Friday, May 24, 2024

Like Google Search, But For Faces: Clearview AI And The Dreams of Facial Recognition

I was at a talk recently by a computer scientist, and the speaker ended on an ominous note, explaining that she does everything she can to keep pictures of herself off the internet. On her slide was an image: a book called Your Face Belongs To Us.

Your Face Belongs to Us is a non-fiction book by NYT reporter Kashmir Hill about Clearview AI -- a facial recognition company -- which I decided to read immediately. I figured I would learn that facial recognition is possible, widespread, and creepy, and I did. It's no secret that Clearview's algorithm matches faces to a database of more than 20 billion images collected from the Internet -- including from Facebook et al. Upload a picture of a face, and it'll not only tell you info about that person and who they are, it will find other pictures of that person -- including pictures where that person is in the background, or much younger, or wearing a mask.

But a few things surprised me. It surprised me -- though it shouldn't have -- that some of the players in facial recognition have beliefs rooted in eugenics. The background dream is not just about recognizing faces, but also about the phrenological predictive possibilities: that the face will reveal the character, so we can prevent crime, and improve the human race, by identifying "degenerates" through their features and eliminating those people from existence. 


Chapter 2 of the book gives a short overview of the history of this idea in western culture -- which we know appeared not only in Nazi ideology but also in the work of "progressive" political thinkers since at least the Victorian era, and has persisted. If you want to read more about what Hill describes as "racism masquerading as scientific rigor" and its connections to AI, check out "The TESCREAL bundle: Eugenics and the Promise of Utopia through Artificial General Intelligence

I was also interested to learn that while Clearview is currently used mostly by law enforcement agencies in the US to identify people from surveillance footage or in crowds and protests, a central aim of its creators has been to market the product to private users -- especially business titans and large companies. Not only could you efficiently determine, for example, who should be allowed into a building, based on their face, the marketing possibilities are extraordinary: every time someone walks by, an ad could appear tailored specifically to them, based on their identify through recognizing their face.

That hasn't happened, but it's not because it can't. There have been lawsuits, and strategic restraint, and social pressure from privacy activists making that not happen. But the world in which a billboard is tailored to your Instagram is closer than you think.

Surprisingly often, people bring up a dream of having a pair of glasses that will tell you people's names to avoid embarrassment or seem suave. A Facebook guy describes the "universal experience" of being at a dinner party and seeing someone you know and forgetting their name. They could fix that, Facebook could! If they put their own facial recognition into virtually reality glasses.

It kind of blew my mind that people want to harness a vast, unpredictable, and potentially unethical technology to avoid a moment of awkwardness. I guess that shows what we've all known all along: that innovation happens where money can be made, which means solving the small problems of wealthy people. The tech has not been made available in that way yet, partly for the obvious reason that having people be able to identify strangers in any context is creepy and terrifying: imagine you're a woman in a bar and a guy can look up everything about you, even your address, just from your face. 

Overall, while I knew people -- and especially police -- were identifying faces, and while I knew that was a troubling, potentially evil thing, I hadn't thought of all the ways people might think identifying a face might be useful. Once I saw them, the world of more facial recognition seemed even bleaker and more dystopian than I thought.

I'll just leave you with one final random item. In terms of faces, the biggest surprise to me was about gender and hair: " if two ears are showing," one expert says, "there is an 85 percent chance that the person is male."  As a woman with long hair, who always wears my hair up, so my ears are always visible, I just found that interesting along various dimensions.

No comments: