Advancing the Next Generation in Scientific Discovery

“Artificial Intelligence (AI) presents myriad opportunities,” says Marin Soljačić, professor of physics at MIT and an expert in nanophotonics—a subfield of nanotechnology that deals with light. Over the past decade, powerful AI techniques have been developed for a variety of applications, such as self-driving cars, language translation, and image recognition, but they are only now starting to be substantially applied in science.

“AI can expedite the discoveries of new materials, new medicines, and potentially even new laws of physics,” Soljačić says. “Moreover, one can think of large parts of modern machine learning as essentially being complex nonlinear dynamics systems. We have many tools in physics suitable for analysis of such systems. So, physics could help us understand machine learning better, and this understanding could help to build even better AI.”

The potential societal implications of AI in science are vast; Soljačić’s specialty of nanophotonics is a good example of that. “Nanophotonics is one of the few fields of science where you can simulate experiments nearly exactly on a computer,” he says. Utilizing AI allows scientists to more efficiently design optimal nanophotonic materials, which in turn allows them to tailor the laws of physics, as they relate to light—to create materials with entirely new optical properties. “Among other applications, AI could enable us to build more efficient solar cells at a lower cost, implement a better electrical grid, and produce enhanced power storage. These advances would help us move further away from fossil sources of electricity, reduce our CO2 emissions, and positively impact climate change,” Soljačić says.

However, a technology as powerful as AI that carries such large potential benefits also carries substantial risks. Soljačić stresses that to fully understand all the possibilities and perils, we need collaboration among experts in many different fields—social sciences, medicine, computer science, technology, economics, natural sciences, and so on. “There are many organizations that have specialists focused on one or a few aspects of this, but not many entities have experts from all of these disciplines like we do at MIT,” he says. “Thus, we at MIT not only have a unique opportunity but also a unique responsibility to help figure out the right way to progress with AI.”


Computing Intelligence