Why Users Seek Explanations Matters for Human-Centered AI

Summary

This position paper argues that human-centered explainable AI should incorporate the psychology of information seeking. Users assess explanations through instrumental, hedonic, and cognitive expected utilities, while cognitive biases shape whether those utilities appear valuable. The authors warn that both excessive and insufficient information-seeking can undermine decision quality. The challenge is particularly important for agentic AI, where explanations must help people anticipate cascading actions, assess risks, and decide when to intervene. The paper therefore advocates designing systems that make explanations appropriately sought, rather than merely making them available.