Latest / AI Ethics with Fexingo: Bias, Safety, and Responsible Artificial Intelligence
How AI Models Are Learning Gender Bias From Performance Reviews
Episode 102 of AI Ethics with Fexingo examines how performance review language teaches AI to perpetuate gender stereotypes. Lucas and Luna dig into a 2025 study by Stanford and Anthropic that analyzed 1.5 million performance reviews from Fortune 500 companies, finding that words like 'aggressive' and 'confident' were disproportionately applied to men, while 'helpful' and 'collaborative' skewed female. The hosts trace how these biases flow into AI models trained on corporate HR data, then into resume screeners, internal promotion tools, and even chatbot personas. They discuss the feedback…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.