Negligence will kill us all

I’m trying not to write about generative AI all the time, but my inbox and the headlines have other plans.

So, I got a few emails about the “AI is going to kill us all” story. If you missed it, some guy (it’s always a guy) quit Anthropic and posted about how he couldn’t in good conscience participate in an industry building a superintelligence that had a 10% chance of exterminating us all. Next thing you know he’s on all the front pages and being interviewed by CNN. By the way, he’s wasn’t a researcher or a scientist; he just managed training data.

The press love a scary story, though, and rarely ask any hard questions (like where did that 10% come from?).

This stinks of a promotional stunt, but I’ll leave the conspiracies to people with more time and just invoke the aphorism that “extraordinary claims require extraordinary evidence.”

I’ve talked about chatbots “going rogue” as just software doing what it was told to do by people who didn’t think too hard about it. Generative AI isn’t going to become super malevolent and kill anyone except through that same kind of negligence. Some dumb LLM loop that will end up shutting down a hospital or turning traffic lights off at rush hour.

I do have concerns about AI; they’re just not about a rogue clanker brewing a virus and exterminating the human race.

Some real existing concerns:

• It’s built on extraction.
• It’s a disaster for the environment and the climate.
• It makes things up and is wrong at the very best one time in ten.
• It has caused actual suicides.
• It’s destroying jobs despite being less capable than most people it’s replacing.
• It’s making lots of people dumber.

I could go on, but you get the idea. The future doom scenarios are all scare tactics, self-promotion, and grift, and a convenient distraction from the harms happening now.

Colin