Event Details

Kush Varshney, TJ Watson Research Center, IBM

Title: Precautionary Human and Machine Decision Making, and Humble Trust

Abstract: Many consequential allocation decisions that have traditionally been made by people are now being supplanted by automated technologies. Whether in employment, lending, criminal justice or other domains, we have observed numerous examples of unwanted bias in both cases. In this talk, we first present a model of human decision making with bounded rationality that predicts racial discrimination that is unfavourable to unprivileged groups when the decision maker is precautionary. Then we turn to machine decision making and point out that perhaps people do not trust machines because we believe that they do not trust us. We present an idea based on the practice of humble trust to improve the situation, which partly translates into a less precautionary machine decision maker.