Right agent, wrong level of hedonism: How high (vs low) hedonic values in AI-performed tasks lead to decreased perceptions of humanlikeness, warmth, and less consumer support

Abstract

Reasons of reactions towards artificial intelligence (AI) performing hedonic tasks invites attention. This research argues that the reference of comparison to assess AI's suitability for the task shifts from machine schema to human schema as the hedonic value of the task increases, leading to the decrease in perceived humanlikeness, perceived warmth, and support for the AI, subsequently. Four studies, including a big data study on GitHub and three experiments, have shown that people are less supportive of AI and its creators when AI performs highly (vs low) hedonic tasks. This effect is mediated by the perceived humanlikeness of AI, since it affects AI's perceived warmth in a way that decreasing perceived humanlikeness when AI is assigned to the tasks with high (vs low) hedonic values decreases the perceived warmth of the AI. Decreased perceived warmth, in turn, lowers people's support for the AI and its creators. This serial mediation mechanism remains robust when AI's baseline humanlikeness increases, when the hedonic value of AI's method of performing the task, rather than the task itself is manipulated, and when the perceived competence and fairness are tested as alternative paths in parallel.