Calibrating Users’ Mental Models for Delegation to AI

Artificial intelligence (AI) has the potential to dramatically change the way decisions are made and organizations are managed. As of today, AI is mostly applied as a collaboration partner for humans, amongst others through delegation of tasks. However, it remains to be explored how AI should be optimally designed to enable effective human-AI collaboration through delegation. We analyze influences on human delegation behavior towards AI by studying whether increasing users\u27 knowledge of AI\u27s error boundaries leads to improved delegation behavior and trust in AI. Specifically, we analyze the effect of showing AI\u27s certainty score and outcome feedback alone and in combination using a 2x2 between-subject experiment with 560 subjects. We find that providing both pieces of information can have a positive effect on collaborative performance, delegation behavior, and users\u27 trust in AI. Our findings contribute to the design of AI for collaborative settings and motivate research on factors promoting delegation to AI