Taming Chaos in Physical Systems Through Artificial Intelligence

Artificial intelligence (AI) has become ubiquitous in day-to-day life. From predictive text to virtual assistants to video games, AI is now embedded in most technologies we use. Its impact on research, though, is yet to be seen.

Lauren B. Dachs Professor of Science and Society Tsampikos Kottos and researchers from five other universities aim to explore that impact. Researchers will attempt to create a physics-based generative AI platform, referred to colloquially as “Physics-GPT.” The purpose of the platform would be to control chaos by, paradoxically, introducing a bit of randomness to the systems.

To develop Physics-GPT, Kottos and his collaborators received a five-year, $9 million Multi-University Research Initiative (MURI) grant from the Department of Defense. The research team consists of researchers from Wesleyan, Yale University, the University of Maryland, Arizona State University, Virginia Tech, and University of Michigan. Kottos’ lab will receive $1.2 million to fund his portion of the research.

“The optimal goal of this type of grant is to provide a new scientific paradigm, which is going to potentially alter our way of thinking,” Kottos said.

Mathematical models have shown evidence that complex systems—like weather or ecosystems—can be tamed by introducing tailored randomness. On the other hand, the use of AI allows for the mining of information encrypted in “big data” generated by complex systems. These data can then be used for the design of complex, seemingly disordered interventions for taming the chaotic evolution of these systems.

“On many occasions people are connecting chaos with noise and loss of information. It is not wrong; in some framework it is fair,” Kottos said. “Our working philosophy is, however, different. We want to embrace chaos and make it work for us.”

First, the researchers will analyze data generated by idealized mathematical models to build machine learning networks and create a universal model; then this model will begin to train itself off datasets generated by physical systems. Once the basic ingredients of the model are formed, it can begin to be applied to different physical realities, Kottos said.

At Wesleyan, Kottos and his group of undergraduate students, graduate students, and postdocs will begin testing the model with radio frequencies. Then, once the model is producing the right results, it will move on to researchers from University of Maryland for microwave frequency testing, then Yale for optical platforms, and so on.

Alongside the potential for fundamental knowledge, he said this novel approach could help synchronize systems of many lasers to act effectively as a single high-powered laser, and improve direct wireless energy transfer for powering remote systems. It could also have an impact on indoor wireless communications, robotics, and medical imaging, Kottos said.

“Everyday life is chaotic and does not seem to follow any rules,” Kottos said. “But there must be some form of order that up to now has eluded us. The availability of big data and the new tool of AI can help us unveil many of the laws that dictate such mysteries.”

Alongside his work in controlling chaos, Kottos also studies how system transformations affect the interactions between waves and matter for a variety of applications like sensors, which are crucial in areas like telecommunications, environmental research, and avionics. This year he received a nearly $500,000 continuation grant from the Simons Foundation to design new types of structures for different functionalities, like hypersensitive sensors, cavities that readily absorb heat, and optical isolators, which allow light to pass through one way but not the other.