Wait Baltimore and Detroit are like ~60-80% black, and people are drawing inferences about set of six where the people were black - instead of say 4-5 black and 1-2 white?!
[this doesn't excuse the police actually arresting the people]
I think you'd need to know the distribution of criminal suspects, which may or may not reflect the general population. For example if 50% of the suspects are white, then 80% of the population being black wouldn't matter. Right?
EDIT: On second thought perhaps not, I think some are arguing for a biased training dataset, which would presumable be the "population."
No idea - my whole point is that without more data you can’t really draw any inferences here, (e.g. it’d be nice to know how many people were sent for facial recognition, and the ethnic composition of that group, the. It would be nice to know the misidentification rate in that BEFORE the police went out arresting people), one answer here may be that facial recognition’s failure rate is quite high, but the police don’t do the requisite follow up work when it’s a black suspect vs a white suspect. Which brings me back to my original point, we need more data, but what limited data that is supplied doesn’t YET give us a smoking gun.
[separately I’m hopeful that those wrongly arrested enjoy success in the civil courts]
[this doesn't excuse the police actually arresting the people]