In most of these discussions we're using "bias" completely wrong. I frequently see processes that result in a disproportionate negative outcome for a particular

In most of these discussions we're using "bias" completely wrong. I frequently see processes that result in a disproportionate negative outcome for a particular group referred to as "biased". This is not what biased means, and very well-designed processes can (and should in many cases) result in disproportionate negative outcomes for one group over another. For example, if you build a psych model that predicted the likelihood of an individual becoming violent, and you used that model to decide who should and should not be allowed to own a firearm, this model would disproportionately impact young males. This is okay since, as a group, young males are significantly much more violent than other groups. The model is giving a negative impact to this group precisely because it is not biased. We should only be discussing "bias" if and when something systematically makes adverse discussion toward a group when those decisions are not rooted in reality about the group. A model that routinely scored old ladies as more violent than young men, might need to be investigated for bias, but the reverse would not be true.