Unstable genius: DeepMind cracks a century-old physics mystery with AI

For over a century, the chaotic behavior of fluid movement has puzzled mathematicians and physicists. The complexities of how air flows around airplane wings or how water moves through pipes have remained a significant challenge. Recently, Google’s DeepMind laboratory has made a remarkable advancement in this area by harnessing the power of artificial intelligence. Despite ongoing debates regarding the cost-effectiveness of AI, the work being done by DeepMind serves as a reassuring example of technology creating tangible benefits. This prominent AI research lab, acquired by Google over a decade ago and led by the talented Demis Hassabis, has been at the forefront of AI innovation, particularly as its applications become increasingly crucial. To gauge the significance of DeepMind’s latest findings, I consulted with my daughter, Nora Woolley, a mechanical engineering student specializing in fluid dynamics at the University of Washington. Sharing DeepMind's research paper and blog with her provided valuable insights into the implications of their discovery. Nora emphasized that this breakthrough could significantly influence the fields of fluid dynamics and physics. The unpredictability of fluids often leads to equations that cannot be completely solved. Physicists typically rely on simplifying assumptions, such as constant viscosity or smooth pressure changes. However, introducing even minor variations can cause "blow-ups" in these equations, predicting impossible scenarios like infinite pressure or extreme velocity surges. These unpredictable events are known as singularities, which can be either stable or unstable. While stable singularities are relatively easier to identify, unstable ones present a much greater challenge. Through the use of machine learning and specially designed AI models focused on physics, DeepMind researchers have identified new families of unstable singularities in three different fluid-dynamics equations. By integrating the structural elements of these equations into their tailored AI models and optimizing the process, the team achieved an impressive level of precision, enabling mathematicians to formally validate their findings. According to DeepMind's research paper, this work offers a fresh approach to addressing longstanding issues in mathematical physics. The accompanying blog post highlights the breakthrough as a novel method for conducting mathematical research. Understanding these newly discovered unstable singularities could enhance scientists’ grasp of turbulence—the unpredictable and energy-consuming behavior of fluids—in both natural and engineered contexts. This knowledge could lead to improvements in various applications, such as reducing aircraft drag, predicting weather patterns, managing blood flow, and optimizing energy distribution. With these insights, future flights may become smoother. Nora elaborated on the potential for better monitoring of "turbidity," a state where fluids behave more according to momentum than their physical properties, complicating predictions. She noted that many existing software tools rely on the assumption that the equations governing fluid behavior are always accurate, but with these new discoveries, scientists can refine their understanding of where these equations hold true. While this breakthrough may not be as revolutionary as curing cancer, it stands as a significant step forward compared to the less impactful generative AI content currently saturating the internet.