32
See the leaked teen social media ban tech trial report that has experts worried
(www.crikey.com.au)
This morning I experienced this tech going wrong first hand. Someone mentioned and I decided to try a site called pimeyes. It lets you upload a photo, then it will try to find other examples of your photos posted online. Well I uploaded one of my photos. And it returned this:
I am a 38 year old woman. I don't recall using this site before, and I sure as hell have never used it to search for kids. I uploaded another photo, and that one worked. But still, that is a fucked up message to send someone just based on some shitty AI age estimation algorithm. Someone with a different face might always trigger the "minor" filter.
And while this was just for an image search site, it's much more serious on other sites. People do a lot of essential communication on social media platforms. What happens if you're completely shut out of all social media sites because you have a face that the algorithms decide look under 18? What happens if it's something even more important, like your bank's website or a government service page?
It's ridiculous that that they would claim 85% accuracy as some great success. That's a horrible success rate in this context. That means millions of people will be incorrectly flagged as minors and potentially lose access to entire regions of the internet. And how long until they start using this facial recognition, not as proof of age, but as proof of identity? How long til you have to scan your face to apply for government benefits, retirement, or access other government services? And what happens to those people who the face scanning algorithm just fails at? A 15% failure rate is awful. When you're imposing something on the entire population, you shouldn't even consider applying it until the success rate is more like 99.99% accurate. If the tech just isn't that good, then it simply shouldn't be used.