AI Aversion or Appreciation? Meta-Analytic Evidence for a Capability-Personalization Framework

摘要

Artificial intelligence (AI) is transforming human life. While some studies find that people prefer humans over AI (AI aversion), others find the opposite (AI appreciation). To reconcile these conflicting findings, our Capability-Personalization Framework posits that when deciding between AI and humans in a decision context, individuals focus on two dimensions: (a) perceived capability of AI and (b) perceived necessity for personalization. We propose that individuals appreciate AI when (a) AI is perceived as more capable than humans and (b) personalization is unnecessary in a given decision context; otherwise, AI aversion occurs. Our Capability-Personalization Framework is supported by a meta-analysis of 442 effect sizes from 163 studies (N = 82,078): AI appreciation occurs (d = 0.27, 95% CI = [0.17, 0.37]) when AI is perceived as more capable than humans and personalization is unnecessary in a given decision context, otherwise AI aversion occurs (d = -0.50, 95% CI = [-0.63, -0.37]). These effects are moderated by AI embodiment (robot vs. algorithm), outcome nature (behavior vs. attitude), and country-level unemployment rate and college degree percentage. Overall, our integrative framework and meta-analysis advance knowledge about AI-human preferences and provide important implications for AI developers and users.