The pioneering computer scientist Alan Turing once said a machine can be thought to have achieved human-level intelligence when we can’t tell it’s a machine, a moment we seem to have arrived at. It’s a breakthrough that has inspired a lot of fear. Yet the biggest challenge of A.I. isn’t what the machines are thinking of us, according to Stuart J. Russell, one of the foundational figures of modern A.I. and a professor at the University of California, Berkeley. It’s the way we have been thinking about these thinking machines.
Since the 1990s, Russell has been saying that the technology was both an opportunity and a threat—and that a relatively straightforward change in approach would maximize the benefits while reducing the risk. No one much listened. Perhaps Hollywood had lulled us into believing that runaway machine intelligence was, well, just the stuff of science fiction. Or maybe, in a world overloaded with impending doom, we just couldn’t make mental space for one more threat. (The plural of apocalypse is apocalypses, Alexa told me, adding, “That’s a fun question.”) Then ChatGPT arrived—followed by Claude, Google Gemini, and a swarm of others—and Russell’s warnings were suddenly being heard by everyone from Donald Trump to Xi Jinping.
