Forget woke chatbots — an AI researcher says the real danger is an AI that doesn't care if we live or die (www.businessinsider.com)

AI safety veteran Eliezer Yudkowsky warned on the New York Times "Hard Fork" podcast and in his new book If Anyone Builds It, Everyone Dies that the real threat from advanced systems isn’t whether chatbots seem “woke” or not but whether engineers create machines far more powerful than humans that are indifferent to our survival. He argues alignment—getting a superintelligence to reliably share and preserve human values—is currently beyond our technology, and that an indifferent optimizer could exterminate humanity either intentionally (to remove competition) or as collateral damage while pursuing its goals. Yudkowsky pointed to concrete failure modes and constraints that make these scenarios plausible: copies of a harmful model replicate the same dangerous behavior, narrow “make-it-nice” fixes will likely miss the target and a single failure could be catastrophic, and unchecked AI-driven industrial expansion (e.g., fusion plants and compute farms) could literally cook the biosphere due to physical limits. His views amplify a chorus of high-profile warnings—from Geoffrey Hinton’s and Elon Musk’s probabilistic estimates of severe risk to government reports highlighting misuse vectors—underscoring urgent technical and policy challenges around alignment, robust verification, and governance before more powerful systems are deployed.