commitment to your new hierarchy.

However, even given these sunk costs, leaving for a healthy faction can easily be more long-term beneficial than staying in an unhealthy faction. For example, I expect that many humanities graduates from the early 2010s would be making more money now (and be happier) if they’d retrained as plumbers after graduating, rather than entering unhealthy white-collar job markets. And sunk costs matter less if other factions will let you rise in them without making additional costly commitments—e.g. the relatively meritocratic hiring policies of Silicon Valley companies make it easier for people in other industries to jump ship.

Another commitment mechanism: requiring people to “burn their bridges” with other factions to rise in their current faction. For example, getting a prominent gang tattoo makes it more difficult to later leave the gang and reintegrate into society. And a politician who viciously attacks another party will tend to have more trouble switching to that party later on. Again, however, this depends on other desirable factions maintaining high barriers to entry, rather than having mechanisms by which former enemies can be forgiven and integrated.

A third way of making credible commitments is by setting up external constraints, like legal contracts. This is difficult to do directly for class membership without that class losing the advantages it gains from illegibility. However, the PMC has managed to weaponize civil rights law to indirectly enforce many constraints that benefit it—e.g. by making it borderline illegal to hire based on IQ tests in America. Overall, legal constraints seem like an important part of the picture, but not the main mechanism by which class loyalty is maintained.

Instead, the main mechanism by which I think college creates class commitment is by rearranging people’s motivational structures, in particular by making their self-esteem dependent on success by the standards of the PMC. This is hard to analyze from an economic perspective; instead, we need to draw on more psychological concepts. In particular, I model most people as being in a constant process of “ego defense”—that is, working to find or maintain internal narratives (like “I’m a well-respected member of my community”, or “I’m a world-class researcher”) which are consistent both with the evidence available to them, and with their sense of social and physical security. A group can induce loyalty in such a person by crowding out their existing narratives in favor of narratives where group membership is a key pillar of their lives. (This seems closely related to what Bourdieu calls “symbolic violence”.)

More concretely, this can happen via putting people through intense experiences which pit their desire to join that group directly against other identities. Narratives in which group membership is crucial (which “predict” compliance with extreme aspects of the initiation ritual) will therefore be reinforced; while other conflicting aspects of one’s identity (like individual dignity) which “predict” noncompliance will be negatively reinforced.

In this post I’ve been using the concept of “identity” in a fairly vague way. Unfortunately, this is because I don’t yet have a clean definition; my best attempt is something like “narratives about themselves that guide agents’ instinctive behavior and conscious deliberation in self-fulfilling ways”. I expect that a better understanding of frameworks like active inference and Garrabrant induction will allow us to pin down this concept much more cleanly (see here, here and here for three steps in that direction). For now, though, it seems better to think of “identities” in a pre-theoretic way, as the kind of thing we appeal to when trying to describe patriotism, or ego, or Stockholm syndrome.

Intense experiences seem to be particularly effective at shaping identities when they demarcate a sharp boundary between who’s in and who’s out of a club (or between different discrete levels of membership). I’ll call such experiences initiation rituals. The binary nature of initiation rituals makes coordination to enforce norms much easier, because there’s less room for (strategic) ambiguity. This helps explain the sheepskin effect: why graduating from college helps your career much more than dropping out one semester early. With a class as large as the PMC, the boundaries will always be fuzzy nevertheless. But the binary distinction between having a degree or not remains surprisingly relevant for many employers even when hiring people with decades of work experience.

Initiation rituals (especially coming-of-age rituals) are a common feature of pre-modern societies, and are often extremely unpleasant. For example, Scott Alexander summarizes one here:

When an Aboriginal man comes of age, he gets mock-abducted from his parents by unrelated elders and forced to spend five to fifteen years going through a series of grueling initiation rituals. During some parts of the rituals, he must spend months isolated from all other humans, living on his own in the wilderness. During others, he endures horrible mutilations, starting with circumcision and only getting more horrifying and creative from there.

Initiation rituals in modern societies are much less intense, but follow some similar principles. Two of the more obvious examples are hazing processes for fraternities, and military basic camp. Each of these can be very unpleasant by modern standards, typically including a combination of sleep deprivation, physical pain, and social humiliation. Candidates might have many reasons for continuing with this process—e.g. a sense of stubbornness, or concern for their reputation. But regardless, they’ll also be reinforcing the internal narrative that, if they’re willing to put up with this much pain, they must really value joining the group. The process of breaking psychological resistance to group loyalty involves making the initiation ritual painful enough that this narrative is hard to question. Even a long time afterwards, changing your mind involves admitting that you made a very costly mistake—often a psychologically painful process.

Compared with these initiation rituals, college lacks intensity, but makes up for it in length. You typically spent three or four years away from your family and community, surrounded by other young people trying to form their own identities. It’s therefore very easy for those identities to end up anchored to an external status hierarchy—especially for the many students who go into a lot of debt to attend college, and are relying on their newfound class privilege to make that choice financially worthwhile. Dan Wang has a very compelling blog post on the resulting Girardian terror, which I’ll quote at length:

It’s hard to construct a more perfect incubator for mimetic contagion than the American college campus. Most 18-year-olds are not super differentiated from each other. By construction, whatever distinctions any does have are usually earned through brutal, zero-sum competitions. These tournament-type distinctions include: SAT scores at or near perfection; being a top player on a sports team; gaining master status from chess matches; playing first instrument in state orchestra; earning high rankings in Math Olympiad; and so on, culminating in gaining admission to a particular college.

Once people enter college, they get socialized into group environments that usually continue to operate in zero-sum competitive dynamics. These include orchestras and sport teams; fraternities and sororities; and many types of clubs. The biggest source of mimetic pressures are the classes. Everyone starts out by taking the same intro classes; those seeking distinction throw themselves into the hardest classes, or seek tutelage from star professors, and try to earn the highest grades.

There’s very little external intermediation, instead all competitive dynamics are internally mediated. The prizes are so obvious. The big companies that come to career fairs soothingly assure high-status jobs; the speakers at convocation tell us that we too will become as successful as them one day; our peers hold leadership positions at clubs, get internships at exciting companies, and earn those chances to have lunch with the university’s president.

In his blog post School is to Submit, Robin Hanson discusses a related hypothesis: that university attendance became so common because it trained students to submit to hierarchies. However, if university graduates acquired a willingness to submit in general, we’d see more of them submitting to (well-paid) blue-collar work, right-wing politicians, etc. Instead, a version of Hanson’s hypothesis which better fits our observations is that modern universities instill the desire to submit specifically to the PMC hierarchy. One indicator of the strength of this effect is how graduates who come from very wealthy families often end up working the same jobs, for the same salaries, as graduates who actually need money. This is very hard to explain in economic terms, but makes perfect sense if those graduates view jobs at McKinsey or Meta (or increasingly OpenAI or Anthropic) as primarily paying them in social status.

Moving beyond individual rationality

So far in this post, I’ve tried to remain reasonably close to the classic “rational individuals” framework. My account of initiation rituals, Dan’s account of Girardian terror, and Robin’s account of submission all describe non-rational mechanisms by which people’s identities change—but they’re still the kind of experience which a rational-ish individual could in principle opt into for the sake of other benefits (e.g. better jobs). I’ve focused on this level of analysis because I think it has the potential to develop into a sociological framework that’s much more rigorous than current academic sociology, as I’ll discuss at the end of this post.

However, while I’m optimistic about the long-term potential of this kind of reasoning, in the short term it can easily miss the forest for the trees. So in this final section of my post I want to zoom out to give a more holistic sense of what’s going on with higher education and the PMC. As one starting point: there’s an important disanalogy between processes like fraternity hazing or basic training and higher education. The former are run by people who want to select and shape initiates to be useful for their institutions. Whereas insofar as college graduation serves as an initiation into the PMC, that happens without any centralized planning or top-down enforcement of class boundaries.

Instead, many people’s individual decisions are steered towards what they think others in their class will support. This kind of distributed coordination can superficially resemble a conspiracy, but actually differs in many ways. In a previous post, I used the term “consensus of power” for the extreme case where individuals’ values don’t matter at all. A more general term, from notable sociologist Talcott Parsons, is “social equilibrium”: a social system which, when disturbed, tends to return to its original state. My understanding (mainly from Fable) is that Parsons didn’t consider social equilibria to be modellable by game theory, precisely because this kind of robustness requires players to internalize shared norms as part of their identities.

Social equilibria are all around us (e.g. tipping culture, or dating culture), but ones with the breadth and influence of the PMC are rare. One that’s somewhat comparable is the Indian caste system, although it’s much more explicit than the PMC. A more implicit (but also very influential) social equilibrium is beauty standards. We typically think of beauty standards as being maintained using relatively little coercion. However, beauty standards linked to class dynamics can motivate extremely costly commitments, like female genital mutilation in Africa or foot-binding in China.¹ I’ll zoom in on the last (and its parallels to higher education) as a case study.

Foot-binding was a horrific practice in which young girls’ feet were compressed hard enough to break bones. After months or years of being bound, their feet permanently deformed to become much smaller and painful to walk on. It also happened a lot: the cultural practice of foot-binding lasted for over a millennium and affected hundreds of millions of women. Wikipedia estimates that “by the 19th century 40–50% of all Chinese women may have had bound feet, rising to almost 100% among upper-class Han Chinese women”.

The nominal purpose of foot-binding was to make girls more beautiful. From our current vantage point, it’s obvious that this was just a fig leaf. One way of systematically identifying fig leaves is asking whether there are much better and cheaper ways of getting the nominal goal. We can also apply this standard to universities: the fact that there are much better and cheaper ways of learning job-relevant skills undermines that explanation for the college wage premium. In turn, though, the fact that there are much better and cheaper ways of signaling desirable traits undermines Caplan’s signaling theory of universities.

Taking the comparison further: husbands who choose to marry women with bound feet are analogous to employers choosing to hire university graduates for much higher salaries than they’d need to pay equally-talented high school graduates. You might object that university students are old enough to rationally opt in to this process, whereas Chinese girls had foot-binding forced on them by their parents. However, the actual reasons why most students go to university seem to have less to do with rational deliberation, and more to do with having already internalized conformist motivations. I like Hazard Spence’s description of primary and secondary school as “lay[ing] the groundwork for stupefaction and domestication” (see also his annotated Open History of American Education).

At this point we’ve moved far enough away from the ideal of individual rationality that I’ll shift to talking about social equilibria as agents in their own right, which are able to propagate and defend themselves using decentralized coordination between their human subagents. One canonical example of this kind of “distributed agent” is Scott Alexander’s ‘toxoplasma of rage’, a reconceptualization of entrenched political conflicts as agents carrying out strategies optimized to reproduce themselves. Compare how the foot-binding process was often carried out by a girl’s mother or other female relatives—once co-opted into the social equilibrium, they played a key role in perpetuating it. Analogously, college graduates often act to further the PMC’s interests directly via roles in academia, and indirectly via their support for the current education system and the elite political consensus.

Another feature of distributed agency is coordinated behavior to defend the current equilibrium. At one point the Chinese government (specifically the foreign Manchu dynasty) tried to ban foot-binding, but was thwarted by the resistance of the Han Chinese population. To explain this, we need to ascribe the foot-binding equilibrium the ability to generalize to defending itself in novel situations, by mobilizing its adherents towards coordinated action.

Modern educated elites can similarly display a level of collective vitriol that’s hard to explain except as a defense of their class interests. The most obvious example is the PMC’s contempt for blue-collar workers, and for the “Red Tribe” of America more generally—not just its politics, but also its pastimes, culture, and religion. When political movements have arisen that oppose the PMC’s interests—most notably Trump, Brexit, and the “far right” across Europe—cultural elites have mobilized in near-lockstep against them, with very forceful (often to the point of hysterical) rhetoric.

More recently, the PMC has also taken an adversarial stance towards what Scott Alexander calls the “Gray Tribe”, an outgrowth of the Blue Tribe which has formed a distinct power structure over the last few decades, and is therefore being treated as a defector by the Blue Tribe. For example, mainstream journalists adopted an across-the-board hostility to Silicon Valley (and especially to initiatives like the Thiel fellowship which offered an alternative to higher education). Meanwhile, academics often dismiss intellectual progress presented in non-standard ways. The example of this that I’ve observed most directly is the ML community’s failure to engage seriously with the concept of AGI risk (as I discuss in more detail here). Caplan and his colleagues in the GMU economics department themselves provide another example. They demonstrate that professors can be far more intellectually productive when they refuse to submit to standard academic norms—and, relatedly, have faced widespread criticism.

When we encounter any given example of such a response, it’s tempting to ascribe it to an individual’s personal values. But the uniformity of these responses is a strong indicator that they’re instead tracking, and being guided by, a kind of class-wide consensus. However, I don’t mean to imply that the PMC is thereby actually furthering its class interests. As the enormous backlash of the last decade suggests, this level of hostility towards its rivals seems like “irrational” behavior from the PMC. More specifically, it seems like the kind of groupthink-driven overreach that consenses of power are prone to. Unfortunately, the Gray Tribe doesn’t understand what’s going on well enough to avoid reimplementing these harmful dynamics themselves. For example, the main purpose of yCombinator now seems to be forming a similar kind of alumni community as top universities do. (Meanwhile, the AI safety community made the opposite mistake of not gatekeeping enough, which led to the breakdown of alignment research as a meaningful concept.)

Although this post has discussed distributed agents in an informal fashion, much of my research aims to develop a formal theory of distributed agency as rigorous as other leading theories of intelligent agency. This will require progress in game theory and epistemology. Ultimately, I hope that it will ground out in a theory analogous to microeconomics or evolution via natural selection, which describe the emergence of complex structure from multi-agent interactions. However, this post has tried to demonstrate how, even in the absence of the full theory, these ideas can help a lot in understanding the shape of the modern world.

Thanks to Michael Vassar, Ben Landau-Taylor and Hazard Spence for conversations which inspired many of these ideas.

1. I’m also very interested in the beauty standards of peacocks, and the extent to which existing evolutionary frameworks are sufficient to describe them. However, I’ll save that for another post.