New AI benchmarks could help developers reduce bias in AI models, potentially making them fairer and less likely to cause harm. The research, from a team based at Stanford, was posted to the arXiv ...