Biometrics Reveal AI Coding Assistants Shift Cognitive Engagement Patterns
EEG and eye-tracking data show developers offload mental effort differently with AI help.
Researchers from universities in Bari, Italy, and Copenhagen, Denmark, recruited undergraduate and graduate students for a within-subjects crossover experiment. Participants completed coding tasks with and without AI-code assistants while wearing EEG caps, eye-trackers, electrodermal sensors, and heart rate monitors. The team also collected performance scores using a rubric and self-reported workload via the NASA-TLX across six dimensions.
Key physiological findings: under AI assistance, the EEG theta/alpha ratio was lower during the first task, and gaze blink rate was higher during the second—both consistent with reduced cognitive engagement when developers offload generative work to the model. Interestingly, this pattern did not differ between undergraduate and graduate students. Electrodermal activity correlated with performance only in the non-AI condition, and among the six NASA-TLX dimensions, only physical demand was linked to performance—again only without AI. The authors conclude AI-assisted programming is not simply faster solo coding but a fundamentally different cognitive activity, with implications for designing AI assistants and using biometric monitoring in augmented development.
- EEG θ/α ratio decreased and gaze blink rate increased during AI-assisted tasks, signaling reduced cognitive engagement.
- Performance correlations with electrodermal activity and self-reported physical demand disappeared when AI was used.
- No significant difference was found between undergraduate and graduate students' physiological responses to AI assistance.
Why It Matters
For developers and tool makers: AI assistants don't just speed up coding—they fundamentally change how we think through problems.