A Solution to Cryptographic Boxes for Unfriendly AI

A researcher on LessWrong proposes a solution to a decade-old AI safety problem: how to safely run a dangerously powerful AI in a cryptographic sandbox. The existing approach, from twenty ten, uses homomorphic encryption—encrypting code so thoroughly that even the person running it learns nothing about it. But this method relies on assumptions that might fail under infinite computing power and runs slowly. LessWrong reports the proposal is Secure Multi-Party Computation, a protocol from the nineteen eighties that distributes computation across multiple parties instead. The advantage: no mathematical assumptions needed, and significantly faster. The principle allows powerful AI to run while keeping its logic hidden from everyone—addressing a core challenge in AI safety.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton