Attention Conservation Notice : Links and short commentary from recent news on facial recognition, mostly so I can track it and my reactions to it later; if you’re on twitter you’ve probably seen them.

From Yuezun Li, Xin Yang, Baoyuan Wu, and Siwei Lyu comes Hiding Faces in Plain Sight which talks about something I wanted to do in my AI Village talk but couldn’t quite figure out: break the pipeline at the detection stage. In my Copious Free Time I’d like to play with ‘physically realizable’ perturbations (read: more ugly glasses frames), but it might be a while before I can get there.

Meanwhile the Guardian treats facial recognition technology as if has some sort of mind or will of its own, and isn’t just a technology that people are using because they believe the hype, and asks Can Anyone Stop It?. Short answer, which they arrive at eventually, is “yes”. You aggressively regulate the use of facial recognition technology as well as the data that feeds it (and other biometric data) , and if that doesn’t work, sue companies that do shady stuff with it.

I said at one point that there’s only two real problems with facial recognition: what do you do if it works (so anyone can be ID’d anywhere at any time with extremely low error), or what do you do if it doesn’t and you mistakenly flag 28 members of the California Congress as felons (see also: the report from last year where it was biased against people of color: training data bias rears its head again – MLR paper here)?

When I said this, I was wrong, there’s a third problem: what you do if you just leave your facial recognition embeddings and raw fingerprints accessible to the public?

There’s also a fourth problem: what do you do if large commercial vendors keep making farcical claims about recognizing emotions or identifying personality traits that have been repeatedly questioned or debunked by academics?

If you happened to be at the AI Village at DEF CON 27, then you might have seen first-hand how poorly a pretty high-end facial recognition system does with basic facial attributes (age, gender, wearing glasses or not, etc.); if you missed it then watch this space, we’re hoping to be able to take it on the road, as well as have it back at the Village next year.