Vignette study on the mobilisation potential of different AI risk areas

This study tests which AI risks drive public concern and willingness to take action on AI safety. We conduct a randomised vignette experiment comparing 11 specific AI risk areas: job displacement, misinformation, disinformation, extinction risk, surveillance, autonomous weapons, social isolation, algorithmic bias, environmental harms, fraud, and cognitive decline. Participants read realistic vignettes depicting specific AI risks/harms (or no vignette in the control condition), then report concern, regulation support, willingness to act, etc. We measure both general AI attitudes and risk-specific responses, including possible spillover effects (does reading about one risk affect concern for others?). Bayesian regression models with pre-registered analyses test which risks are the most galvanising and through what psychological mechanisms (anxiety, anger, personal relevance, temporal relevance).