Terence Tao: “We have to slow down AI”

Haider clips Terence Tao arguing that the pace of AI development is unjustifiably fast: society is changing everything without knowing what follows, in a system governed by “extremely nonlinear dynamics.”

Links

• Thread on X
• Terence Tao on AI — a living summary
• Related: tao-fields-medalists-ai-mathematics-declaration

The interesting part: the responses

The replies show how poorly the nonlinear-dynamics argument lands in an accelerationist frame. Critics recast Tao's call for caution as fear of professional irrelevance, insist that progress and geopolitical competition cannot be slowed, or point to prospective benefits such as medical research and assistive robotics. Another response asks why human adaptation should constrain mathematics at all if its goal is truth.

Those objections mostly answer a simpler claim—that progress is bad or AI should stop—rather than Tao's actual systems point. In a rapidly changing nonlinear system, uncertainty is not merely ignorance about a predictable trajectory: feedback loops, thresholds, and small interventions can produce disproportionate, irreversible effects. One notable counter-response therefore shifts the target from AI capability to the incentives of labs racing for valuation and prestige, suggesting that the speed is a product of the bubble rather than a social necessity.