28 What Happens When 10,000 AI Agents Tackle a Million-Dollar Physics Problem?

For nearly two centuries, scientists have relied on the Navier-Stokes equations to model fluid movement, yet mathematicians could never prove whether the underlying math remains valid in every possible scenario. The central tension exists between approximate computer simulations used in daily engineering and the strict mathematical proof needed to guarantee that fluid velocity never reaches infinity. While OpenAI claims to have demonstrated a velocity breakdown using AI agents, mathematicians note that the core unforced scenario remains unsolved.

When predicting the trajectory of a thrown ball, standard rules of motion apply easily because the object moves as a unified whole. Pouring a bucket of water presents a fundamentally different challenge, as every portion of the liquid moves independently while interacting with adjacent sections. To manage this complexity, mathematicians track velocity fields using differential equations rather than attempting to track individual droplets. However, fluid speed creates internal feedback loops and turbulence, making it difficult to prove that the equations will not encounter an infinite value or mathematical breakdown.

• Standard physics equations yield direct values, whereas differential equations calculate rates of change over continuous time.
• Viscosity functions as an internal brake within fluids, counteracting turbulence and vortex stretching.
• OpenAI's multi-agent computation consumed approximately 130 billion output tokens during its proof generation process.
• Charles Fefferman's formulation categorizes solution paths based on the presence or absence of external forces.
• Human expertise remains necessary to verify whether AI-generated logical steps correctly address the original physical question.

While AI systems can process massive datasets and generate lengthy formal proofs, the deep conceptual understanding required to integrate these results into broader scientific frameworks remains a human capability.

If AI systems can identify mathematical singularities under forced conditions, will they eventually resolve unforced natural systems without human guidance?