Safety Perceptions of Generative AI Conversational Agents - Uncovering Perceptual Differences in Trust, Risk, and Fairness

Explore research findings on how users perceive the safety of generative AI conversational agents through this 12-minute conference presentation from SOUPS 2025. Discover how researchers from The George Washington University, University of Maryland, and Tel Aviv University conducted an exploratory study using factor analysis and latent class analysis on survey responses from 123 U.S. participants to measure and understand the dimensionality of safety perceptions regarding AI chatbots. Learn about the three distinct user groups identified through the research: The Hesitant Skeptics, The Cautious Trusters, and The Confident Adopters, each displaying unique and sometimes counterintuitive perception patterns regarding trust, risk, and fairness in AI systems. Understand how usage frequency of AI chatbots correlates with higher trust and fairness perceptions but lower perceived risk, and examine how demographic factors including sexual orientation, income, and ethnicity significantly influence group membership and safety perceptions. Gain insights into the complex relationship between user experiences, interactions, and the development of safety perceptions in generative AI conversational agents, highlighting the need for more nuanced approaches to measuring and understanding public attitudes toward AI safety in the context of trust, fairness, and risk assessment.