“Subjects who used AI assistance over a six-month period showed measurable declines in unaided performance on the same tasks.”
The MIT study is the first long-horizon experiment on what AI assistance does to human capability over time. The result is the one anyone who has used the tools for a year already suspected. The skill atrophies. The model fills in. When the model is unavailable, the user is worse than they were before they started using it. The cognitive cost is not free. It just shows up later.