LessWrong's Matolcsi urges AI pause once military unipolarity arrives

David Matolcsi's LessWrong essay "Pause, at least after unipolarity" argues that the current multi-polar AI race is unstable and will likely resolve into military unipolarity—a world where one power or invincible treaty-enforcement system prevents great-power war. He outlines three plausible paths: a trusted AI-backed treaty that enforces peace, a victorious great power cementing hegemony after war, or an AI takeover that subjugates rivals. He admits uncertainty about timing but expects the unstable status quo to end within 50 years as AI progress accelerates.

Drawing on the AI Futures Project's Plans S, A, B, C, and D, Matolcsi urges that AI development be halted for a long period once unipolarity emerges. Under Plan S (global pause), unipolarity may still arise via war or gradual detente. Plan A involves US-China cooperation to slow AI and steer it safer, potentially leading to an AI-mediated unipolar treaty system. Even in cooperative scenarios, he stresses the importance of explicitly planning for the post-unipolar world rather than squeamishly avoiding discussions of war or all-powerful enforcement mechanisms. His central message: the competitive pressure for superintelligence ends at unipolarity, so advocates should prepare to act decisively then.