Researchers at the MIT Media Lab asked a question every teacher has been asking since late 2022: what actually happens in a student's head when an AI does the drafting? Using EEG to monitor brain activity while participants wrote essays — some unaided, some with a search engine, some with a large language model — they found a clear gradient.
The findings
- Writers using the LLM showed the weakest neural connectivity of the three groups across networks associated with attention and memory.
- They struggled to quote or even recognize sentences from "their own" essays minutes after submitting them.
- Most concerning: when LLM users were later asked to write unaided, their engagement remained depressed — the researchers called the lingering effect cognitive debt.
The work of forming a thought is not overhead to be optimized away. It is the learning.
What this means for assignment design
The instinctive response — detection tools and bans — misreads the problem. Students didn't stop thinking because they are lazy; the assignments allowed the thinking to be outsourced. The design answer is tasks in which the reasoning has to happen in the room: positions that must adapt to new evidence mid-unit, work defended orally, drafts assessed as seriously as final products.
That is a design problem — and design problems can be solved.
