In science, there is always more to be learned. Scientists are often making new discoveries. But what would it be like if the time it took to discover these things was reduced by half?
Last December, engineers at Iowa State University received an award for a proposal furthering science through artificial intelligence (AI) granted from the Defense Advanced Research Projects Agency (DARPA), in the category of Artificial Intelligence Exploration (AIE).
The goal of the DARPA project, led by the Department of Electrical and Computer Engineering’s Professor Chinmay Hegde and the Department of Mechanical Engineering’s Assistant Professor Soumik Sarkar and Professor Baskar Ganapathysubramanian is “to get AI to the level where it can enable humans to do research.”
In the project, the team “will develop a new machine learning framework called Physics-Aware Learning for Microstructure Design (PALM).” The team specifically created the proposal for an application enabling the creation of new materials, like silicone or rubber, where the microscopic structure (or microstructure) plays an important part.
“We are trying to leverage advances in AI to discover new materials,” Hegde said. “That is the motivation behind our project.”
Their project will potentially enable the next wave of AI technologies.
“One of the key goals in engineering is to design systems (in this case materials) that achieve a desired property,” Ganapathysubramanian and Sarkar said. “In this project, we integrate classical engineering approaches (i.e., physics models) with sophisticated concepts in machine learning to transform and accelerate such design exploration process.”
Hegde explained that the process humans go through now when completing scientific research is similar to the way their AI will further scientific discoveries specifically related to materials.
“The idea that we are pursuing is to encode into AI the same principles that you would teach to a human scientist. AI would be able to learn and apply laws into the discovery process,” Hegde said. “The AI will be able to tell the human scientist what kinds of knobs or experience will be need to be changed, therefore accelerating the process of scientific discovery.”
This research project may not stretch far beyond the walls of the lab yet, but in the near future it could easily impact important aspects of the world, such as the environment and energy.
“The need for materials with optimized properties is universal,” Sarkar said. “Specifically, our PALM project seeks to improve electronic properties of materials with long-term implications for the energy security of our nation, primarily in the area of renewable energy generation.”
While the three engineers have been working together a lot on the research proposal, graduate students and postdocs have played a large role, Ganapathysubramanian and Sakar said.
“We have a very strong team of graduate students and postdocs involved in this project who are gaining multidisciplinary education and research experience crucial for the next generation workforce,” Ganapathysubramanian and Sakar said.
The first phase of this project revolves around demonstrating proof-of-concept of the idea, which is in effect now. Once phase one is complete, the team plans to begin applying the research on a greater scale.
Discovering new laws, substances or methods in science is a lot of trial and error. But with AI, the errors could be made significantly less often.
“When you come up with new materials, you go to the lab, synthesize a material and test its properties and then make it stronger if it is too weak, for example. The experiment could work, or you could have to go tweak it again,” Hegde said. “There is a scientific loop of discovery. The goal is to accelerate this loop, which we believe will be done by introducing AI. According to our proposal, AI will be matured to the level that it can enable humans to do better research.”