AI isn't going to kill all humans. But the entitled billionaires and wanna-be...

AI isn't going to kill all humans. But the entitled billionaires and wanna-be billionaires might be if we refuse to apply our laws to them.

The Hugging Face hack (which OpenAI just calls an "incident" in their August 26th blog) is just one more in a series of careless incidents that don't prove anything about the "intelligence" of frontier LLMs, but do prove that the companies developing them are careless.

If you or I had ordered a computer to breach another company's computer without an AI involved we would likely face a criminal investigation and prosecution. Consider how we have treated cyber criminals like the famous Kevin Mitnick who wasn't even allowed near a phone because they feared he would launch a nuclear weapon by whistling into the phone (an absurd proposition to be sure, but is the hype around AI not this breathless?).

Yet no hints of criminal prosecutions or investigations have been made when it comes to OpenAI hacking Hugging Face or other similar incidents, which are not the result of some intelligence doing something that it was not asked to do, but are in fact the result of the tools doing EXACTLY what they were told to do.

Years ago when getting my hunting license the instructor told a story of how they had traced back a bullet that a hunter fired that had struck a boater he couldn't see in the middle of a lake through quite a bit of trees and distance. The instructor told this story because the importance of checking your direction of fire isn't just a good idea, you can be held criminally and civilly liable when you're the one operating the weapon, regardless of your intent.

A while back I wrote about the Watson Grinding / 3M case (how the person hired as an expert witness used ChatGPT to do his job poorly) and how 3M, the company behind one of the sensors that could've prevented the tragedy, was found to be 30% responsible for it. What percent liable should an AI company be when it can't be bothered to monitor its own training processes? I believe the people who carelessly set up these training/testing systems may have committed multiple crimes. I also believe that until we're willing to apply our laws seriously that the same sorts of behavior will continue.

The machines aren't coming for you. The lawsuits will. "The AI did it" is going to hold up about as well as "the computer did it" would for you or me. Build and document your work better than OpenAI or you'll end up in 3M's shoes.