Bridging the Human-AI preference gap: A persuasion-oriented tripartite framework

The remarkable performance of artificial intelligence (AI) systems on a wide variety of complex tasks has fueled its rapid adoption in recent years. Yet, many individuals still exhibit a human-AI preference gap, such that they prefer human-generated over AI-generated outputs and follow recommendations made by generative AI (GenAI) less than human advice. We conceptualize the rejection of a superior AI system in favor of an inferior human as a persuasion failure that necessitates psychological rather than technological resolution. Recent research has identified various interventions that compress the human-AI preference gap. We codify these interventions by introducing a persuasion-oriented tripartite framework that delineates three distinct paths by which the preference gap between humans and AI systems may be narrowed: (1) AI-centric interventions, which directly bolster perceptions and evaluations of AI systems; (2) human-centric interventions, which reduce the (often inflated) perceptions and evaluations of humans; and (3) hybrid interventions, which offer an integrated or reformulated alternative that combines human and AI elements. Ultimately, our tripartite framework illustrates how different persuasion tactics fundamentally reshape beliefs and attitudes about both AI and humans, thereby informing system design elements and communication strategies that can bridge the human-AI preference gap.