A Critique of OpenAI’s AI “Brake” Argument and the Debate Over Control

Summary

This opinion article responds to OpenAI chief scientist Jakub Pachocki’s essay “An Alien Mind,” which argues that machine intelligence is largely grown through scaling compute, remains difficult to understand, and may eventually require recursive self-improvement safeguards, voluntary slowing, international coordination, and mandatory safety thresholds. The author focuses first on a tension in Pachocki’s proposal: chain-of-thought monitoring is described as becoming less reliable, yet an automated AI researcher is presented as a way to solve alignment. The article also argues that using rapid model development to build defenses against other AI creates an arms-race logic that conflicts with a broad call for everyone to slow down. It rejects the idea that AI is principally “grown” rather than designed, stressing that people choose the loss function, data, optimizer, compute budget, training stop, and release decision, even if internal representations remain poorly understood. From this, it distinguishes limited scientific understanding from fundamental uncontrollability. The article then interprets AI catastrophe narratives through Western cultural stories of creations turning against their makers, contrasting that framing with the Chinese tale of Yan Shi taking apart a mechanical figure to make its construction visible. It cites US polling and reported cancellations of data-center projects as evidence that AI infrastructure opposition has become a cross-party domestic issue tied to electricity, water, and land. Its policy conclusion for China is to pursue safety, interpretability, auditability, and open verification as capabilities, while resisting externally imposed slowdowns. The closing argument is that opaque systems should become more open, reproducible, and subject to multi-party verification.