# College Students Using AI Face a Learning Crisis, Brown Professor Warns
Brown University economics professor Roberto Serrano observed something alarming in his classroom last spring. When he administered his first take-home midterm exam, the grade distribution seemed impossible. Students who typically earned Cs suddenly submitted work worthy of A-level performance. Serrano suspected what many educators now fear: widespread reliance on artificial intelligence tools is replacing actual learning with what he calls "cognitive surrender."
The phenomenon Serrano identified reflects a broader crisis spreading across American higher education. Students armed with ChatGPT, Claude, and similar large language models can now produce college-level writing and analysis in seconds. The problem is not that they obtain correct answers. The problem is that they bypass the cognitive struggle that builds genuine understanding.
When students use AI to complete assignments rather than consult it as a reference tool, they skip the mental processes that encode knowledge. Retrieval practice, struggle with difficult concepts, making mistakes and correcting them, explaining ideas in their own words—these mechanisms drive learning. Remove them, and students graduate with credentials but not capabilities.
Serrano's "absurd" grade distribution offers concrete evidence. In take-home exams, students who submit AI-generated work or minimal original thought produce suspiciously uniform quality. The work shows no variation in voice, reasoning process, or the messiness of genuine problem-solving. A student who wrote a C-level paper last month does not suddenly write an A-level paper the next week without intervention or effort. When this happens across an entire class, AI is the likely culprit.
The stakes extend beyond individual courses. Employers and graduate programs rely on college grades and work samples as signals of competence. If those signals no longer reflect actual learning, hiring decisions and admissions become unreliable. A mechanical engineer who used AI to complete thermodynamics assignments may not understand heat transfer principles necessary for safe design. A business student who outsourced financial analysis to a chatbot cannot interpret a balance sheet independently.
Some colleges have responded by banning AI use outright. Others have attempted integration policies that designate when and how students may use these tools. Both approaches struggle. Students can bypass prohibitions through VPNs and secondary accounts. Integration policies require faculty consensus and consistent enforcement, which most institutions lack.
The deeper issue is pedagogical. Faculty designed assignments assuming students would do the thinking. Take-home exams, papers, projects, and problem sets worked when submission required genuine effort from the student. Now they don't. Institutions must redesign assessment itself.
Some professors have shifted toward in-class, proctored work. Others use oral exams where students must explain and defend their reasoning in real time. These methods cost more time and resources, but they measure what matters: whether students actually learned.
Serrano's experience at Brown illustrates the tension many institutions face. They cannot ignore AI. It exists. Students will use it regardless of policies. But neither can they accept a system where students graduate without developing the competencies their degrees claim to certify. Higher education faces a choice: adapt assessment practices to verify genuine learning, or accept that many bachelor's degrees no longer mean what they say.