MIT's research committee on AI use in teaching put a name to something many educators have observed but couldn't quite describe: cognitive surrender. Students reach for AI "at the first hint of struggle", and what they receive in return is the illusion of learning. The data makes the cost clear: homework scores rise, exam scores fall.
Surrender as a Decision, Not a Slip
MIT's five-month consultation with students and faculty arrived at a precise definition. Cognitive surrender happens when a learner reaches for an AI tool not because they've exhausted their own effort, but at the moment the work becomes uncomfortable—usually when a deadline looms. The answer on screen looks like understanding, the homework gets handed in on time, and the student won't discover that nothing has stuck until the exam arrives.
It's not forced. Nobody removes the thinking from a student's hands. It is surrendered, which means something different: it is a choice made in a moment of friction, when the immediate relief of an answer outweighs the delayed cost of not having learned.
I find it fascinating that most people I know who are not particularly keen on AI have three main concerns. The first, and interestingly the smallest group, fear loss of control and doom for humanity. The second covers most of my creative friends in the arts - the soullessness of AI, the outright theft of intellectual property and the generated slop. The third concern is educational, that human learning is greatly diminished through cognitive surrender or outsourcing or offloading, whatever you may call it. All are legitimate and real concerns and we would do well to deal with them with the seriousness and effort it requires because our humanity is at stake.
The Numbers Tell the Story
A study of 26,000 secondary-school students in China provides the hard evidence. After AI tools became available, homework scores rose by 18%. Exam scores fell by 20%. The work submitted got better. The students got worse. That gap—between what is turned in and what is understood—is not a minor glitch. It is the problem itself.
The disconnect matters for anyone building tools that measure learning: completion rates, scores on take-home tasks, and time spent are all easy to record. All of them can rise while actual understanding falls. A learning platform that celebrates higher homework scores may be celebrating a form of cognitive surrender, not progress.
Any metric that measures only what a student produces (not what they know) can be optimised by tools that produce on their behalf. Progress data needs at least some checks that the learner does alone: a quiz without AI, an exam, an unassisted problem, or an explanation in their own words tested in real time.
What Schools Can Actually Do
MIT's recommendations avoid drama in favour of practicality. They call for: a review of the curriculum to clarify which skills require unaided work, more teaching assistants (who actually interact with students), designated spaces where AI is not used, and communities where teachers share what is working.
Notably, the committee does not call for a ban. It does report that students are confused by AI policies that differ from one course to the next, which is a fair warning to any school writing policy for the first time.
Students will use AI regardless of what rules say. The sensible task, then, is to decide which parts of a course must be done unaided—and to say so plainly, consistently, and in advance.
A Question for Every Educator
If you teach, the question is concrete: which skills in your curriculum must a student demonstrate alone? Not which skills are old-fashioned or outdated, but which ones require unassisted struggle as part of learning them. That struggle is not waste—it is where learning lives.
Building Schools That Teach Thinking, Not Just Answer-Giving
Cognitive surrender is not inevitable. It is a system problem with system solutions. The first step is naming it clearly and deciding, at your school, which work stays unaided.