What happens when we make an artificial intelligence that's smarter than us? Some AI researchers have long warned that moment will mean humanity's doom.
Now that AI is rapidly advancing, some "AI Doomers" say it's time to hit the brakes. They say the machine learning revolution that led to everyday AI such as ChatGPT has also made it harder to figure out how to "align" artificial intelligence with our interests, which means greater risk for humans once AI can outsmart them. Researchers into AI safety say there's a chance — a "p(doom)" where the "p" stands for "probability" — that such a superhuman intelligence would act quickly to wipe us out.
NPR's Martin Kaste reports on the tensions in Silicon Valley over AI safety.
For a more detailed discussion of the arguments for — and against — AI doom, please listen to this special episode of NPR Explains:
And for the truly curious, a reading list:
The abbreviated version of the "Everyone Dies" argument, in the Atlantic.
The "useful idiots" rebuttal, also in the Atlantic
The potential timeline of an AI takeover
Research into "AI Faking" and deception
Maybe AI isn't speeding up smarter AI, at least not yet. Research from METR
Analysis — and skepticism — from experts about the near-term likelihood of human- or superhuman-level artificial intelligence
Copyright 2025 NPR
