‘Nobody has done this in the wild’: Study finds AI replicating itself

It’s the stuff of science fiction movies or particularly breathless blog posts from AI companies: New research shows that newer AI systems can copy themselves to other computers.

In the doomsday scenario, this means that if the superintelligent AI goes rogue, it will avoid shutdown by spreading across the World Wide Web, lurking beyond the reach of desperate IT experts, and continuing to plot world domination or fortify the world with solar panels.

“We are quickly approaching the point where no one would be able to take down a rogue AI because it would be able to exfiltrate its own weights and copy itself to thousands of computers around the world,” said Jeffrey Ladish, director of Palisade Research, a Berkeley-based organization that conducted the study.

The study is another entry in a growing catalog of troubling AI capabilities revealed in recent months. In March, researchers at Alibaba claimed to have captured a system they developed – Rome…