Safety Perceptions of Generative AI Conversational Agents: Uncovering Perceptual...
summary
Survey of n=123 participants uses exploratory factor analysis to create multi-dimensional measures: trust splits into benevolence, low personal risk, reliability, fairness into decision-making integrity and politeness, and risk into discrimination, misinformation/unhelpful outputs, offensive language. Latent class analysis identifies three classes: Hesitant Skeptics, Cautious Trusters, Confident Adopters with usage frequency strongly predicting class membership: more frequent users show higher trust and fairness and lower perceived risk, while sexual orientation, income, and ethnicity also significantly correlate with group assignment.
AI-written summary · about
chapters (10)
- 0:00Safety perceptions study intro
- 1:13Common uses of GenAI chatbots
- 2:17Defining AI safety research
- 3:42Limitations of risk measurement
- 4:43Multi-dimensional measurement scales
- 5:44Trust fairness risk subdimensions
- 7:08Clustering perceptual user groups
- 8:10Comparing user group perceptions
- 9:13Demographic differences in perceptions
- 10:35Summary of findings
related talks
07mCybersecurity Analysts' Perception of AI Security Tools and Practical Implications
09mBuilding Trust in Conversational AI
26mAI Agents and the Future of Human-AI Collaborative Intelligence
26mBuilding ethical AI: Transparency, bias mitigation, and trust
44mToxic Trust: The Psychology of Automation Bias in Generative AI
28mEthical AI?!
45mAI, Corporate Responsibility & Democratic Legitimacy: Extended Q&A
34mTrust in the Age of AI Agents | Building Trustworthy AI