For over a century, mathematicians and physicists have wrestled with the chaotic nature of fluid movement — how air whirls around airplane wings or water churns in a pipe. Google's DeepMind lab recently made a significant breakthrough in this field, utilizing artificial intelligence.
While investors and others question whether AI will be worth the astronomical cost, it's reassuring to see DeepMind working on important stuff like this. This is an example of AI producing something of real value.
As a reminder, DeepMind is a pioneering AI research lab that was acquired by Google more than a decade ago. It's led by Demis Hassabis, a math and gaming whizz who has risen quickly up Google's ranks as AI has become more powerful and important.
I got a "G" grade in my first year of physics in high school (yes, you read that correctly). To explain the importance of DeepMind's recent discovery, I consulted my daughter, Nora Woolley, who's studying mechanical engineering and fluid dynamics! — at the University of Washington.
I shared DeepMind's blog and research paper with Nora, and she wrote back with her takeaways and some explanations for those of us who struggle with math and physics.
"This could have a big impact on fluid dynamics and physics as a whole," Nora told me.
Concentrate on this part
OK, on to the details, with Nora's help.
Fluids are so unpredictable that the equations used to model their behavior are impossible to solve completely. To use these equations, physicists must make assumptions such as constant viscosity or a smooth change in pressure.
Introducing even simple scenarios can lead to "blow-ups" where the equations predict extreme outcomes such as infinite pressure or an unbelievable surge in velocity. These are called singularities, and they represent the moments when math is no longer able to predict the physical behavior of fluids.
Singularities can either be stable or unstable. Stable singularities are easier to find, while unstable ones are far harder to pin down.
Unstable genius
Using machine learning and bespoke, physics-focused AI models, DeepMind researchers uncovered new families of unstable singularities across three distinct fluid-dynamics equations.
By embedding the structure of the equations directly into these specialized AI models and optimizing them in stages, the team achieved near machine-level precision, enough for mathematicians to verify the results formally.
"This work provides a new playbook for… tackling long-standing challenges in mathematical physics," the DeepMind researchers wrote in their research paper. An accompanying blog said, "This breakthrough represents a new way of doing mathematical research."
Why it matters
The discovery of these new unstable singularities may help scientists better understand how turbulence, the unpredictable and energy-draining behavior of fluids, happens in nature and engineering.
This unlocks a deeper understanding of areas such as aircraft drag, weather systems, blood flow, and energy distribution. Who knows, these discoveries might make future flights less bumpy.
"A lot of the software we use to monitor turbulence assumes that these equations are fully accurate across all values," Nora said.
Now that DeepMind has discovered new unstable singularities, scientists may be able to monitor flows better, "Because we have a better understanding of the ranges where these equations are valid," she added.
This is not AI curing cancer, but it's a lot better than the generative AI slop currently infecting the internet.
Sign up for BI's Tech Memo newsletter here. Reach out to me via email at abarr@bjinnox.com.
