Artificial Intelligence (AI) Can't Feel Me: Lower Perceived AI Perspective–Taking Impedes Human–AI Cooperation
TL;DR
It is found that individuals are generally reluctant to cooperate with AI possessing partial decision‐making authority, and this reluctance is shaped by perceived AI perspective‐taking, and it is suggested that psychologically anthropomorphic design may foster more effective human–AI interaction.
Abstract
As artificial intelligence (AI) becomes increasingly autonomous, its interactions with humans increasingly resemble social relationships, giving rise to a novel form of cooperation—human–AI teaming. Drawing on social cognition theory, the present research examines how individuals’ perceived AI perspective–taking shapes their willingness to cooperate. Across five studies, we find that individuals are generally reluctant to cooperate with AI possessing partial decision‐making authority (Studies 1 and 2), and this reluctance is shaped by perceived AI perspective‐taking (Studies 3 and 4). When AI agents appear to accurately—or even slightly overestimate—the thoughts and feelings of individuals experiencing unfair situations, perceived AI perspective–taking increases, thereby enhancing their willingness to cooperate with AI (Study 5). These findings extend perspective–taking research from human–human to human–AI contexts, highlighting that cooperation with AI depends not only on instrumental outcomes but also on social–cognitive inferences about being understood, and suggesting that psychologically anthropomorphic design may foster more effective human–AI interaction.