Algorithm Aversion

Algorithm aversion is the tendency for people to distrust and avoid using algorithms for decision-making, especially after observing them make mistakes.

The phenomenon appears across high-stakes and low-stakes domains: medical recommendations, financial advice, employee selection, even joke recommendations. After seeing an algorithm err, people often prefer human judgment even when the algorithm demonstrably outperforms humans on aggregate.

Algorithm aversion is not irrational exactly, it reflects people’s difficulty in calibrating appropriate trust. Seeing an error is salient; understanding base rates of error is not. The design challenge is helping users develop appropriate rather than blanket distrust.

Related: 01-atom—calibrated-trust-vs-high-trust, 05-atom—automation-bias-regulatory-recognition, 05-molecule—self-assessing-ai-pattern