I can't help but worry that our AI death robots are going to be hamstrung against chinese AI death robots because ours won't take prior probabilities into account.
That would be a terrible implementation. The bias reduction is about answering "is the Jewish or black man guilty" without more context. It should not affect "tell me about crime rates grouped by race in (region) and (period)".
I don’t understand you. What do you mean by this?
It's a little dogwhistle implying prejudice is good, actually.