"We don't know how 12 years of schooling really teaches people how to become mathematicians either," he says. "You can't do brain surgery on the neural network and let's take out the part where it's really sinister because it doesn't exist." The idea that just as the brain can become capable doing lots of other things beside what you learn in school? So could this perhaps learn many other things besides the autocomplete function? Is that your claim at root in some sense? Yeah, absolutely. I mean, this, that sort of claim is already well empirically proven?"
They operate according to rules we can never fully understand. They can be unreliable, uncontrollable, and misaligned with human values. They're fast becoming as intelligent as humans--and they're exclusively in the hands of profit-seeking tech companies. "They," of course, are the latest versions of AI, which herald, according to neuroscientist and writer Erik Hoel, a species-level threat to humanity. Listen as he tells EconTalk's Russ Roberts why we need to treat AI as an existential threat.