Abstract: The rapid development of artificial intelligence (AI) has shifted human-AI interactions from instrumental to collaborative partnerships, yet public distrust and resistance remain major obstacles. Meanwhile, anthropomorphism (attributing human characteristics to non-human entities) is widely used to enhance human-AI collaboration, but existing findings are contradictory: anthropomorphism may trigger negative reactions. This study integrated the Stereotype Content Model and Intergroup Threat Theory to explore how and when anthropomorphism influences collaboration intentions.
This research comprised four sequential studies (incorporating six sub-experiments; N = 1128), recruiting working adults from the credamo platform to systematically investigate the impact, mediating mechanisms, and boundary conditions of AI anthropomorphism on human-AI collaboration intention. Diverse anthropomorphism manipulation methods were employed: Studies 1a and 1b used passive textual descriptions (anthropomorphic vs. non-anthropomorphic AI); Study 2 utilized an active guided imagination task (high vs. low anthropomorphism); Studies 3a and 3b introduced realistic threat scenarios (high vs. low) based on the imagination manipulation; Study 4 employed ecologically valid robotic images (high vs. low anthropomorphism) and measured the moderating role of uniqueness threat.
Results from Studies 1a and 1b indicated that passive textual anthropomorphism successfully enhanced perceived warmth, which in turn mediated collaboration intention, though the direct effect on intention was non-significant. Study 2, employing active imagination, strengthened the manipulation effect. It demonstrated that high anthropomorphism significantly boosted both perceived warmth and competence, which in turn fully mediated its positive effect on collaboration intention. Studies 3a and 3b replicated the findings of Study 2 under both high and low realistic threat conditions, revealing that realistic threat was not an effective boundary condition; the positive effect of anthropomorphism and its social cognitive mediating pathway remained stable regardless of realistic threat level. Study 4 revealed that objective visual anthropomorphism (based on ABOT database ratings) not only failed to promote collaboration but significantly reduced intention under high uniqueness threat. However, when using subjective perceived anthropomorphism as the indicator, the mediating pathway through perceived warmth and competence was confirmed again.
This study represents the first systematic investigation into social cognitive mechanisms underlying how anthropomorphism influences human-AI collaboration. By extending the applicability of the Stereotype Content Model and Intergroup Threat Theory to human-AI interaction contexts, the study provides an integrated theoretical framework for understanding the complex operational mechanisms and boundary conditions of AI anthropomorphism in human-AI collaboration—specifically within the context of in-group and out-group dynamics—and further deepens reflections on the question of how and when to use anthropomorphism in the context of human-AI collaboration. Furthermore, drawing on the perceptual pathways of warmth and competence, as well as the boundary defined by uniqueness threat, the study offers practical implications for optimizing the anthropomorphic design of AI and facilitating human-AI collaboration.

Key words: anthropomorphism, human-AI collaboration, perceived warmth-competence, uniqueness threat, social cognition