Curious minds gathered at Napper Tandy’s in Bay Shore, New York, on Nov. 19, 2024, for the latest installment of PubSci, a science café presented by the U.S. Department of Energy’s (DOE) Brookhaven National Laboratory. The event connects scientists and the community for a casual chat about the Lab’s research. This time, the conversation turned to artificial intelligence (AI) with the topic “Reimagining Scientific Discovery with AI.”
At Brookhaven Lab, scientists are advancing ways to use artificial intelligence as a research tool. Examples include AI for materials discovery, autonomous experimentation, data processing, brainstorming bots, virtual research assistants, and much more.
After the night kicked off with a toast to all who gathered to discuss this latest frontier in science, audience members fueled a lively discussion with their questions for three Brookhaven Lab experts: Carlos Soto, who leads the AI Theory and Security Group in the Computing and Data Sciences Directorate; Kevin Yager, who leads the Electronic Nanomaterials group at the Center for Functional Nanomaterials (CFN) — a DOE Office of Science user facility at Brookhaven; and Esther Tsai, a staff scientist at CFN working at the intersection of measurement science and data methods.
Among the many questions, guests wondered about the reliability of information collected and output by AI tools, how AI models are developed for science, the amount of time AI technology saves researchers, and whether bots can teach themselves. Throughout the discussion, the panel of scientists made connections and comparisons to AI tools community members may encounter on a day-to-day basis, such as Google Maps and virtual assistants, like Siri or Alexa, but noted major distinctions and considerations that go into AI for science.
“We are all in the space of trying to adapt AI methods to science,” Yager said. “One of the things we work on is adapting AI methods to leverage trusted science artifacts, like vetted publications or software. As we learn how to do that correctly, that same methodology can be applied by other people working on general chatbots for the public.”
This includes, for example, training research-specialized chatbots using vetted scientific papers and appropriately attributed data sources. Those tools employ generative AI models, which produce content based on patterns from provided data.
“One potential vision for the future is to integrate different AI models, each trained to carry out a specific task, into a single ecosystem of AI agents that will accelerate science workflows,” Yager said.
Tsai is using generative AI to create an assistant that can operate complicated instruments on behalf of researchers. Still, Tsai noted, it’s important that humans aren’t totally outsourcing all their decisions to computers.
“We should still try to learn things and be curious,” Tsai said.
Beyond exploring possibilities for AI assistants, Brookhaven scientists have applied AI methods to run autonomous experiments — where AI steers advanced instruments and adapts on the fly. This technology was recently used, for example, to discover new nanostructures. There are rigorous methods that help determine what achieves success in these experiments, Yager noted.
Machine learning — key to handling huge amounts of data — also falls under the AI umbrella. It can help scientists dive further into patterns and nuances in data, which can be difficult for humans to interpret on their own.
“Most of AI in science doesn’t look anything like ChatGPT,” Soto said. He added, “Data compression and data filtering with AI is a very capable and very useful application.”
That’s especially relevant for a facility like the Scientific Data and Computing Center (SDCC) at Brookhaven Lab. The SDCC recently hit a major milestone — storage of more than 300 petabytes (equal to about seven million movies) of important data from experiments at the Relativistic Heavy Ion Collider (RHIC), a DOE Office of Science user facility for nuclear physics research at Brookhaven Lab and the ATLAS experiment at the Large Hadron Collider (LHC), located at CERN, the European Organization for Nuclear Research.
“We see AI as another tool and another resource for getting more value out of that data that we already have,” Soto said.
As the conversation came to an end, the crowd heard straight from the scientists what it’s like to work in a rapidly evolving area.
Tsai emphasized that it’s up to humans to decide how to use AI technology.
“I’m very excited about having this great tool that can really accelerate science discovery,” Tsai said.
For Yager, the field of AI is exhilarating and fun.
“Sometimes we build solutions that use a model, but then a new model comes out that’s better, and you swap in that new model, and everything just gets better, your whole workflow,” he said. “It’s exciting to be able to take advantage of these accelerations in real time.”
Soto agreed, adding, “The rate of progress and the capabilities that are getting into the hands of more people who can do interesting and powerful things with these methods is incredibly exciting.”
Since 2014, PubSci has offered the Long Island community a chance to see a casual side of the cutting-edge research happening every day at Brookhaven Lab and chat with the Lab’s scientists over a drink. The series hops around Long Island covering different topics from the Big Bang to tomorrow’s technologies.
Stay tuned for the next PubSci and sign up to be alerted to the next event!
Brookhaven National Laboratory is supported by the Office of Science of the U.S. Department of Energy. The Office of Science is the single largest supporter of basic research in the physical sciences in the United States and is working to address some of the most pressing challenges of our time. For more information, visit science.energy.gov
Follow @BrookhavenLab on social media. Find us on Instagram, LinkedIn, X, and Facebook.
