Skip to main content

Artificial intelligence research must remain an American priority

ORNL’s Frontier, the first and fastest exascale supercomputer in the world, will accelerate advances in AI, computational sciences, HPC and data analysis. Credit: Carlos Jones/ORNL, U.S. Dept. of Energy

By Prasanna Balaprakash, Director of AI Programs, Oak Ridge National Laboratory

In every scientific field, researchers draw conclusions by managing the generation, organization and analysis of data — whether that information comes from computer simulations or experiments.

Artificial Intelligence, or AI, is an essential tool for streamlining this discovery process. It is revolutionizing the domains of scientific discovery and complex systems, experimental facilities and national security. Advancements in AI enable scientists to work with far larger datasets and do so far more quickly than ever before.

Two of the most promising AI techniques are machine learning, in which sophisticated algorithms learn from data to ensure accurate decision-making, and deep learning, which is a subset of machine learning that focuses on training artificial neural networks to learn about and understand complex patterns and representations in data.

These tools help make sense of modern research projects, which tend to generate so much data that learning complex patterns and extracting insights from the results would be impossible without AI.

Experts at the Department of Energy’s national laboratories are working with university and industry partners to improve this process by reducing costs and times to solution. And as AI systems become increasingly integral to critical decision-making processes and more deeply integrated into research workflows, ensuring their reliability and accuracy is more important than ever.

Prasanna Balaprakash portrait
Prasanna Balaprakash. Credit: Genevieve Martin/ORNL, U.S. Dept. of Energy

Establishing safe and trustworthy AI systems will enable researchers to pursue scientific discovery while earning the confidence of researchers, policymakers and members of the public by providing safeguards against incorrect and improper use. To address AI’s continually escalating computational demands, scientific institutions must also adopt energy-efficient methods that not only align with sustainability goals but also enhance the longevity and scalability of AI-driven accomplishments. By developing and strengthening the foundations of these three pillars of AI — safety, trustworthiness and energy-efficiency — we aim to ensure that any progress made in this promising field remains ethically and environmentally responsible. 

Oak Ridge National Laboratory, which I recently joined as director of AI Programs, launched a lab-wide AI Initiative in 2018 to accelerate AI Applications in fundamental and applied research and to help strengthen the status of the U.S. as a formidable scientific and economic competitor on the global stage.

The initiative encompasses three key thrust areas. The first, AI for Scientific Discovery and Complex Systems, seeks to utilize translational AI to accelerate scientific discoveries and model, optimize and control complex engineered systems. The second, AI for Experimental Facilities, aims to automate science workflows at experimental facilities by using AI-driven methods. The third, AI for National Security, aims to support national security efforts by developing foundational AI technologies required for advanced threat detection, decision-making and planning.

Specific objectives of this ongoing internal investment range from developing high-quality AI for enhancing national security infrastructure and improving energy distribution practices to designing innovative manufacturing processes and informing best practices for public health.

Accomplishing these goals requires collaboration among scientists to harness the power of ORNL’s eight DOE user facilities, which generate enormous amounts of data. That data can be analyzed in-house with high-performance computing, or HPC, resources provided by the lab’s Compute and Data Environment for Science and the Oak Ridge Leadership Computing Facility, a DOE Office of Science user facility and home of world-leading systems.

In 2022, ORNL launched Frontier as the world’s first exascale supercomputer. This energy-efficient HPE Cray EX system can complete calculations up to 10 times faster than its contemporaries and follows in the footsteps of ORNL’s Jaguar, Titan, and Summit systems — all of which were the fastest supercomputers of their time.

From modeling the lifespan of a nuclear reactor, to uncovering genetic components responsible for the manifestation of various diseases, Frontier will enable groundbreaking simulations and accelerate advances in AI, computational sciences, HPC and data analysis.

After 2 years as the world’s fastest supercomputer, Summit remains fifth on the TOP500 list. An IBM system, Summit’s architecture and memory bandwidth are optimized to integrate AI algorithms that quickly return accurate results.

Recent developments at ORNL have reinforced the importance of overcoming challenges and exploring opportunities related to AI. For example, the ORNL team that developed an AI algorithm called the Distributed Accelerated Semiring All-Pairs Shortest Path, or DSNAPSHOT, was a finalist for the Association for Computing Machinery’s annual Gordon Bell Prize, one of the most prestigious awards in the world of HPC. Part of ORNL’s Advances in Machine Learning to Improve Scientific Discovery at Exascale and Beyond project, DSNAPSHOT can quickly identify meaningful connections between data points buried in millions of medical documents. Two years later, running the first exascale graph-AI scientific demonstration on Frontier allowed the researchers to extract key connections between medical concepts related to COVID-19 and other coronaviruses, earning them another spot as a Gordon Bell finalist.

ORNL researchers were also responsible for creating AtomAI, an open-source software package that uses deep learning to assist with the characterization of complex materials, thereby helping extract information that would be difficult or impossible to obtain with traditional analysis techniques. This invaluable information can then be applied to optimize materials and facilitate the development of next-generation technologies.

Meanwhile, ORNL and the National Cancer Institute collaborated on a project called Modeling Outcomes using Surveillance Data and Scalable Artificial Intelligence for Cancer, or MOSSAIC, which focuses on using computational resources and deep learning algorithms to automate data extraction and harmonization for cancer research. Near real-time extraction of patient information allows for population-scale epidemiologic studies for cancer, rapid case ascertainment for clinical trials, and evaluation of the effectiveness of screening and treatment options. The overall goal of MOSSAIC research is to provide a library of predictive algorithms that will ultimately be used in the real world to improve patient outcomes.

Research efforts such as these are aligned with the recently released “AI for Science, Energy, and Security” report, which details the findings from a series of summer workshops at various national laboratories and outlines DOE’s vision for AI in the coming years. ORNL was a key contributor to the report and is positioned to play a leading role in many high-priority areas, including autonomous laboratories and exascale computing.

Through policy prioritization, upgraded facilities and other improvements on the horizon, we must continue to expand the role that safe, trustworthy and energy-efficient AI plays in America’s scientific and technological pursuits to improve the world around us by meeting or even surpassing our most ambitious objectives.

To get there, we must establish guardrails that ensure the accuracy and reliability of next-generation AI models, improve the algorithms and infrastructure that enable the validation and verification of these models throughout their lifetimes, and maintain scientific integrity by protecting the data used to train them.

And by pioneering an integrated approach that embraces these elements, we will be better equipped than ever before to exploit the full potential of AI capabilities aimed at advancing scientific discovery and improving national security.

Only then will the science fiction of today become the science fact of tomorrow.

To learn more about how ORNL is accelerating scientific discovery and national security with artificial intelligence, please visit the AI Initiative website.