In the summer of 2023, Prasanna Balaprakash of the Department of Energy’s Oak Ridge National Laboratory was invited to speak at a roundtable discussion – hosted by the White House Office of Science and Technology Policy and the National Science Foundation – that focused on the importance of academic artificial intelligence research and development.
Later in the year, Balaprakash spoke at the Task Force on American Innovation’s briefings for U.S. Senate and House staffers. These briefings focused on the ways in which federal investments in mathematics, computer science, materials science and microelectronics have helped advance the development of AI in a variety of scientific fields.
The White House roundtable event, which included talks from several experts and a Q&A session between speakers and attendees, was part of a series of AI roundtables organized to help curate content for an upcoming strategic plan that will serve as a roadmap for future AI research in the U.S.
Balaprakash, who serves as director of ORNL’s AI programs, primarily focused on the safety, trustworthiness and energy efficiency of AI and its primary applications in scientific discovery and national security.
He explained the value of implementing guardrails while training AI models to ensure that these resources are used only for legitimate research endeavors and not being corrupted for nefarious purposes. Then, he emphasized the need to develop advanced algorithms and supporting infrastructure to sufficiently validate and verify these models’ decision-making capabilities throughout their entire lifecycle.
To determine how these increasingly large and complex black box models are making their predictions, he said it’s ideal to incorporate causal AI capabilities. He also recommended enabling AI safety and data model protections to guard sensitive data and maintain scientific integrity. In a similar vein, Balaprakash advocated for the advancement of federated learning approaches, which train models on decentralized data to preserve privacy and keep information secure as it’s moved from one location to another.
Balaprakash’s penultimate point, prioritizing the energy efficiency of these models, can be achieved by using training methods such as spiking neural networks, which are inspired by the inner workings of the human brain. Finally, he stressed the importance of DOE’s national network of supercomputers — including the Frontier exascale system located at ORNL’s Oak Ridge Leadership Computing Facility, a DOE Office of Science user facility — which empowers researchers to push the boundaries of AI and accelerate advancements in the field.
In addition to these topics, Balaprakash noted the need to expand workforce development and retention efforts. This expansion is one of many key aspects that require sustained and specific investments in AI techniques and technologies.
“With an integrated approach that focuses on safety, trustworthiness and energy efficiency, we will be better equipped to exploit AI’s capabilities for scientific discovery and national security,” Balaprakash said.
The Task Force on American Innovation briefings highlighted three major pillars of AI: large data, specialized hardware and advanced algorithms and software. Participants discussed success stories, including improvements to data storage technologies, as well as how systems such as Frontier showcase the capabilities of sophisticated AI algorithms and software resources.
Speakers also discussed how the convergence of various federally funded technologies have made significant contributions to the current state of AI. These advancements include the transformation of GPUs from standard gaming infrastructure to essential AI assets, the role of robust networking in distributed AI, improvements in search and indexing algorithms and innovations in energy management for sustainable AI growth.
Through these conversations, the participants identified key areas related to privacy, security and trust that they plan to prioritize in the coming years: differential privacy, secure multiparty computation, defense against adversarial attacks, explainability, robust authentication protocols and real-time anomaly detection. They also explored how quantitative changes in data handling, hardware speed, algorithms, storage, connectivity and scalability have led to qualitative improvements in AI capabilities, thereby enabling more complex and efficient applications.
In all of these contexts, Balaprakash reiterated the vital role that ORNL has played throughout the history of AI research. By enacting strategies such as diverse collaborations, iterative learning, infrastructure investment, open-source contributions and ethical guidelines, ORNL’s ongoing efforts — along with those of other U.S. institutions — push the boundaries of the field to ensure the U.S. remains competitive on the global stage.
“AI has the potential to help address global challenges such as climate change, so efforts to advocate for continued federal funding are important for maximizing power efficiency and harnessing AI’s full potential for societal benefit,” Balaprakash said.
ORNL has a rich tradition of AI research dating back more than four decades and garnering more than 10 patents. The laboratory’s AI Initiative is dedicated to ensuring safe, trustworthy and energy-efficient AI in the service of scientific research and national security.
UT-Battelle manages ORNL for DOE’s Office of Science, the single largest supporter of basic research in the physical sciences in the U.S. The Office of Science is working to address some of the most pressing challenges of our time. For more information, please visit https://energy.gov/science. — Elizabeth Rosenthal
