Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
AI Research
arXiv CS.AI·Aug 28·5 min read
Is Your Neighborhood Safe? Place-based Stigma in Large Language Models' Urban Safety Judgments
TL;DR
Large language models (LLMs) exhibit place-based stigma in urban safety judgments, with neighborhood names influencing safety ratings more than actual crime data. This bias is particularly pronounced in areas with higher populations of marginalized groups.
✦ Why It Matters
Evaluate the impact of neighborhood names on safety assessments in your AI applications today.
Key Takeaways
- Assess how neighborhood names influence AI safety ratings.
- Consider demographic factors when interpreting model outputs.
- Test models under different conditions to identify biases.
- Implement strategies to mitigate place-based stigma in AI.
How It Works
The study dissociates neighborhood names from geographic coordinates to analyze their individual impacts on safety ratings. By comparing models' responses under different conditions, researchers identified that names carry significant weight in shaping perceptions of safety, often reflecting underlying demographic stereotypes.