Dimensions of Human Likeness in Conversational AI: Effects on Human Identity Threat and Dehumanization

Abstract

Recent advancements in conversational artificial intelligence (CAI) have ushered in a new era of human-AI coexistence, where AI systems exhibit humanlike cognition, emotions, and behaviors. While the anthropomorphism of CAI fosters trust and intimate relationships, it also raises concerns about threatening human identity and dehumanizing humans, highlighting the need to understand which human characteristics in AI influence these outcomes.

This study aims to (1) identify the dimensional structure of human likeness in CAI through factor analysis and (2) examine how the perceptions of these dimensions affect human identity threat and self- and other-directed dehumanization using Partial Least Squares Structural Equation Modeling (PLS-SEM).

A multi-step study was conducted. Firstly, an extensive literature review was performed to identify 35 key human characteristics relevant to CAI. Next, an online survey form was created, including an explanatory passage and a video of CAI-human interactions. English-speaking adults in the United States were recruited and valid data from 323 participants was analyzed for the study.

Factor Analysis with oblimin rotation identified four latent dimensions of CAI human likeness: Experience, Communication & Memory, Social Connection, and Agency. PLS-SEM was then used to explore their effects on human identity threat and dehumanization, with separate models for self- and other-directed dehumanization.

The results revealed that the Experience dimension significantly increases both perceived human identity threat and dehumanization, with an indirect effect mediated by the perceived threat. In contrast, the Agency dimension didn’t directly influence the perceived human identity threat but significantly reduced dehumanization, while marginally moderating the human identity threat induced by Experience. In addition, the Social Connection dimension decreased the perceived human identity threat and indirectly mitigated dehumanization. However, it amplified the threat when paired with high Experience. Lastly, younger individuals reported greater human identity threat and dehumanization. The threat also medicated the effect of age on dehumanization.

These findings suggest the complex psychological and societal implications of anthropomorphizing CAI, emphasizing the need for ethical CAI design and development. By systematically identifying key humanlike traits in CAI and assessing their impact on our human identity and dehumanization, the present study deepens the understanding of CAI anthropomorphism’s psychological and societal impacts. It provides critical insights for responsible CAI development and human-AI interaction design that balance anthropomorphic benefits with the preservation of human distinctiveness.