Study: Making AI 'human' backfires for experts, helps novices trust it more

A new study of 1,256 users reveals that making AI assistants appear human (anthropomorphism) significantly alters risk perception, but the effect depends on user expertise. For users with low domain knowledge, a human-like design increases trust and reduces perceived risk. However, for experts, the same design can have the opposite effect, increasing risk perception. This creates a critical calibration challenge for designers aiming to build responsible, trustworthy decision-support systems.