AI could end Thucydides traps via hegemon or AI-negotiated pacts

In a speculative essay on LessWrong, djbinder explores how advanced AI could end the Thucydides trap—the structural competition between rising and declining powers that often leads to war. The author outlines two plausible outcomes: first, one actor achieves AGI and grows so rapidly it becomes a hegemon, ending any meaningful competition. Second, if no hegemon emerges, rival nations could use AIs to negotiate positive-sum agreements, surrendering hard power in exchange for enforceable cooperation. Unlike human institutions, AI systems can credibly commit to rules indefinitely and detect defection, solving the security dilemma that has historically made disarmament unthinkable.

The essay notes that commitment failures have driven wars for centuries: states cannot credibly promise not to exploit a disarmed neighbor, making any disarmament deal impossible. AI could serve as a global enforcer, much like a federal government does within a state—able to punish violations and verify private information. The author acknowledges challenges, such as human leaders blocking AI bargaining or locking themselves into rigid positions. If AI can indeed construct and enforce these deals, the result could be a global singleton wielding a permanent monopoly on hard power. Whether that outcome is desirable remains an open question, but it offers a novel path beyond the historical cycle of competition and conflict.