"What if racist facial recognition algorithms result in false arrests for people of color?" We've known of this speculative harm for years. But did not take the step to mitigate harm by banning use of this tech by law enforcement. So now it's happened. npr.org/2020/06/24/882…
Very recent steps taken by Amazon, Microsoft, and IBM are great but also too late - law enforcement already has this technology and smaller companies with less scrutiny are continuing to distribute it. I'd like to know what @DataWorksPlus is doing to mitigate harm. 
I'm reminded of @robotsmarts' work on overtrust for AI. dl.acm.org/doi/abs/10.114… That seems to be happening here - well the computer said it's him and computes don't lie. So why should they waste their time investigating?
What really concerns me about trust in AI, in the context of racism in law enforcement, is a perception that AI is LESS racist and therefore creates a safe, unbiased decision. A police officer might think all Black people look the same, but so does the AI.
I'm sure some think the cost-benefit analysis weighs in favor of facial rec - "but think how many criminals they'll catch and false positives are rare!" - but if false positives are always POC, then harm hits them disproportionately and reinforces existing power structures.
During a talk I saw @schock elegantly answer a question about why designers should care about "edge cases" - e.g., if there aren't a lot of trans people, why do we care if this tech harms them? I suspect a lot of technologists are thinking that way. Anyway, read "Design Justice."
I've been going on tears about ethical speculation recently (you have to anticipate harms so you can mitigate them!) and in that context this is a pretty depressing example, because we've KNOWN about the harms of facial rec for a long time... and have done little about it. :(
in conclusion... 👇👇👇
