Agency Decay and the Risk of AI's Asymptomatic Harms |
Artificial intelligence (AI) does not take away human agency. And agency does not vanish in one dramatic moment. More often, it dissolves—gradually, politely, and with measurable gains along the way. That’s what makes the newest wave of AI risk harder to govern: Many harms arrive asymptomatically. Performance metrics go up. Output accelerates. Errors may drop. Yet the worker’s capacity to think, judge, and recover from stress quietly thins out.
A useful example comes from a year-long field study in radiation oncology. Researchers tracked 42 specialists across interviews, workshops, and “think-aloud” sessions and observed a paradox: AI improved efficiency while eroding underlying expertise and agency—moving from subtle “intuition rust” to chronic skill degradation and, eventually, fear of being “hollowed out.”
These findings transcend the disciplinary field of medicine. It is the pattern: AI can raise the floor of performance while lowering the ceiling of mastery—especially when systems become seamless enough to remove the very friction that used to keep humans engaged or mentally “in the loop.”
Traditional AI governance is built around visible failures: biased outputs, safety........