In ancient Greek mythology, the hero Theseus is famous for cleverly solving a labyrinth and defeating the Minotaur at its center. That story, as well as a famous maze experiment from the history of machine learning, is the inspiration for a new generative AI tool that helps researchers navigate the domain of computer modeling. 

As science, engineering, and other fields rely more and more on powerful computers to generate digital versions of molecules, machines, and other complex systems, researchers must devote significant time and resources to produce these simulations. With a more than $7 million grant from the Department of Energy, the Theseus research team aims to change that. 

“The AI tool we are developing is meant to reduce the time and labor that go into computational modeling, thus accelerating the pace of research and discovery. We envision an AI assistant that acts as a collaborator, suggesting and running simulations, and helping researchers accomplish what could normally take a year in a matter of weeks,” said Shaowu Pan, Ph.D., co-principal investigator on the grant and an assistant professor in the Department of Mechanical, Aerospace, and Nuclear Engineering at Rensselaer Polytechnic Institute. 

The three-year grant is led by Patrick Emami, Ph.D., of the National Renewable Energy Laboratory and includes co-principal investigators Sameera Horawalavithana, Ph.D., of the Pacific Northwest National Laboratory and Jason Eisner, Ph.D., of Johns Hopkins University. 

With his expertise in computational science, Pan will focus on testing and refining the applications of an advanced generative AI assistant in computational science. 

“Generative AI tools such as ChatGPT have laid the groundwork for this kind of specialized tool,” Pan explained. “Let’s say someone is designing a new kind of wind turbine that will generate more clean energy. To do this, the researcher needs to simulate their design on a computer. The Theseus assistant adds another layer of scientific intuition to that process, not only checking the code for bugs, but also making suggestions about possible hypotheses and pointing the user to relevant existing research. Ultimately, we want our AI model to enhance the work of computational scientists.”

Trained on a large body of studies, existing computer models, figures, images, and more, the tool could have broad implications for many disciplines, not just fluid dynamics and computational science, Pan said. 

“The transition from the slide rule to the calculator, and, more recently, the adoption of computer-assisted design, are two examples of technological advances that increased research productivity. We hope our generative AI assistants represent the next such leap and start a wave of revolution for the next decade,” Pan said.