How AI Framing and Level of AI Involvement Influence Trust

Abstract: As artificial intelligence (AI) is increasingly integrated into consequential decision-making in fields like medicine and education, organizations face a critical dilemma concerning how to disclose the division of labor between human experts and AI systems. While highlighting the role of advanced technology signals innovation, it risks overshadowing the human oversight that is critical for trust. Across four preregistered studies (N = 6,371), we investigate how the framing of this division of labor shapes decision recipients’ trust. Participants evaluated collaborative processes where the contribution was described either by the percentage of work done by the AI or by the human across varying degress of AI involvement. We consistently find that the effect of framing is non-monotonic. At low levels of AI involvement, framing the work in terms of the human’s contribution (human framing) builds more trust than highlighting the AI’s minor role. Conversely, at high levels of AI involvement, emphasizing the AI’s contribution (AI framing) can increase trust relative to human framing (Studies 1-4). We show that this shift is driven by perceived competence, which favors AI framing as involvement increases, whereas perceived warmth favors human framing until high levels of AI involvement (Studies 2 and 3). Furthermore, we find that the benefits of AI framing are amplified when the task is perceived, even if illusorily, as well-suited for AI. This research contributes to the literature on human-AI collaboration by demonstrating that trust is not just a function of how much AI is involved, but also of how that involvement is described. Our findings suggest that organizations can foster trust not simply by hiding or revealing AI, but by matching the disclosure frame to the dominant contributor.

Keywords: Trust; Artificial Intelligence (AI); Algorithm Aversion; Framing