Rethinking AI Safety: Provocations from the History of Community-based Practices of Road and Driver Safety

As AI safety becomes a central research and policy focus for organizations interested in addressing the risks of advanced AI systems, historical analogies to road and driver safety have figured as a means to illustrate the importance of certain safety measures such as safety-by design and regulatory interventions. While examples like road regulations, seatbelt design, and automobile laws have been the dominant conceptualization of safety in these efforts, our paper draws on alternative case studies to highlight the value of a different paradigm: that of community-based safety. We present three historical vignettes—on the Pan-American Highway system, the Green Book guide for Black motorists, and the patron saint of transportation workers La Virgen del Carmen—to illustrate alternative, on-the-ground, and bottom-up ways of knowing and practicing safety. Bringing these cases to bear on current AI safety efforts, we forward the provocation that, more than a measurable, objective fact, "safety" is a felt and lived practice that relates not only to system features, but also to social, cultural, economic, and political conditions. We argue that those living in the midst of sociotechnical unsafety hold the knowledge needed to envision appropriate community-based safety tools and practices whenever system-based interventions turn out to be insufficient. These considerations raise important questions about how we conceptualize agency and expertise in emerging research on AI safety.