Task Delegation from AI to Humans: The Impact of AI Transparency on Human Performance

This study examines the effects of AI transparency in AI-human task delegation. According to the principal-agent theory (PAT), increasing transparency can help address the problem of hidden action. Through a between-subjects experiment with three conditions (AI advantage information, AI function information, vs. no information), we explore the impact of AI transparency on human task performance and consider the mediating effect of human epistemic uncertainty and AI trust. Results show that AI function information enhances accuracy, while AI advantage information encourages more task completion. Epistemic uncertainty fully mediates the relationship between transparency and task performance (both accuracy and image number), while AI trust only mediates the effect on image number. The results have implications for the future work of the design of AI transparency in human-AI collaboration.