The Iron House

This project was conducted as part of SPAR Spring 2025, a research fellowship connecting rising talent with experts in AI safety or policy.

Abstract

This article examines the accelerating race between the United States and China to develop Artificial General Intelligence (AGI) and its implications for global security. Drawing on Lu Xun's metaphor of the iron house – a sealed structure where sleepers face suffocation – we argue that humanity stands at a critical juncture, where the pursuit of AGI supremacy may paradoxically lead to the loss of human control over our collective future.

The analysis identifies three key dynamics that shape this competition. First, the race is driven by the perception that AGI represents the ultimate determinant of national power, creating a winner-take-all scenario. Second, an "Alignment Trilemma" forces both nations to prioritize rapid development over safety and international coordination. Third, the year 2027 emerges as a convergence point where AGI timelines, military preparedness, and Taiwan's semiconductor dominance create a window for potential conflict.

We examine how both nations are preparing for an AGI-dominated future: the US through reshoring critical supply chains, securing resources, and building massive computational infrastructure; China through innovation around Western restrictions and potentially destabilizing open-source strategies. The race's escalatory logic suggests that to maintain competitive advantage, nations may eventually cede decision-making to AI systems, crossing an "AI event horizon" beyond which human control becomes impossible.

The article concludes that without immediate global cooperation on AGI development, humanity faces an existential security dilemma. We propose five policy interventions: acknowledging the constraints of the security dilemma, implementing a global development pause, establishing robust international cooperation similar to those for nuclear proliferation controls, creating accountability mechanisms for AI developers, and maintaining meaningful human control. However, we acknowledge the near-impossibility of achieving these goals given current geopolitical tensions.

As Baker warns in The Centaur's Dilemma, if we fail to get the human part of human-AI collaboration right, the AI's capabilities become irrelevant. We may be the last generation with the agency to shape our relationship with artificial intelligence—the choice to break open the iron house remains ours, but not for much longer.