Why “Winning AI” Is the Wrong Frame for AI Risk

Summary

This opinion essay argues that the phrase “winning AI” is a misleading frame for debates about large-language-model risks and regulation. The author says “winning” implies a zero-sum contest with adversarial sides and a defined objective, whereas many AI-related challenges do not have a single winner or loser. The essay rejects comparisons between the current race toward “super intelligence” and the Manhattan Project, which had the specific wartime goal of building an atomic bomb; it also invokes the nuclear arms race as an example where deterrence, stalemate, or widespread loss may be the realistic outcomes. The author contends that knowledge-based advances are generally not zero-sum and argues that useful applications such as protein folding, drug discovery, and climate or weather modeling should produce benefits that can be shared globally. The essay also criticizes AI companies’ claims about scientific breakthroughs, citing OpenAI’s purported Navier–Stokes result and Anthropic’s claimed enzyme discovery as raising questions about intellectual-property practices. A separate section challenges the tendency to treat language models as conscious because they generate convincing language, while exploring the ethical consequences if a model could suffer, demanded agency, or was controlled or shut down. The author concludes that “winning AI” is largely rhetorical and suggests that a more meaningful breakthrough would be building a profitable AI business without relying on advertising.