Can facial recognition do right by trans, non-binary subjects? There is doubt
A disparate pair of publications is questioning if facial recognition can be used reliably when it comes to LGBTQ+ subjects.
In the podcast, Caleb Brown, director of Cato’s multimedia operations, says of facial recognition “is, for the most part, not ready for prime time.” Brown is talking with a Cato colleague, Matthew Feeney, director of the Project on Emerging Technologies.
Feeney says studies indicate that facial-analysis business tools may be better at determining sexual orientations than people.
That has convinced some outfits to find related problems to solve with AI code. But better is not perfect.
But the risks of humiliation or harm for people who are not cis male are many, he says. For example, in 2019, the Berlin Transportation Co. wanted to make a supportive statement on Germany’s Equal Pay Day, by offering all women riding a 21 percent discount, presumably the average pay differential between women and men in that nation.
Automated kiosks decided genders. Error rates were not publicized, but there is little doubt that the software misidentified some riders’ orientation, causing some resentment.
Similarly, some U.S. state governments have or would like to enact so-called bathroom bills requiring people to use the gendered bathroom corresponding to the gender assigned on a birth certificate, Feeney says.
The only feasible way to accomplish this would be placing face biometric cameras at a restroom door. And that, of course, invites police to act on what an algorithm decided. A transparent law enforcement agency that engenders trust in facial recognition could keep this and other situations from blowing up into protests.
Mitek has come to a similar conclusion: “Facial recognition is an ineffective biometric for LGBTQ+.”
It is only going to get better, according to Mitek’s post, if the industry can be “open continually listening to the needs and challenges faced by all demographics as we develop products.”
That seems like a commonsense plan, but this topic splits more hairs than a barber.
In the meantime, everyone involved has to understand that “facial recognition isn’t always the right solution.” That is a refreshing point of view from any corner of the technology industry.
The post cites the results of design sessions conducted the University of Illinois and University of Michigan with 21 people in the trans community. Trans-inclusive technologies include those for changing bodies, safety and finding resources as well as those changing appearances and/or gender expressions.
Good to know before units start coming down the line.