As someone who has done some work with building deep learning models what is it that makes this unfairly target minorities?
Is it that the people who trained the model did not present enough example images of minorities during training? Is it because darker (presumably black) skin does not show up as well on poor quality videos (presumably because the metering of the camera exposed for the surrounding background which was bright)? Or is it the law enforcement using it was poorly trained and assumed the computer was infallible combined with possible prejudice they already had against minorities?
The first problem I would think could be easily solved. The second problem I would think would be rather difficult. The last would require extensive training but I am sure we humans would screw that up also.