Latest / AI Ethics with Fexingo: Bias, Safety, and Responsible Artificial Intelligence
How AI Models Are Learning Racism From Historical Redlining Maps
Episode 96 of AI Ethics with Fexingo dives into a specific, troubling case: how modern AI models used in mortgage lending and insurance underwriting are absorbing discriminatory patterns from 1930s redlining maps. Lucas explains how the Home Owners' Loan Corporation graded neighborhoods by race, and how those grades now appear in training data for automated valuation models. Luna pushes back on the idea that we can simply 'de-bias' the data, pointing to a 2025 study from the Federal Reserve Bank of Philadelphia that found residual race proxies even in models trained without race labels. The…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.