Algorithm Aversion
Definition
Algorithm Aversion is the phenomenon where individuals lose trust in algorithmic decision-makers faster than human decision-makers after seeing them make a mistake, even when the algorithm performs better overall.
Real-world examples
A forecaster chooses to manually adjust or entirely discard statistical models after seeing them predict a minor stock drift incorrectly, reverting to less accurate gut-based decisions.
- A clinician may discount a diagnostic model after one visible error, even when it outperforms them on average.
- A driver overrides a navigation app that once routed them poorly, trusting their own memory instead.
How to design for it (nudge strategy)
Restore trust by giving users small adjustments or 'control overrides' (e.g., allowing them to adjust the algorithm's recommendation by +/- 5%), which mitigates aversion and boosts adoption.
The evidence (3)
- Radiology Diagnostic Integration and Human Underweighting
Healthcare · Diagnostic Classification Accuracy: 82.5% → 79.8% (+-2.7 pts) · n = 200 radiologists, 10,000 cases
- Pre-Trial Release Decisions and Machine Crime Forecasting
Public Policy · Rearrest Rate of Released Defendants: 18.2% → 13.7% (+-4.5 pts) · n = 750,000 cases in New York City
- Mitigating Algorithm Aversion Through Output Control
Finance · Adoption Rate of the Algorithmic Forecast: 15% → 88% (+73.0 pts) · n = 3 experimental studies, 1,200 trials
Key research
- Algorithm Aversion: People Erroneously Avoid Algorithms After Seeing Them ErrBerkeley J. Dietvorst, Joseph P. Simmons, Cade Massey · Journal of Experimental Psychology: General (2015)
Related biases
Cite this page
Behavioral Economics Lab. "Algorithm Aversion – Definition, Examples & Evidence." Behavioral Economics Lab, https://behavioraleconomicslab.com/biases/algorithm-aversion.