I think you're handwaving the difficulty of identifying and addressing bias and discrimination. I'm not sure it's any easier to do for people than for computers. Computers have a huge advantage here because we can open up their brain and examine how they think, and once we create one that suits our tastes, we can replicate it faithfully. Humans are much more of a black box than a neural net would ever be.
If some parameter is well correlated with the bias, the model will discount that parameter.
It's not magic, it isn't going to just erase the bias, but a reasonable model will reduce bias rather than reinforcing it.
A far cry from "many algorithms" like its some kind of well established practice and fact.
https://arxiv.org/abs/1708.01785
https://arxiv.org/abs/1710.00935
https://arxiv.org/abs/1702.04595
https://arxiv.org/abs/1801.01693
https://arxiv.org/abs/1503.02531
https://arxiv.org/abs/1712.03781
https://arxiv.org/abs/1710.07535
https://arxiv.org/abs/1702.02540
https://arxiv.org/abs/1603.02518
http://ieeexplore.ieee.org/document/977279/
etc
It is kind of far cry from established practice but how far cry from established implementation are your fears of unintended abuse? It's not nearly at the point yet to be paranoid. Having bugs in ANN before debugging makes them no different from usual software that is already used in forensic research. And nobody in their right mind will deploy networks that impossible or prohibitively difficult to debug for any kind of critical tasks. Otherwise irrational people can abuse power even without any software or ANNs.
Before you reply that detectives might/would have done that anyways, the point is that at least that is identifiable and addressable, how do you determine how a neural net makes its decisions?