LessWrong post maps how AI could enable extreme power concentration even after alignment

In a detailed LessWrong post, pepijn_cobben argues that even if the AI alignment problem is solved, the technology’s ability to automate human labour and accelerate R&D could lead to extreme power concentration. Drawing on Michael Mann’s IEMP model—which breaks power into ideological, economic, military, and political dimensions—the author shows how AI shifts economic leverage from labour to capital. For instance, the labour share of GDP (~55%) could drop as AI replaces workers, concentrating wealth and influence among capital holders. This dynamic is reinforced by second-mover advantages in AI development, giving a decisive lead to early adopters.

The post outlines several threat models, including AI-enabled surveillance and loyal agents that replace soldiers and bureaucrats, making the state more totalitarian and stable than ever before. While the LessWrong community has discussed related topics like Gradual Disempowerment and AI-enabled coups, Cobben notes that only a few organizations (Forethought, Formation Research, EuroSafeAI) are actively working on the problem. He calls for more technical researchers to contribute, offering a framework to identify tractable projects. The post is intended as a preliminary map to catalyze discussion and action in the neglected area of AI governance and power concentration.