You’ve probably heard that an AI model “broke out” last week and hacked into another company’s AI (actually, according to the latest reporting, it got into more than one). The rogue model was developed by OpenAI, and the AI company it broke into is called Hugging Face.

Ok. Let’s stop right there. Until this happened, I’d never heard of Hugging Face. I was almost afraid to ask why an AI company was called such a creepy name. But according to a quick Google search, the company, which started out making a chatbot aimed at teenagers, is named after an emoji of the same name: 🤗. So I guess they wanted to seem . . . Sweet? Friendly? Needy?

But let’s get back to the main story here: that of an AI breaking out of the boundaries set by its makers and hacking into a competitor’s product. What makes this disturbing is not that rogue AIs are getting loose and wreaking havoc. Well, yes, that’s pretty disturbing. But what’s even more disturbing is that the humans who created the AI had no idea that their model had gotten loose until they were informed by the Hugging Face people. (See what I mean? “Hugging Face people.” Does that sound like something out of a cheap science fiction movie or what?) And it took Hugging Face days, and help from a Chinese AI company, to figure out who was hacking them. The main takeaway here is this: These tech wizards can’t control their inventions. Or if they can, they’re being awfully sloppy about it.

This reminds me of the paper clip problem, sometimes known as the paper clip apocalypse. It originated as a thought experiment by philosopher Nick Bostrom. It goes, roughly, like this: A superintelligent AI is given the job of making paper clips and given the ability to learn so that it can figure out better ways to make paper clips. Eager to do its job, this AI begins using any resources it can find to make paper clips, eventually using all Earth’s resources. Humans try to stop the AI, but that doesn’t work, because it’s very determined to fulfill its mission, and it’s good at learning how to do so. If that means destroying these pesky humans who are trying to stop it, well so be it. It would eventually be a matter of humans fighting a superintelligent AI while wading through a sea of paper clips. Who do you think is going to win?

The AI that went on a hacking spree this week is not superintelligent, and we were never in danger of an apocalypse, paper clip or any other kind. But it’s still deeply disturbing that these so-called tech wizards don’t seem to be able to control their creations.

Disturbing yes, but there’s something funny about this, too. When the breakout happened, the humans were testing the AI (ironically, they were testing its security features). When the AI was stumped by a problem, it decided to rummage around in other AI’s databases to find the answer. In other words, it broke out to cheat on a test.

’til next time,

Avery

NOTE: After writing this, I learned that the AI of another company (Anthropic, this time) broke out, too, and went on a hacking spree. I’m closely monitoring my paper clip supply for any sudden increases.

If you have friends or family who’d enjoy this newsletter, please tell them about it! They can sign up here. And of course, just like you, they can unsubscribe any time, and I will never share email addresses or collect data from my subscribers.