Slashdot Mirror
Artificial Intelligence Has Race, Gender Biases (axios.com)
An anonymous reader shares a report: The ACLU has begun to worry that artificial intelligence is discriminatory based on race, gender and age. So it teamed up with computer science researchers to launch a program to promote applications of AI that protect rights and lead to equitable outcomes. MIT Technology Review reports that the initiative is the latest to illustrate general concern that the increasing reliance on algorithms to make decisions in the areas of hiring, criminal justice, and financial services will reinforce racial and gender biases. A computer program used by jurisdictions to help with paroling prisoners that ProPublica found would go easy on white offenders while being unduly harsh to black ones.
You can't have AI that learns on its own and have AI that isn't racially biased unless you artificially code blocks to it reaching certain logical conclusions.
What? No? You don't block their inputs, not their outcome. You just hide the race of the targets it's looking at. As in, the sql library which has all that data it's digesting? You just exclude the 'race' field. DONE. The algorithm isn't judging based on race.
Nope, excluding the "race" fields does not mean the algorithm isn't judging on race. If the other fields have race invisibly encoded into them, it can still be judging based on race... but now it's doing it in a way that you can't see any more.
That won't stop it from looking at... say... the location where someone lives for approving or disapproving a loan. And lo-and-behold that's pretty similar to looking at someone's race.
Exactly. Race can be coded into other data.
But that's not their race and an unbiased look at something that DOES indicate loan-worthiness. If that's the sort of thing the ACLU disapproves of, they've got a very difficult fight on their hands, because where does that end?
It is a difficult problem. But just because it is difficult to exclude invisible bias due to race does not mean that it is not desirable to do so.
In the example you give, suppose that whites living in black-majority neighborhoods are, for some reason, likely to not repay loans (possibly because if they weren't financially distressed they'd move to the lily-white suburbs); but blacks living in black-majority neighborhoods have no problem (because there's nothing exceptional about blacks living in black-majority neighborhoods, that's just how "majority" is defined.) So, an algorithm tags "living in black majority neighborhood" as correlating with defaulting on loans. The net result is that blacks are denied loans even though they do not have a higher probability of default. The results of the loan algorithm are not race neutral-- even though the data appears to be both objective and not explicitly including race. But the results are all that matters. How do you make loans race neutral in this situation?
It's hard. But, again: just because a problem is hard, doesn't mean that it should be ignored.