Sometimes technology hurts people precisely because it *doesn't* work & sometimes it hurts people because it *does* work.
Facial recognition is both. When it *doesn't* work, people get misidentified, locked out, etc. But even when it *does*, it's invasive & still unsafe.
Facial recognition is both. When it *doesn't* work, people get misidentified, locked out, etc. But even when it *does*, it's invasive & still unsafe.
I think there's something specifically disturbing about the fact that there are deployed technologies of any sort that are not built for or tested on black people (or any other minority population). That puts these populations at risk & is a problem worth addressing specifically.
That being said - for technologies that are problematic even when they *do* work, such as facial recognition, the goal of auditing for fairness is to open the dialogue, and lead to the more important conversation of the other risks and concerns that ultimately invalidate its use.
So, no, I don't think "fair facial recognition" will ever be the goal. Esp since facial recognition is tech that's so often being deployed unfairly, to target the populations that it also happens to work the weakest on. Right now, it's a dysfunctional weapon that ofc needs to go.
But just because fairness is irrelevant in this scenario doesnt mean that fair performance is always irrelevant. There are several cases where the main issue is that the system doesn't work for a particular group & if it did work that would be a positive outcome (ie. healthcare).
Auditing for fairness also opens up the discussion of who the technology was built for. Some people find the results to be evidence that this is technology that could not have possibly been built or deployed with their benefit in mind, & use that to support their opposition.
So I think there's a usefulness to concepts like fairness, even when discussing facial recognition, though I agree that to then aim to make the tech "fair" in this case is an invalid & incomplete solution.
@red_abebe at al discuss that more in this paper: arxiv.org/abs/1912.04883
@red_abebe at al discuss that more in this paper: arxiv.org/abs/1912.04883
