Disuse and Misuse of AI-Generated Suggestions: Examining Human–AI Conflicts in AI-Assisted Decision-Making
研究了AI辅助决策中，决策者因先验信念和任务知识导致弃用或误用AI建议的人机冲突，通过贝叶斯模型和实验揭示其成因。
The increased integration of artificial intelligence (AI) technologies in human workflows has brought about a new paradigm of AI-assisted decision-making, in which human decision-makers (DMs) aspire to gain enhanced insights by reviewing AI-generated suggestions. However, achieving effective synergies in human–AI teams often proves challenging due to the misuse or disuse of AI advice by DMs. This article investigates how DMs’ prior beliefs and task knowledge employed in decision-making processes may affect such inappropriate behaviors, referred to as human–AI conflicts. We model the cognitive process of AI-assisted human decision-making within a Bayesian hypothesis testing framework, where the DM evaluates information using either their own knowledge or AI-generated suggestions to make expectedly optimal choices. We analytically characterize the impact of AI input on the human decision process and identify circumstances that contribute to human–AI conflicts. We then perform a scenario-based experiment to provide empirical support for our theoretical findings. We show that skewed prior beliefs about the state of the world promote the disuse of AI advice, while difficulties that humans have in explaining observed data lead to the misuse of AI advice. These findings provide insights into the interplay of prior beliefs, task knowledge, and AI advice in human–AI teams. Our research can help practitioners be aware of task environments in which human–AI teams perform poorly and better understand the optimum modes of use under different conditions so that organizations can derive value from the deployment of AI systems.