For a few generations, heading off to college after high school became a rite of passage for many Americans. The academic landscape grew competitive, the names of the oldest and most illustrious universities entered common parlance, and the four-year degree was a ticket into a financially secure middle-class life.
For the longest time it was simply a matter of course that the bright, aspiring young ought to attend university; the fancier and more credentialed the institution, the better.
Even in the 2010s, it was rare for brilliant young people to opt out of the foreordained path that was college. Programs like Praxis — which matched young, entrepreneurial people with business through apprenticeships — offered an alternative, as did starting a business, learning a trade, or simply entering the workforce. But around American dinner tables, the question remained: “Where is your kid going to college?”
Not anymore.
Or, at least, not obviously.
Some survey results have hinted at this for a while, even though the financial logic behind the college boom has long been straightforward. Give up four years of wages, pay tuition and living expenses (perhaps borrow to do so), and receive a credential that lets you earn considerably more for the rest of your working life. The degree was an investment in human capital — the noncynical story goes — and depending on your chosen profession (and how lavishly you spent on those credentials, e.g., high living expenses versus working on the side and graduating early), you could reasonably expect to earn back what you lost fairly quickly.
As I described in a piece for The Daily Economy last year, the calculations were never quite that simple: The relevant comparison isn’t the lifetime earnings of graduates and non-graduates, but the incremental earnings attributable to the degree, net of tuition, financing, and four years of foregone income. Regardless of how much of higher education’s investment returns reflected raw ability and how much was a signal, the underlying wage premium was still a real, observed phenomenon. And for decades it was large enough to make the college bet look fairly safe.
A new pre-print by researchers from UPenn and the University of Navarra, titled “The College Wage Premium in the Generative AI Era,” has some bad news for the college-bound kids: the economic case for a degree has rarely looked this uncertain. The wage premium has declined precipitously in just a few years, bucking a long-term trend in the US. The authors, José Azar, Mireia Giné, and Javier Sanz-Espín, explain the significance:
“The post-2022 contraction in the US college wage premium marks a structural shift in the returns to higher education. In long-run supply-and-demand accounting, 2022–2026 represents the first documented historical period characterized by a sustained, absolute decline in relative demand for college labor.”
A four-year college degree was a scarce good whose value (the wage premium it attracted over what was otherwise available) rose as the economy demanded more professionalized, high-cognitive labor. For decades, the supply of that sort of worker exploded out of the universities. Now, many of the cognitive tasks their jobs require are becoming easier to automate or reproduce via the AI revolution. That’s contributing to the falling college wage premium. “Employers and officials have long argued that as the numbers attending university have ballooned, a degree has ceased to be a good proxy for job-relevant abilities outside of elite institutions and vocational subjects,” John Burn-Murdoch summarized for the Financial Times’ “Data Point” column.
The (unadjusted) wage premium peaked at about 80 percent in 2001, after which it hovered around that level — until the post-pandemic work-from-home and AI revolution have recently reshuffled the playing field.
At some level none of this is unsurprising (markets work!): rising returns to education were contingent on a specific postwar, turn-of-the-century set of economic circumstances. Of course. The race between education and technology told us as much decades ago; Claudia Goldin and Lawrence Katz’s series of scholarly articles dates from the 1990s! The book that developed the framework, The Race between Education and Technology, came out in 2008.
Writes Burn-Murdoch again: “there is nothing inevitable about more education being met with more demand for those skills — it all depends on how the supply of particular skills interacts with the technological paradigm of the day.” The college premium wasn’t a fact of nature or an economic law, just the result of particular technological change fleetingly favoring one set of skills.
The workplace shift in previous centuries, from at-home artisans to intricate business environments and from manual labor to the service sector, created a premium for educated workers, and gave them good pay. The widespread schooling and high school education in the first half of the twentieth century eroded that. The information/digitalization era did something similar again, pushing technical competency way up the income ladder — with growing education efforts following. America did what rich societies tend to do well: make more of the scarce, highly rewarded thing. The new tech complemented educated workers and enriched them.
In that informational revolution, the computer didn’t replace the college graduates but instead made them more productive. Today’s infinite-generation and agentic-AI revolution might turn the tables on that dynamic; technological change is not obliged to remain skill-biased in the same way forever. Importantly, the authors find a minority role for AI in causing the falling premium; they find that wage growth in AI-exposed occupations has been weaker than elsewhere. Some 70 percent of the skill premium contraction, they write, remains unexplained: competing explanations involve labor market tightness in low-wage sectors, market power, firm sorting, or shifts in remote work flexibility.
It shouldn’t surprise us either that college enrollment is stagnating as the premium has weakened, even though intuitively slower growth of college-educated labor shouldn’t coexist with falling wage premium for that labor. The difference is that potential college attendees make a rough estimate of lifestyle and lifetime income with or without a degree. In a digital world of AI and many nontraditional opportunities, they observe that what pays is excellence and value, not credentials.
In a version of the same question that every AI commentator has brought up in recent years, Azar, Giné, and Sanz-Espín conclude their paper by wondering whether generative AI will replace “high-cognitive tasks” (and thereby permanently reduce the returns to schooling) or just temporarily displace workers, which then “ultimately yields skill-reallocation productivity gains” for everyone.
Parents of college-age kids learned the wrong lesson from what the world had shown them during their lives: send the kids to college and they’ll earn higher wages — paying back even eye-popping loans won’t be a problem. College was the way to gain an advantage in a competitive labor market.
Consequently, there was an explosion of exactly the credentials everyone thought they needed to escape the competitive labor market pressures from everyone else.
College may still pay: The wage premium remains large. For decades, a college degree was associated with a more and more valuable kind of skill for a technologically hungry labor market. Then America produced millions of people with those skills, and now AI is making some of the cognitive output abundant, too.
The question for the next generation is which kinds of education will prove scarce enough to command a premium and generate a positive return on educational investments, all things considered.
