MIT's own AI Education Committee has put a name to the worry drifting through lecture halls: cognitive surrender, the state students reach when they hand a chatbot the thinking before the thinking has started. The committee's report warns the habit may be eroding exactly the critical faculties a degree is meant to build, and it arrives as American universities split into two camps over what to do about it. The timing is not quiet, either. AI in classrooms has been one of this week's loudest education arguments, and here is an institution that helped build the technology weighing in with a warning rather than a brochure.
Watch Desk analysis
What happened
The report came out in August. The New York Times wrote it up on 15 September, and DongA Science's English-language account followed on 17 September, carrying the detail that matters.
Its central concern is cultural as much as technical. Students, the committee found, are turning to AI from the very first moment of difficulty, skipping the office-hour queue and the group study session where confusion used to get worked through. The skill being lost is not any particular answer. It is the habit of sitting with a problem.
The evidence gathered in the report is pointed. A June study of 26,000 middle and high school students in China found homework scores rose 18% after AI was introduced, while exam scores, sat without devices, fell 20%. MIT's own work from June last year found people who drafted essays with a chatbot remembered their own writing poorly minutes later, with researchers concluding the brain's memory networks had barely been recruited.
To its credit, the report refuses to be a ban manifesto. It concludes AI has substantial potential to improve university administration and learning, and the campuses are behaving accordingly. Dartmouth, Harvard and Ohio State are among the institutions whose presidents treat AI fluency as an essential graduate competency. The University of Chicago has banned it from certain required social science courses. UC Berkeley's law school bars it from ideation, outlining, drafting, revising and editing assignments. Harvard, refusing a central rule, is weighing more in-person exams and assignments a machine cannot simply finish.
Oregon State's Anita Sarma supplies the pragmatism worth keeping: students are going to use AI anyway, she argues, so what matters is teaching them to use it responsibly.
Why it matters
The Chinese numbers are the whole dispute compressed into two figures. An 18% homework rise paired with a 20% exam fall measures assistance beautifully and learning not at all, and every university on that list is really choosing which of the two numbers it is optimising for.
There is a knock-on consequence too. If some graduates spend three years building judgement the slow way while others spend them orchestrating chatbots, employers and doctoral programmes inherit the difference without knowing which transcript belongs to which.
Our read
Sarma's line is the honest centre of this. Prohibition is a losing bet against a tool that lives in every pocket, but delegation without literacy is cognitive surrender wearing a productivity badge. The universities quietly redesigning assessment, the in-person exams and the machine-proof assignments, are the ones acting on the evidence rather than issuing statements about it.
If you teach, grade the process a machine cannot fake. If you study, do the struggle first and let the model check it afterwards. The struggle is not the tax you pay on learning. It is the learning.
What to watch
- Whether the full committee report and any follow-up work reach a wider audience beyond the newspaper write-ups.
- Whether Harvard's drift towards in-person exams becomes a stampede, given its signalling power over other institutions.
- Whether larger studies replicate the Chinese homework-exam split, the figure this argument will be fought over.
- Whether campus rules converge on AI literacy requirements or keep fragmenting course by course.
Discussion spark: If AI lifts homework scores by 18% while exam scores fall 20% the moment devices are removed, should universities grade what students can do unaided, or accept that assisted performance is now the skill that counts?
Sources and evidence
- MIT report warns AI triggers cognitive surrender among US college students – DongA Science (17 September 2026, 06:43 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.