Fight for Earth Health with AI | MIT News

For Priya Donti, a childhood trip to India was more than just an opportunity to visit a large family. The biennial journey inspired her and continued to influence her research and teaching.

Compared to her home in Massachusetts, Donti was shocked by the differences in people’s lifestyles. She is now a Silverman Family Career Development Professor in the Department of Electrical Engineering and Computer Science (EECS), a department that is co-studyed by the MIT Schwartzman School of Computing and EECS and is also the principal investigator at the MIT Laboratory of Information and Decision Systems (LIDS).

“I am very aware of the severity of inequality around the world,” Donty said. “From a young age, I knew I absolutely wanted to solve this problem.”

This motivation was further inspired by a high school biology teacher, whose curriculum focuses on climate and sustainability.

“We understand that this huge and important issue of climate change can exacerbate inequality,” Donty said. “It really made me unforgettable and caught my stomach on fire.”

So when Donty attended Harvey Mudd College, she believed she should focus on research in chemistry or materials science to make the next generation of solar panels.

However, these plans were abandoned. Donty “loved” computer science and then discovered the work of British researchers who believe that artificial intelligence and machine learning are crucial to help integrate renewable energy into the power grid.

“This is the first time I’ve seen these two interests come together,” she said. “I was fascinated and have been working on this topic ever since.”

While pursuing his PhD at Carnegie Mellon University, Donty was able to design degrees including computer science and public policy. In her research, she explores the need for fundamental algorithms and tools that can manage power grids that rely heavily on renewable energy on a large scale.

“I want to participate in the development of these algorithms and toolkits by creating new machine learning technologies based on computer science,” she said. “But I want to make sure that my way of working is based on the real-world energy systems and working with people in that field” to provide what is actually needed.

While studying for her PhD at Donty, she co-founded a nonprofit called Climate Change Artificial Intelligence. Her goal is to help people involved in climate and sustainable development — “whether they are computer scientists, academics, practitioners or policy makers” — come together to get resources, connect and educate “to help them on this journey.”

“In the climate field,” she said, “you need experts in the field of climate change-related, experts in different technical and social science toolkits, problem leaders, affected users, policy makers who understand regulations – all of which – to have a scalable impact on the ground.”

When Donti came to MIT in September 2023, she was attracted by MIT’s initiatives to guide computer science applications to solve the biggest problems of society, especially the current threat to the health of the planet.

“We are really thinking about where technology can have a longer-term impact and how technology, society and policy work together,” Donty said. “Technology is not only completed and monetized in just one year.”

Her work uses deep learning models to combine physical and hard constraints of power systems, leveraging renewable energy for better prediction, optimization, and control.

“Machine learning has been widely used in areas such as solar forecasting, which is a prerequisite for managing and balancing the power grid,” she said. “My focus is, in the face of a series of renewable energy that changes over time, how can I improve the algorithm that actually balances the grid?”

One of Donti’s breakthroughs is providing grid operators with a promising solution that can optimize costs while taking into account the actual physical situation of the grid, rather than relying on approximations. While the solution has not been deployed, it seems to work 10 times faster than previous technologies, and is much cheaper, and has attracted the attention of grid operators.

Another technology she is developing is designed to provide data that can be used to train machine learning systems to optimize power systems. Generally speaking, much of the data related to the system is private, either because it is proprietary or for security reasons. Donty and her research team are working to create comprehensive data and benchmarks, which Donty says “can help uncover some potential problems” and thus improve the efficiency of the power system.

“The question is,” Donti said, “can we take our datasets to a level that is enough to drive progress?”

Due to her efforts, Donti was awarded the U.S. Department of Energy Graduate Scholarship for Computational Sciences and the National Science Foundation Graduate Research Fellowship. She is considered MIT Technology ReviewThe 2021 “Innovators Under 35” list and Vox’s 2023 “The Future Perfect 50” list.

Next spring, Donti will teach a course called “Artificial Intelligence for Climate Action” with Assistant Professors at EECS, with a focus on AI for biodiversity and ecosystems; Abigail Bodner is an Assistant Professor at EECS and an Assistant Professor at Earth, Atmospheric and Planetary Sciences, with a focus on AI for climate and earth sciences.

“We are all very excited about it,” Donty said.

After coming to MIT, Donty said: “I knew there would be an ecosystem of people who really care about not only the success metrics like publications and citations, but also the impact our work has on society.”