Nutrition Facts, Drug Facts, and Model Facts: Putting AI Ethics into Practice in Gun Violence Research

Transparency labels can make AI systems more trustworthy: The paper proposes "Model Facts"—similar to nutrition labels on food—to help non-experts understand how AI models work and what biases they might contain, particularly for sensitive applications like predicting gun violence and suicide risk.

AI used in health research on vulnerable communities needs special accountability: Because firearm injury research relies on data from Black and Brown Americans who have historically experienced exploitation, the authors argue that clear communication about model accuracy and demographic biases is essential to rebuild trust with affected communities.

Complex AI information can be simplified for public understanding: The Model Facts template breaks down technical accuracy metrics and demographic performance into "standardized and minimally complex values," demonstrating that general users can assess AI validity without requiring deep technical knowledge.

Current AI transparency efforts fall short for real-world users: The paper identifies a gap where existing model documentation is too technical for public health practitioners and community members who actually use these tools, suggesting that better communication frameworks are needed beyond academic papers.

Implementation requires ongoing refinement and user testing: The authors acknowledge that Model Facts labels will need educational resources and human-computer interaction experiments to ensure they actually help people make better decisions about whether to trust and use AI systems in practice.