Predictions that generative artificial intelligence will make universities obsolete are becoming a staple of today’s technology commentary. If large language models can tutor students cheaply and at scale, the argument goes, the economic case for mass higher education collapses. Why pay for classrooms, professors and campuses when an AI can explain concepts, generate exercises and answer questions on demand? In this narrative, universities will survive only as credentialing bodies or elite social clubs, while mass college attendance becomes a relic of the pre-AI economy.

I am skeptical of such predictions. In fact, if the labor-market effects of generative AI prove to be as revolutionary as some predict, there is good reason to expect demand for higher education to increase rather than fall.

Predictions about generative AI replacing professors fail to recognize that most students lack the self-discipline required for self-directed study. Deadlines, evaluation, and regular accountability are central to the learning process. While it is possible that generative AI systems can develop some of these features, they are unlikely to replicate the full force of in-person instruction, where expectations are clearer, feedback is harder to ignore, and disengagement is more visible and costlier.

Some might counter that the accountability function of higher education could be separated from instruction itself. That is, colleges could simply employ monitors to ensure students are using generative AI productively but not dishonestly, while leaving the actual learning largely self-directed. But knowing how and when to use generative AI productively is not a simple skill; it requires domain-specific judgment that depends on disciplinary standards, developmental stage, and the goals of a particular assignment. In some contexts, reliance on AI short-circuits learning; in others, it enhances it.

Designing curricula and assessments that exploit such complementarities, diagnosing misunderstandings that AI use can obscure, and cultivating students’ capacity for independent judgment all require subject-matter expertise. A drill-sergeant model can’t teach students what counts as good reasoning, credible evidence, or responsible use of powerful tools.

By contrast, there is a strong case to be made that generative AI could increase the demand for colleges and universities. Inexperienced workers in high-skill occupations are widely expected to bear the brunt of AI-driven disruption, while demand for more experienced workers is unlikely to fall in the near term. This creates a potential dilemma for firms: slower hiring of junior workers today implies a smaller pool of more skilled workers tomorrow.

But, creating more skilled individuals out of less skilled ones is a key reason why educational institutions exist in the first place. Firms have little incentive to provide broad, general training to workers, because they cannot fully capture the returns when workers are free to leave for competitors. As a result, general training opportunities on the job are undersupplied. Higher education addresses this gap by shifting the cost of such training to workers themselves. By making inexperienced individuals even less attractive to employers, generative AI strengthens, not weakens, the economic rationale for higher education.

This doesn’t mean that every college-goer will now need to pursue a graduate degree. With deliberate effort, colleges can (and should) adapt their curricula to emphasize skills that generative AI complements, such as problem-solving, judgment, and creativity, while placing less weight on tasks that are likely to become relatively less valuable. These adjustments need not require longer time to degree. Colleges should also recognize that, over time, firms will adapt as well, likely in part by redesigning entry-level roles to pair inexperienced workers with generative AI in ways that raise the productivity of both.

If colleges respond successfully to these labor-market shifts, the result could be a win-win: a lifeline for institutions facing declining enrollment as the demographic cliff sets in and a safety net for young workers confronting reduced demand in the early stages of their careers due to generative AI.