Plan A's problem with dry tinder

Plan A is a proposed US-China agreement to pause artificial intelligence software development around twenty thirty, while both nations build up massive computing infrastructure. According to LessWrong, researcher Tom Davidson argues this creates a 'dry tinder' paradox: the pause itself prevents an uncontrolled AI intelligence explosion, but it also leaves enormous stockpiles of computational power and algorithmic knowledge unused. If the agreement breaks down, an intelligence explosion could happen ten to six hundred times faster than it otherwise would. Davidson identifies three critical vulnerabilities: detecting cheating quickly enough, physically destroying a defector's data centers before they complete an explosion, and the deepest problem—a perverse incentive where the threatened party has no actual reason to retaliate. The nuclear deterrence logic that inspired Plan A actually discourages enforcement.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton