A new pair of AI benchmarks could help developers reduce bias in AI models, potentially making them fairer and less likely to ...