AI is getting too good at doing homework, MIT wants colleges to rethink exams

We already know about the student-AI relationship when it comes to homework. It is no longer a headache for students, with AI doing much of the work for them. Across the world, educational institutions and universities are baffled trying to address this crisis. And now, one of the world’s most prestigious universities – MIT – is raising serious alarms about what happens when machines do the thinking for the students. In a blog post, it states that generative AI can now "credibly complete most undergraduate assignments."

MIT President Sally Kornbluth addresses this issue to students on her campus in a candid letter. There she shared the final report from the Institute's Ad Hoc Committee on AI Use, calling this a historic "watershed moment."

MIT leadership acknowledged that while generative AI feels like an academic superpower, letting algorithms take over the heavy lifting introduces disturbing risks to genuine education. When chatbots solve the equations, draft the essays, and debug the code, the traditional college homework model collapses under its own weight.

The hidden toll on the student brain

Who doesn’t like shortcuts? And as for students, the majority of them have embraced automated help very eagerly. Surveys from the College Board revealed that over 80 per cent of high schoolers now use generative AI tools for schoolwork at least occasionally, with usage running the gamut from brainstorming and outlining to straight copy-pasting.

A comprehensive study by the RAND Corporation and the Center for Reinventing Public Education backed this up, finding that homework use among middle and high school students has climbed past 60 per cent.

That gut feeling is supported by hard scientific evidence. In a massive field experiment titled "Generative AI Can Harm Learning," researchers from Wharton and the University of Pennsylvania tracked thousands of high school students tackling math problems with and without AI tutors. The results were striking. While students with AI access completed 48 per cent more practice problems correctly, they scored 17 per cent worse on subsequent exams when the bots were taken away. The technology had essentially become an academic crutch, allowing students to skip the mental friction that makes concepts stick.

The neurological cost appears even steeper. In a study from the MIT Media Lab, "Your Brain on ChatGPT & Cognitive Debt," researchers used electroencephalography (EEG) to monitor 32 neural regions during writing exercises. Students crafting essays with ChatGPT exhibited significantly lower cognitive engagement and reduced neural connectivity compared to those writing unassisted. The authors warned that this trend fosters a dangerous build-up of "cognitive debt,"a slow, invisible atrophy of long-term memory formation and independent synthesis skills.

How MIT plans to reclaim real learning

Instead of waging an unwinnable war on AI with buggy AI detection software, MIT is overhauling the way education happens on campus. The vision as President Kornbluth put it, "to define and foster the highest-quality residential education – of humans, by humans, in support of human flourishing, and for the betterment of humankind."

That strategy means dismantling vulnerable routines and putting the value back on the human experience. MIT is moving away from traditional, easily automated problem sets and take-home exams in favour of oral defences, semester-long project portfolios, and hands-on collaborative engineering challenges that demand authentic reasoning.

Every professor is being asked to provide clear, justified policies on how and why AI may be used for a specific assignment, tying rules directly to learning goals rather than blanket bans. Most importantly, the Institute is actively investing in vibrant collaborative spaces and scheduling dedicated "Tech Free Times" to revive the late-night dorm study groups, lively whiteboard arguments, and peer-to-peer mentoring that made an MIT education transformative long before algorithms entered the classroom.