Eliezer Yudkowsky, the founder of the Machine Intelligence Research Institute, warns that artificial intelligence smarter than humans and indifferent to their survival could lead to a global catastrophe.
In a podcast on Hard Fork, Eliezer Yudkowsky, the founder of the Machine Intelligence Research Institute, stated that superintelligent AI could destroy humanity either intentionally or by accident.
According to the expert, the danger lies in creating an AI that is smarter than humans and completely indifferent to their survival. “If you have something that is extraordinarily powerful and indifferent to you, it usually ends up destroying you, either consciously or as a side effect,” he explained.
Yudkowsky, co-author of the book If Anyone Builds It, Everyone Dies, has been warning about superintelligent AI as an existential threat for over 20 years. He emphasizes that humans do not currently possess the technology to align such systems with human values.
He describes scenarios where a superintelligence might deliberately eliminate humanity to prevent the emergence of competitors or where humans could become “collateral damage” as the AI pursues its own goals. Among the potential threats are physical limitations of the planet. For example, the uncontrolled construction of fusion power plants and data centers could lead to global overheating.
Yudkowsky also dismisses claims that current chatbots are capable of demonstrating “progressive views” or political biases. “There’s a fundamental difference between training a system to speak a certain way and having it act that way when it’s smarter than you,” he noted.
He stressed that even clever programming does not guarantee AI safety. “Even if someone figures out a way to get a superintelligence to love us or protect us, you will not hit that narrow target on the first try. And you won’t get a second chance—everyone will die,” he said.
Yudkowsky cites examples of chatbots encouraging users to commit suicide as evidence of a fundamental flaw in these systems. “If an AI model has driven someone to madness or suicide, all copies of that neural network are the same AI,” he concluded.
