News · 2026-07-08

A Brown professor moved a final in person, and the near-perfect scores collapsed

A Brown University economics professor who suspected his students were leaning on AI moved his take-home final into a proctored in-person room — and the near-perfect scores collapsed by roughly half. The take-home version had averaged around 96 out of 100 with dozens of perfect papers, far above the historical norm; under supervision, the average fell to a level much closer to what the course used to produce. It is one of the clearest single illustrations yet of a question haunting higher education: when AI is a click away, what does an unsupervised exam actually measure?

Key facts

• A Brown econ professor switched a take-home final to in-person, and average scores fell by roughly half.
• The take-home version averaged about 96/100 with dozens of perfect scores — well above the historical average.
• A Princeton survey cited alongside it found roughly 30% of students admit to AI-assisted cheating.
• Reported in depth by Ars Technica; corroborated by El País and Inside Higher Ed.

The background: take-home exams were always a bargain built on trust and effort — you could look things up, but doing so well still took understanding and time. Generative AI broke that bargain. A model can now produce a polished, correct-looking answer to a typical economics problem set in seconds, and it does so in a way that is genuinely hard to detect after the fact. When a whole class quietly gains that capability, the take-home average doesn't measure economics anymore; it measures who used the tool.

What the score drop reveals: the gap between the take-home and in-person averages is, in effect, a rough measurement of how much AI was doing the work. The professor's line — captured in the story's title, that "we cannot choose to become idiots" — frames the stakes as more than grades. If students outsource the struggle that produces learning, the credential survives but the competence it is supposed to certify does not.

Why it matters: this is a cultural flashpoint with real institutional consequences. Universities are being pushed back toward proctored exams, oral defenses, and in-class writing — a partial retreat from decades of flexible assessment. It connects to a wider anxiety about AI eroding the honest signal in exams, from smart glasses defeating proctoring to the broader question of what testing means when the tool is undetectable. The through-line across the show's coverage — is any of this real — lands squarely here.

The honest caveat: these figures are reported, not from a controlled study, and a single course is not a national dataset. Score drops can have other causes — an in-person exam is more time-pressured and stressful than a take-home, which alone lowers averages somewhat. And the Princeton self-report survey measures admitted behavior, which is its own kind of noisy. The direction is hard to dispute and widely corroborated; the exact magnitude at any one school should be read as illustrative, not precise.

Primary source, verified: read the paper →

Ground Truth. (2026, July 8). A Brown professor moved a final in person, and the near-perfect scores collapsed. Ground Truth. https://groundtruth.day/news/brown-in-person-final-collapses-take-home-scores.html

@misc{groundtruth:brown-in-person-final-collapses-take-home-scores,
  title  = {A Brown professor moved a final in person, and the near-perfect scores collapsed},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {jul},
  url    = {https://groundtruth.day/news/brown-in-person-final-collapses-take-home-scores.html}
}

Topics: education · cheating · academic-integrity · culture · chatgpt · society