Can similarity-based customization of advice increase cooperation with algorithmic decision aids?

Abstract: Interpersonal research highlights similarity's role in conflict resolution and advice-taking. We apply this to algorithm aversion, where users, particularly experts, experience identity threat by AI decision aids. Three studies (N=820) investigated Perceived Outcome Similarity (POS). Findings reveal a discrepancy: similarity-focused AI advice is rated as more accurate and helpful than accuracy-focused advice, but users do not follow it more. We propose that similarity might reinforces users' confidence in their opinions, providing affirmation rather than being a catalyst for change.

Keywords: Algorithm Aversion; Advice Taking; Human-AI Interaction; Laboratory Experiments