🧠When GPT writes your notes, what’s your brain doing?
A new MIT Media Lab experiment wired 54 volunteers to EEG caps while they wrote multiple SAT‑style essays under three conditions: no tools, Google Search, or ChatGPT. Those relying on ChatGPT showed the weakest pre‑frontal and parietal connectivity—the neural circuits tied to working memory and executive control. Even after the researchers took the AI away, those brains stayed under‑engaged, a phenomenon the authors dub “cognitive debt.” MIT Media Lab
Quality and ownership dipped, too. Teachers rated the ChatGPT essays the least original, and participants later struggled to recall or quote what they had “written,” confirming that speed came at the cost of synthesis and memory. TIME
Why this matters in clinic
As an orthopedic trauma surgeon I still let an ambient scribe draft every routine fracture‑clinic note—but many in other fields like internal medicine and less procedural specialties have told me that they let generative AI only draft the clerical parts: meds, template language, today’s HPI. The assessment & plan they write line by line. That split keeps paperwork out of a doctor’s head during the visit yet forces them to wrestle with the medical reasoning afterward.
Think of AI as a middle pass:
* Dictate a scaffold. Voice out the structure in your own words.
* Let the model fill. Off‑load rote phrasing and formatting.
* Interrogate the draft. Edit until every decision reflects your judgment.
If composing the note is how you think , keep writing. But if note‑taking just burns cognitive calories you’d rather spend on the patient, teaching, or research, outsource it—then reinvest that bandwidth where it counts.
Open question
Would the same neural slump appear during real‑time clinical reasoning with ambient scribes running? Until someone reruns the study in exam rooms, be intentional: deploy AI where it frees you, guard the zones where insight lives—and don’t let your hard‑won clinical muscle atrophy for a few extra clicks.
Link to full study here https://arxiv.org/abs/2506.08872
