How would that solve anything? Most machine learning algorithms being used today are black boxes and the people implementing them often have little understandin

How would that solve anything? Most machine learning algorithms being used today are black boxes and the people implementing them often have little understanding of exactly what transformations the model ends up making between the inputs and the outputs. Its entirely possible for seemingly-innocuous inputs to pick up outright problematic features (i.e. skin color from a photo id) or indirectly problematic features (i.e. income level and address located in an area predominantly populated by minorities) that its outputs hinge on. If anything, the issues with machine learning are even more problematic that what the OP posted, as simply identifying that something like systemic racism in a machine learning model can be very hard to identify.