Anthropomorphic cueing, social evaluation, and threat in stated willingness to collaborate with AI

Abstract

AI systems are increasingly presented as team members, but humanlike cues may not uniformly increase willingness to collaborate. We examined how assigned cue packages are related to stated collaboration intention, social evaluations, and threat. Across six two-condition studies (N = 1,128), we examined textual cues, imagination instructions, and lower- versus higher-ABOT image-set assignments in imagined collaboration contexts. We distinguished assigned cues from subjective anthropomorphism and measured intention before warmth and competence. Higher-cue conditions increased intention in Studies 1b, 2, 3a, and 3b (d = 0.36–0.68), but not in Studies 1a or 4. Bootstrap 95% CIs excluded zero for positive warmth products in five studies, positive competence products in Studies 2 and 3a, and a negative competence product in Study 4. As evaluations followed the outcome, these products are descriptive decompositions rather than causal mediation effects. The exploratory Study 3 comparison could not isolate threat because article, study form, and sample were confounded. The intention interaction was not detectable. In Study 4, the higher-ABOT image set yielded higher subjective anthropomorphism and lower competence, with no detectable overall intention difference. The image-set coefficient became more negative as post-outcome uniqueness threat increased (b = −0.396, p = 0.008). This interaction remained detectable in an image-fixed-effects sensitivity model conditional on 12 recorded image codes. Assigned cue packages yielded distinct within-study patterns of social evaluation and intention, but did not establish a general benefit of anthropomorphism, a causal threat process, or changes in collaborative behavior.