Nervous about artificial intelligence? You’re not alone: half of Americans say now they are more concerned than excited about the ways AI is changing everyday life. But most people surveyed also recognize that AI is changing some things for the better. Take science: With its ability to detect never-before-seen patterns in complex systems, AI is accelerating the search for everything from new drugs to new galaxies.
We spoke to researchers from the University of California who are using AI to advance health, energy, agriculture and meteorology. It turns out they’re just as skeptical of AI as anyone else – and that’s actually a good thing. Even as they push the boundaries of research and discovery, UC scientists are asking thoughtful questions about the transformation this technology brings: Can we trust AI to make high-stakes decisions? How can we ensure that its benefits outweigh its risks? And how can we ensure that human wisdom and well-being remain central to scientific progress?
Why extreme weather is difficult for AI to predict
Every time you look at a weather app, you’re tapping into one of humanity’s most impressive technological feats: a global network of instruments that collect hundreds of billions of observations every day and send the data to supercomputers, which take hours to solve trillions of equations that approximate the physics that govern all atmospheric and oceanic dynamics on Earth.
All this makes weather forecasting one of the most computationally demanding tasks that humanity routinely performs, says Ashesh Chattopadhyay, an applied mathematician at UC Santa Cruz. Chattopadhyay’s research aims to use AI to more accurately see further into the future, using a fraction of the time, energy and computing power of current methods.
Using supercomputers at the UC-managed Lawrence Berkeley National Laboratory, Chattopadhyay worked with NVIDIA, CalTech and Rice University to develop FourCastNet, the first AI that can be on par with traditional prediction methods for accuracy and range. Just as ChatGPT takes text and images from the Internet to suggest a response to a user’s prompt, FourCastNet looks at the weather over the past 40 years and predicts what will happen next.
Because it doesn’t solve trillions of equations from scratch every time, AI can generate a prediction in seconds instead of hours, using thousands to hundreds of thousands of times less computing power. The venerable European Center for Medium-Range Weather Forecasts now uses FourCastNet and similar tools from Google and Huawei in its daily operations.
But recent research from Chattopadhyay’s group, which includes collaborators at the University of Chicago and NYU, suggests it’s probably too early to hand predictions over entirely to AI.
“AI works great for the daily weather in Houston, for example,” says Chattopadhyay. “But what about when Houston is faced with something never seen in history, like Hurricane Harvey?” The 2017 storm dumped more than five feet of rain in parts of South Texas, a once-in-two-millennium event.
“The fact that such a storm could happen is hardwired into the physics of the system, so the traditional physics-based models predicted it, and that’s the beauty of it,” Chattopadhyay says. If an AI model had been trained only on data from 40 years ago, would it have been able to predict Harvey?
To find out, Chattopadhyay’s group trained a version of FourCastNet using decades of weather observations, but filtered out any hurricanes stronger than a Category 2. Then they fed it an atmospheric condition that they knew would generate a Category 5 hurricane within a few days. The AI model clocked the storm but seriously underestimated its intensity and predicted it would reach a Category 2.
“We found that it couldn’t really extrapolate beyond what it had seen in the training data,” says Chattopadhyay. “Despite how good these models are with routine weather, getting the extremes right is still a problem. And those extremes are actually what scientists and forecasters care about most.”
The research highlighted an important consideration for meteorologists integrating AI into their forecasts, and pointed Chattopadhyay and his colleagues to their next task: improving AI forecasts. Now they are experimenting with integrating algorithms designed to model longer-term climate trends into the shorter-term weather forecasting pipeline, an approach that looks to boost AI’s ability to predict dangerous, unprecedented storms.
More from UC Santa Cruz: AI is good at weather forecasting. Can it predict bizarre weather conditions?
