Can we build an early warning system for loss of control to AI?

David Johnston, writing on LessWrong, explored whether we can build a mathematical model to forecast AI loss-of-control risk. The challenge is real: the concept is contested and observing deceptive AI might be infeasible. Yet he came away cautiously optimistic about feasibility. His model predicts high loss-of-control risk on our current trajectory, though Johnston cautions the predictions aren't robust. More valuable than the forecasts themselves, he argues, would be using such a model as a monitoring layer for AI development—a system to detect drift toward misaligned behavior before it becomes catastrophic.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton