A Slow Algorithm Improves Users’ Assessments of the Algorithm’s Accuracy

We consider the possibility that traditional UX metrics such as response time and usability may be less appropriate for some modern contexts in which algorithms are deployed. Response time, in particular, should be low if you are optimizing for user satisfaction for a word processing app. However, when the algorithm in question is helping a human make an important decision, such as whether someone should be considered for parole, perhaps slowing down the response time is worthwhile. We find that slowing response time can indeed help users make better assessments of algorithm accuracy.