The male–female gap in higher education will widen, and the men opting out are not who you think.

The premise

More women than men attend college in nearly every developed country, and the gap has been widening for three decades. In the US, women now earn roughly 58% of bachelor’s degrees; in Sweden the picture is similar. This is usually framed as a male-underachievement problem in the K–12 pipeline, occasionally as an equity concern in reverse, and rarely as something with serious downstream institutional implications.

The structural drivers are well understood: the collapse of male-coded blue-collar wages (Autor et al., 2019), the expansion of female-coded licensed professions — nursing, teaching, social work, psychology, all gatekept by credentials — and earlier-life sex differences in school engagement. None of this is new.

What I want to argue is that generative AI is going to accelerate the trend non-linearly over the next two decades, that the kind of men opting out will shift in a way that matters more than the headline numbers, and that the consequences run through the credentialing and personnel-assessment system in ways that aren’t yet obvious.

The acceleration

The simple version of the AI story starts with personality. Men are, on average, somewhat less conformist and more risk-tolerant than women. The conformism literature is modest but real (Eagly & Carli, 1981; Bond & Smith, 1996), with effect sizes around d = 0.2–0.3. Sex differences in Agreeableness point the same way. So when a cheap substitute for formal education appears — large language models that will answer your questions at 3am for $20/month — men should be marginally more willing to treat AI as a substitute for the credentialed path, while women are marginally more likely to treat it as a complement.

There is also an asymmetric-returns version (Goldin, 2021), which probably does more work than the personality version on its own. Women bear children, face statistical discrimination in less-formal hiring contexts, and benefit more from portable formal signals. Men can lean on revealed performance in less-gated fields. AI changes that arithmetic against male enrollment at the margin without changing it much for women.

Both stories predict the same thing: AI widens an existing gap. The interesting question is which men are exiting.

The right-tail opt-out

The reflexive answer is to picture men drifting into Eberstadt’s (2016) Men Without Work — screens, games, declining labor-force participation, mostly low-skill. That tail is real and probably grows. But it is not the tail that matters most.

University enrollment has always been a tail phenomenon. The modal man in 1955 didn’t go. What matters for the institutional consequences is the upper tail of the male distribution — the men who went to college historically because the cost–benefit was clearly positive for them. AI changes that calculation specifically for this group, because they are exactly the men most able to extract value from the tools. The Thiel Fellowship logic — that the smartest, most agentic young people are better off not going — generalizes from the 99.9th percentile downward as the cost of self-directed skill acquisition drops.

The prediction that follows is sharper than “more men opt out.” It is that differentially more high-g, low-risk-aversion men opt out, while the remaining male graduate pool is filled by tuition-dependent universities admitting marginal candidates from below. The compositional change at universities understates the underlying shift, because attrition at the top is masked by substitution from below.

This is empirically tractable. Aptitude scores at non-elite universities, conditional on enrollment, should decline. The σ_m/σ_f ratio in graduates’ cognitive ability should compress. The male share in elite postdoc and faculty tiers — already visibly declining — should continue to fall. And the variance of observed competence among male hires from a given university tier should widen, because the right tail of the distribution increasingly does not hold the credential at all.

The signal degrades asymmetrically

Spence (1973) and Caplan (2018) describe the degree as a composite signal of three things: general cognitive ability, conscientiousness, and willingness to submit to institutional norms. AI used inside the university already degrades the conscientiousness signal — students offload work onto models, so the degree carries less information about effort. That part is symmetric across sex.

The right-tail opt-out adds a sex-asymmetric layer. If high-g men disproportionately exit the male pool, then for men specifically the degree increasingly signals conscientiousness and conformity, and less general cognitive ability. The informational content of a male degree drops faster than a female degree. Employers who care about cognitive ability in fields with high g-loaded performance variance face a worse filter when screening men than when screening women.

This sharpens the case for direct cognitive and personality assessment in hiring. The awkward feature is that the differential informativeness functions, in effect, as a male-disambiguation tool. Legal regimes vary in how much room they give for explicit testing — Griggs v. Duke Power (1971) and disparate-impact doctrine constrain US employers; most European jurisdictions, Sweden included, have substantially more headroom. The likely equilibrium involves standardized work samples, structured interviews, and where the law permits, ability and personality testing of the kind that has been quietly refined in industrial-organizational psychology for a century.

I work in this area, so discount the prediction accordingly. But it is also the path of least resistance.

Second-order effects

Assortative mating is already strongly polarized along educational lines. A widening credential gap thins the pool of educationally-matched partners for credentialed women, but the right-tail opt-out makes this stranger than the headline gap suggests: increasingly, the smart, productive, financially-comfortable man is not credentialed at all. The matched-population graph then depends on which women learn to read alternative signals — income, observed competence, founder/maker reputation — and which don’t. Sorting on these instead of credentials redraws the dating market in ways that probably correlate with social class and city. The fertility consequences are starting to be documented in the East Asian and Nordic literatures (Birger, 2015, is a readable popular treatment of the demographics).

Political sorting follows the same line. Educational polarization is already among the strongest single predictors in many Western elections, and a more visible category of high-status non-credentialed men sharpens it further.

Where this leaves us

The interesting claim isn’t that men have personalities that make them go their own way. It is that AI shifts the cost–benefit calculation against credentialing in a sex-asymmetric way that bites especially hard at the upper end of the male ability distribution. Universities respond by admitting marginal candidates. The male degree loses its informational content faster than the female degree. The institutions built around degree-based signaling — including hiring — are about to be put under pressure they weren’t designed for.

Personnel assessment — the unglamorous psychometric infrastructure built up over the last century — is going to get more important, not less. That’s an empirical prediction, and one I’d bet on.

References

  • Autor, D., Figlio, D., Karbownik, K., Roth, J., & Wasserman, M. (2019). Family disadvantage and the gender gap in behavioral and educational outcomes. American Economic Journal: Applied Economics, 11(3), 338–381. https://doi.org/10.1257/app.20170571
  • Birger, J. (2015). Date-onomics: How dating became a lopsided numbers game. Workman.
  • Bond, R., & Smith, P. B. (1996). Culture and conformity: A meta-analysis of studies using Asch’s (1952b, 1956) line judgment task. Psychological Bulletin, 119(1), 111–137. https://doi.org/10.1037/0033-2909.119.1.111
  • Caplan, B. (2018). The case against education: Why the education system is a waste of time and money. Princeton University Press.
  • Eagly, A. H., & Carli, L. L. (1981). Sex of researchers and sex-typed communications as determinants of sex differences in influenceability: A meta-analysis of social influence studies. Psychological Bulletin, 90(1), 1–20. https://doi.org/10.1037/0033-2909.90.1.1
  • Eberstadt, N. (2016). Men without work: America’s invisible crisis. Templeton Press.
  • Goldin, C. (2021). Career and family: Women’s century-long journey toward equity. Princeton University Press.
  • Spence, M. (1973). Job market signaling. Quarterly Journal of Economics, 87(3), 355–374. https://doi.org/10.2307/1882010

Citation

Persson, B. N. (2026). Credentials in the age of the prompt [Blog post]. https://bjorn-persson.github.io/thoughts/credentials/

@misc{Persson2026Credentials,
author = {Björn N. Persson},
year = {2026},
title = {Credentials in the Age of the Prompt},
note = {Blog post},
url = {https://bjorn-persson.github.io/thoughts/credentials/}}