Detroit police chief pushes back against Rashida Tlaib's racist suggestion

[aj3647]

She was saying, basically, that because white people aren't good at discerning the faces of black people, they shouldn't be involved with the facial recognition program.

Note that the studies that show this effect are pretty shaky. The controls they used were dodgy and, if I recall correctly, were amongst the social sciences studies that were unable to be reproduced when a researcher went through a bunch of these studies a couple of years ago. That's not to say the effect doesn't exist, but there isn't a lot of good evidence that it does.

Incorrect. The "cross-race effect" has been extremely well-researched in literally hundreds of scientific studies and has been shown to be present across all races/ethnic groups [[it's not limited to white people, this is true of everyone). Similar effects have also been shown in the literature for age and gender, i.e. we're better at distinguishing between people who are similar in age than we are as opposed to people who are much much older or younger.

https://en.wikipedia.org/wiki/Cross-race_effect

So from a scientific perspective, it actually makes sense to have black officers handle facial recognition for black subjects [[as they are less likely to make a mistake than white officers, on average). Where Tlaib is wrong is that black officers would do a better job distinguishing between non-black subjects, as the cross-race effect also applies to African-Americans.

The best solution would be to have a diverse team of analysts, but probably one that is weighted more heavily towards African-Americans given the racial demographic makeup of the city.

[Wesley Mouch]

Incorrect. The "cross-race effect" has been extremely well-researched in literally hundreds of scientific studies and has been shown to be present across all races/ethnic groups [[it's not limited to white people, this is true of everyone). Similar effects have also been shown in the literature for age and gender, i.e. we're better at distinguishing between people who are similar in age than we are as opposed to people who are much much older or younger. ...
The best solution would be to have a diverse team of analysts, but probably one that is weighted more heavily towards African-Americans given the racial demographic makeup of the city.

Your point is well taken. No dispute. And not surprising. People from a given group can see subtle distinctions between other members of their group. Well gee, they see them more, so much more opportunity.

My question is how much this system relies on humans. My personal experience with facial recognition is via Apple Photos. No humans involved -- unless you consider Siri to be human. And the system does an amazing job of identifying people in pictures. Sure, I have to identify people here and there as she learns, but I've yet to see a mistake. Sure, my photos are all a bunch of old white males, so maybe my mileage will vary.

Point is that this technology is evolving fast, and will increasingly not involve humans except as a last-step check. So do we really need to hire a team of world-class, racially, ethnically, and religiously diverse screeners to verify ever-improving software? Or can we just let Chief Craig do his job, and prove to prosecutors and eventually judges and juries that they 'got the right man', so to speak.

[[When a Uzbek transgender citizens photo is being checked, do we need to make sure our team as the right proportion of Uzbek transgender analysts, or can we just hire some well-qualified people and get on with chasing the bad guy/gal/etc.?)