We all know that artificial intelligence (AI) is dramatically changing how we interact with computers and with each other. But AI still needs guidance from experts to be useful. Joining the brain power of more than 1,000 leading experts with some of the top AI models, a first-of-its-kind event promises to supercharge AI even more. 

The “1,000 Scientist AI Jam Session,” held Feb. 28 at nine national laboratories, including Pacific Northwest National Laboratory (PNNL), provided a focused, one-day effort where researchers could put the most advanced available AI models through their paces by asking them complex scientific questions.

“We brought together scientists and engineers to assess how well new reasoning models from frontier AI companies advance science and national security,” said Court Corley, chief scientist for artificial intelligence at PNNL and director of the Center for AI. “This experience will help us understand gaps and opportunities. Then we can work with other national labs and frontier AI companies to develop AI capabilities that meet nationally strategic goals. That way, we’ll transform and advance science, energy, and national security faster.”

OpenAI and Anthropic

In this first of several planned AI Jams, researchers used frontier AI models from the companies OpenAI and Anthropic to seek answers to problems in their respective scientific and mission domains, evaluate model responses, and share feedback to improve future versions of AI systems so that they develop with scientists’ needs in mind. Future sessions will feature models from different AI companies.

Findings from the event will be shared in a follow-up report about how AI models can be used by the scientific community. In addition to PNNL, participating labs included Argonne, Berkeley, Brookhaven, Idaho, Livermore, Los Alamos, Oak Ridge, and Princeton Plasma Physics. U.S. Secretary of Energy Chris Wright joined the event from Oak Ridge with OpenAI president and co-founder Greg Brockman.

At PNNL, more than 100 scientists gathered in Discovery Hall, huddling over laptop computers for the day interacting with the models in focused sequential inquiries on topics ranging from cybersecurity and power grid optimization, to difficult chemical conversion problems, to how to most efficiently select new soil microbe sampling techniques. 

Changing the world through AI

“AI has the potential to dramatically accelerate the pace of scientific work, including background research, forming hypotheses, automating experiments, improving theory, assisting in computer code development, and dozens of other tasks,” said Wendy Shaw, the associate laboratory director for physical and computational sciences at PNNL. “National labs are ideally suited to tap the power and potential of AI.”

“The national labs are just the gem of research in the United States and really around the world,” Shaw added. “Going forward, it will be interesting to see the outcomes of having all of us working together to solve problems because there's so much intellectual brain mass in the national labs.”

Shaw, whose research career has been focused on chemistry, was a Jam participant.

“I'm working on bio-inspired catalysis,” she said. “I'm trying to understand how something called an enzyme works and how we can take the features of that, which are faster—more efficient—and put them in synthetic catalysts to provide better industrial-based catalysts. A better, bio-inspired catalyst could potentially reduce the large use of energy in the chemical industry.”

Ample opportunity to partner with industry

Corley, one of the primary PNNL AI Jam organizers, said the Jam could help define how national laboratories approach deploying AI to accomplish their mission. And, Corley pointed out, the Jam underscored the need to partner with industry to use and improve the best U.S.-developed AI models.

“We are working with the best models from AI companies to make science advances and see where more research on these models is needed,” Corley said. “Beyond this event, it’s important to have public–private partnerships that leverage the national labs’ deep domain science and energy knowledge. This is what’s going to really drive AI dominance and U.S. innovation.”

Corley was one of four panelists on a closed-circuit video conversation about the future of AI, joining colleagues at Argonne, Oak Ridge, and Lawrence Livermore in a conversation with Alex Carney of OpenAI. The group discussed AI’s potential to accelerate scientific discoveries, improve health care outcomes, enhance business productivity, and address complex global challenges.

A Jam participant, Neeraj Kumar, a chief data scientist in the Advanced Computing, Mathematics, and Data Division, saw great potential for AI and reasoning in autonomous experimentation, coding, and the chemistry/biology domain.

“We work on mission-driven problems at PNNL,” said Kumar. “We solve problems that are critical for humanity. The idea here is to understand how we can make these models more efficient to solve these problems.”

There’s a second time for everything

“This is just the beginning,” said Nancy Washton, a PNNL chemist helping guide AI literacy for PNNL’s Center for AI. “First, it’s amazing that nine Department of Energy labs came together. I have never seen anything quite like this. The next time we have something like this, and there will be a next time, I hope the turnout is tenfold greater. This is our reality. As a scientist, I can do five times as much working with AI. It's not just about productivity. It’s about a tool that gives us superpowers, cognitive superpowers that I could never have imagined.”

Washton and others chuckled about spontaneous visceral reactions from colleagues throughout the Discovery Hall meeting room as the performance of the AI tools surprised normally reserved scientists.

Excitement in the room

Perhaps none was more demonstrative than Bob Runkle, the division director for physical detection systems and deployment in the National Security Directorate, leaping to his feet to show Nathan Hodas, data scientist and lead of PNNL’s Generative AI Initiative, the result of his research.

“This day won over a lot of AI skeptics,” said Hodas.

Runkle said he was stunned at the quality of data dispensed by AI tools at the Jam. Previous AI experiences left him unimpressed. 

“I'm glad that I came here today because as a manager I was wondering, what is the benefit to me?” Runkle said. “But now I have tangible examples that I can take back to my team and say, ‘Even I did this, as a manager. As a researcher, just imagine what you could do.’”