Experts in Automated Science Convene at CMU

Experts in automated science from across the country came to Carnegie Mellon University Oct. 23-25 to attend "Creating a National Network of Academic Cloud and Self-Driving Labs," a National Science Foundation (NSF) workshop on the future of science.

The workshop, organized by the Office of the Vice President for Research and the Mellon College of Science, gathered stakeholders to address the short- and long-term priorities and transformative potential of cloud labs and self-driving labs. Cloud labs, like the soon to open lab at CMU, allow researchers to conduct experiments remotely and increase access to a wide variety of highly specialized tools. Self-driving labs use artificial intelligence to inform which experiments should be conducted; AI can be used to make cloud labs self-driving.

Workshop topics included open science policies, training the next generation of researchers, lowering barriers to access, and the technological infrastructure investments needed to support a national network of these cutting-edge technologies.

Theresa Mayer, CMU's vice president for research, said the group had the potential to provide input that will set the foundation for critical investments at a unique moment in time in automated science.

"We have brought together representatives from a broad section of the community, including university experts, industry leaders, end users, federal labs and agencies, to achieve the goals of the workshop," she said. "We want to think about our long-term goals in automated science, and also about our priorities in the near term, one to three years, to ensure the full promise is realized."

Mayer emphasized the potential of automated science to help address some of society's biggest challenges.

"To be able to harness the data collected and then to close the loop, to then leverage that data to drive the next round of experiments is critically important ... This should not just increase throughput, but allow us to think differently to solve problems that we currently cannot," she said.

In his keynote, Benji Maruyama of the U.S. Air Force Research Laboratory said that automated science could help researchers move more quickly on problems that can't wait, like global climate change.

"We need to move toward a teaming model between humans and autonomous robots where autonomous robots are tools and teammates to help us go faster in our human-led research," he said.

In several sessions, panelists spoke of the opportunity for automated science labs to learn from the success and challenges faced by other fields. Tom Kalil, the chief innovation officer of Schmidt Futures, cited the application of machine learning to protein folding and protein design as an example. 

"Advances like AlphaFold2 and RFDiffusion happened as a result of having a well-defined problem, a large open dataset, the Protein Data Bank and an agreed-upon benchmark for evaluating progress," Kalil said. "I think the research community should be trying to identify other areas where this combination of a 'dream dataset' and a benchmark could have a similarly transformative impact on the acceleration of scientific discovery."

James Barr von Oehsen, director of the Pittsburgh Supercomputing Center — a joint center of CMU and the University of Pittsburgh and a partner in the NSF cyberinfrastructure program — and a research professor of electrical and computer engineering at CMU, noted the automated science community needs to work closely with its user base to make sure they understand the benefits.

"Campuses already have experimental labs," he said. "All that equipment is out there, so you need to understand what's happening on campuses before building a national network."

 The workshop included a tour of a large-scale, automated lab — the Carnegie Mellon University Cloud Lab — based on the platform of the CMU-alumni founded Emerald Cloud Lab. The remote-controlled, automated lab will allow researchers at CMU, partner universities and the local research community to program experiments from anywhere in the world. Those experiments will be executed by robots and technicians on the more than 130 unique pieces of laboratory equipment contained in the lab.

The workshop was supported by an award from NSF's Technology, Innovation and Partnerships (TIP) Directorate. A report of its findings is forthcoming.