AI Poses Opportunities, Risks for Higher Education | American Enterprise Institute
Generative AI is bringing Schumpeterian disruption to the stodgy world of higher education. As in other areas of the economy, the AI revolution is changing the university landscape. It is exposing long-standing flaws in the institution, perhaps most fundamentally its overreliance on exams, term papers, and grades as proxies for learning. This moment of creative destruction poses both challenges and opportunities for administrators like me and the students we serve.
Generative AI can help struggling students close the gap with higher-performing peers. Back in 1984, Professor Benjamin Bloom found that average students who receive one-on-one tutoring perform two standard deviations better than students educated in a classroom environment. For four decades, educators have sought to solve “Bloom’s two-sigma problem” by searching for group instruction models as effective as one-on-one tutoring. Generative AI offers a solution by providing a scalable replication of expensive personalized instruction. Two recent studies, one from the Center for Economic Policy Research (CEPR) and another of business school students, show that AI use raises university grades, particularly for lower-performing students, and compresses the curve.
Of course, real learning requires careful attention to how students use generative AI tools. Students learn little by merely asking ChatGPT to answer their problem sets. As my AEI colleague John Bailey writes at length, the most effective AI tutors “orchestrate the conditions for productive struggle.” This means understanding the individual student, identifying his or her area of difficulty, and tailoring lessons that guide the student through the learning process rather than simply revealing the solution. This dovetails with the studies referenced above: The CEPR study cautions that “some of the apparent progress comes from outsourcing cognitive effort to the machine,” while the business school study found that grade compression resulted both from lower performers improving and high performers declining. Useful instruction focuses not on answers, but on the learning process.
This same insight should prompt a fundamental rethinking of higher education. As Dartmouth Provost Santiago Schnell argues, in an essay that should be required reading for every professor or administrator, generative AI exposes a longstanding problem. Universities have long relied upon written outputs such as exam answers and term papers as evidence of education. This has always been an imperfect proxy for learning. Talented students can submit polished prose without truly grappling with a question, and vice versa. Generative AI spotlights this inconvenient truth by divorcing learning from outputs completely. Any student with an internet connection can produce high-quality outputs without engaging with the underlying material, as I discovered when I asked Claude to answer my 2025 Property exam. Unsurprisingly, its response received top marks, which prompted me to shift to a proctored in-class exam this year.
The solution is not simply better forensics to detect AI use. As Associate Dean, I am responsible for investigating instances of suspected generative AI misuse, which violates our university code of conduct. My experience is that there are telltale signs of AI authorship, such as excessive em-dashes, lists of three, and “not X, but Y” sentence constructions, that can reveal clumsy AI-related cheating. But it is sometimes difficult to distinguish carefully prompted AI prose from quality academic writing. A viral 2024 article noted that AI detection software flagged the Declaration of Independence as 98 percent AI-generated. While tools such as Pangram and Winston have improved since then, they are still easy for a careful student to avoid and are susceptible to false positives. This should not be a surprise, as formal, academic prose that our students mimic is similar to materials upon which these models were trained to write—an observation that rings true for the many academics who have scrubbed their drafts of em-dashes for fear of being accused of malfeasance.
The answer, Provost Schnell explains, is a renewed focus on the academic mission. Ultimately, the university’s mission is not to create graduates who can merely produce good outputs. It is learning, the formation of a student’s mind into a tool for discernment, judgment, and the discovery of truth. In the AI age, this means redesigning assessment to better detect and measure what is happening in a student’s mind. Courses should make students struggle with difficult material, ask them to challenge their preconceived notions, and determine what they believe and why. As I tell my students at orientation, if you graduate here believing everything you believe today, you won’t have gotten your money’s worth.
The university is one of Western civilization’s most venerable institutions. As Schumpeter predicted, this moment of creative destruction may be painful but ultimately regenerative. By revealing higher education’s overreliance on outputs as a proxy for learning, the AI revolution is prompting universities back to their mission: forming educated, discerning citizens who pursue truth.
