Events

Algorithm aversion captures the idea that people may avoid algorithms even when the algorithm generally performs better than their own judgment. Dietvorst et al. (2015) argue and find experimental evidence that a key cause of algorithm aversion is algorithm feedback asymmetry: people lose confidence in algorithms after seeing it err more quickly than they lose confidence in themselves after seeing themselves err. We interpret and test this hypothesis using field data from 221,000 insulin dosing decisions and find the opposite – patients decrease algorithm use less after an algorithm’s mistake than they increase it after their own mistake. A controlled laboratory experiment reconciles this contradiction, showing two distinct forms of algorithm feedback asymmetry: one favoring humans in one-shot decisions (as in Dietvorst et al. 2015) and another favoring algorithms in repeated settings (as in our field data). Our findings suggest that performance feedback does not fundamentally drive algorithm aversion and may, in repeated interactions, increase algorithm adoption.