Cognitive Debt
Your Brain on AI: What EEG Shows About
A groundbreaking study by MIT researchers (Kosmyna et al., 2025) used EEG—and a digital obstacle course—to uncover the hidden cost of relying on LLMs versus traditional search or writing without tools.
The Experimental Setup
-
Participants (n = 54) were split into:
-
Brain‑only: wrote essays with no assistance.
-
Search engine: used Google.
-
LLM: used ChatGPT or similar.
-
-
After three sessions, groups swapped tools to observe learning and rebound effects in a fourth session
Key Findings from the Study
-
Neural Connectivity Declines with AI Use
-
Brain‑only writers showed the strongest, most widespread EEG connectivity across alpha, beta, and delta bands.
-
Search engine users displayed moderate connectivity.
-
LLM-assisted writers had the weakest brain connectivity, indicating lower cognitive engagement
-
-
Switched Tool, Switched Brain
-
Participants moving from Brain‑only to LLM (Brain→LLM) showed increased visual and executive functions, similar to search engine users.
-
LLM-to-Brain participants, however, failed to regain strong connectivity, showing ongoing cognitive dampening
-
-
Lower Ownership & Recall
-
LLM users reported lowest sense of authorship, struggled to quote their own content.
-
Brain-only users showed full ownership and scored highest in essay coherence and originality
-
-
Cognitive Debt Accumulates
-
Over 4 months, LLM users underperformed in writing quality, neural effort, and memory compared to other groups.
-
The study introduces “cognitive debt”—the gap between convenience and sustained mental engagement
Why It Matters
-
Tools reduce brain discipline: Like calculators teaching fewer math basics, LLMs reduce deep cognitive exploration.
-
Effort fuels learning: True skill—writing, creating, reasoning—is nurtured through active struggle.
-
Mindful adoption is vital: There’s value in AI—but only when paired with deep cognitive work, learning mechanisms, and periodic unplugging.
Practical Takeaways
-
Limit AI use for drafting only—do heavy editing, recall work manually.
-
Alternate brain states—rotate between “tool-on” and “tool-off” sessions.
-
Track cognitive engagement—use shared reflection, memory tests, or short EEG-like metrics where possible.
-
