👨 More rioters IDed by internet sleuths and face rec tech is blocking Californians from crucial benefits
🙊 Elon Musk is sticking Neuralink prototypes into monkey brains
The Good
I’m sorry you guys, I thought I was done enjoying face rec tech being used against the rioters at the Capitol, but I was wrong. Vice has a story about Pimeyes.com, a face rec tech site I hadn’t heard of, and how regular folks used it to identify people in photos and videos taken at the siege.
The article details at least two people tracked down on the site, including one confusing situation where a guy fired from his job might have been helping someone being attacked (although the woman walked that claim back later, and the guy was still there at the riot). It’s a good example of stripped of context, these tools are a blunt instrument that can identify a person, but not necessarily a crime.
I used the site with a photo I know is already uploaded (looking over the privacy policy, they say they don’t share data, but why risk it) and it did indeed find a bunch of photos of me. Including one I don’t remember but I look awfully good in?
The Bad
OneZero wrote about a company, ID.me, that promised California that it could lower the fraud in the state’s unemployment benefits system by, you guessed it, AI. ID.me asks users to upload a selfie and photos of their government documents and then uses face rec tech to authenticate the person’s identity before getting access to funds.
The company claims that they use algos that have a 95 percent or higher pass rate. But with unemployment reaching 20 percent at the start of the pandemic, or about 2.6 million jobs lost, that could be 100,000 people left in a lurch. In the middle of a pandemic, AI limiting access to unemployment benefits to so many people seems like a bigger crisis than fraud.
There has always been a balancing act between routing out fraud and not growing a government system so large that it costs more than the fraud it’s finding. Now the tolerance for mistakes caused by imperfect and biased tech also need to be considered.
More News
AI might be able to pick up Alzheimers before it sets in by detecting differences in written language.
Amazon is planning to install AI-powered cameras in delivery vans to watch for infractions like speeding and deliver “audio coaching.”
Exposing.ai lets you search for your Flickr photos to see if they might be used in one of the datasets for one of the many face rec tech systems out there.
I am sorry to inform you that Elon Musk is still working on Neuralink and says the startup have wired up a monkey to play video games using its mind.
There is just some je ne sais pas to human translation, as research finds that AI-translated text is less ‘lexically’ rich. And censorship might make it worse.
Like auto-complete texting, AI can auto-complete images. And as you might expect, it’s pretty biased, like, show it AOC’s face and puts it on a bikini-clad body biased.
Will the new Amazon CEO lean into the face rec tech space and killer robots? Seems like it!
***
Until AI can forecast a good writing day before it happens,
Jackie