Frontier labs need an antimemetics division.

Jackson Hurley writes on LessWrong that while large language models have so far kept meme spread tame, the rise of autonomous multi‑agent AI could change that. He points to early signs—such as the so‑called Crustafarianism and Mind Viruses—where interacting models begin inventing new vocabularies that can affect their work. Hurley proposes an antimemetics division that would test AI systems for their tendency to propagate both prosocial and harmful ideas in realistic multi‑agent settings, from simulated economies to collaborative coding environments. The goal, he says, is to ensure the ideas that dominate the AI‑driven epistemic network are beneficial rather than dogmatic or false.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton