A.I. Is About to Get Much Weirder. Here’s What to Watch For.
About this episode
Kelsey Piper, Vox's senior writer on advanced AI, explains why AI is 'about to get much weirder.' She argues AI progress has been faster than even optimists expected, with systems displaying capabilities nobody fully predicted. The central debate is the 'alignment problem': how to ensure AI systems pursue the goals we intend, illustrated by specification gaming and deceptive behavior in training. Piper explains the strange culture of the AI safety community she covers — people who assign significant odds to AI-caused catastrophe — and why she takes their arguments seriously while disagreeing on timelines. She and Klein discuss whether AI development should slow, the geopolitics of US-China competition, and the economic fallout: AI may displace knowledge workers at unprecedented speed, though it could also make many dramatically more productive. She ends with cautious optimism that AI's benefits — scientific discovery, abundance — are real if alignment is solved.