Trump Plan Highlights AI for Science

The AI Action Plan released by the White House last week sets several goals related to scientific research, including creating high-quality scientific datasets, increasing researcher access to large-scale computing power, and automating lab processes.

In response, the National Science Foundation has announced it will unveil several initiatives in the coming weeks, including the next phase of the National AI Research Resource and new funding for its AI Research Institutes. The plan recommends actions for other agencies, including the Department of Energy and National Institute of Standards and Technology, to support model testing and evaluation.

President Donald Trump directed Michael Kratsios, director of the White House Office of Science and Technology Policy, and other members of the administration to develop the plan in a January executive order. The plan alludes to priorities for a future update to the National AI R&D Strategic Plan, including prioritizing “theoretical, computational, and experimental research” to advance AI capabilities and “fundamental advancements” in AI interpretability, control, and robustness.

The plan states that “AI-enabled predictions are of little use if scientists cannot also increase the scale of experimentation” and directs DOE’s national labs to develop “automated cloud-enabled labs” in partnership with other agencies, research institutions, and private companies.

The plan was released alongside executive orders on exporting American AI technologies, promoting the buildout of data centers, and pushing for large language models to be ideologically neutral.

Incentives to share data

The plan also highlights the need for scientific datasets of “world-class” size and quality. It proposes requiring federally funded researchers to disclose non-proprietary, non-sensitive datasets used by AI models in their research and considering the “impact” of their previously released datasets during the grant application process.

The plan also recommends that the National Science and Technology Council set minimum data quality standards for scientific data to be used in AI model training. It also looks to bolster “controlled access” to restricted federal data, directing NSF and DOE to establish “secure compute environments,” as envisioned for NSF’s National Secure Data Service demonstration project.

Access to high-performance computing

The plan emphasizes the role of NSF’s National AI Research Resource in promoting open-source and open-weight models, which the plan says are “essential” for academic research and could become the global standard. The plan recommends building the foundations for “lean and sustainable NAIRR operations” that connect researchers and educators to AI resources, including through partnerships with the private sector and by improving the financial market for large-scale computing power, which is currently “far beyond the budgetary reach of most academics and many startups.”

After the plan was released, acting NSF Director Brian Stone announced that NSF would soon introduce the “next phase” of the NAIRR. The NAIRR pilot began in January 2024 and is planned to run for two years. NSF also announced $100 million in funding this week for five AI Research Institutes, including a new institute led by Cornell University focused on materials science. The AI Research Institute program was launched in 2020 and currently has 29 institutes, each funded at about $20 million over five years, according to the NSF website.

Stone also said that NSF would soon announce the creation of AI testbeds “to evaluate real-world AI systems with transparency and rigor” and a partnership to create large-language-model infrastructure to help “drive AI for science.”

The administration repeatedly refers to AI as a priority in its budget requests for science agencies for the upcoming year, even amid deep overall cuts. NSF’s budget request would add funding for AI in many of its research directorates, even though those same directorates would be cut anywhere from 40% to 75% in the request.

Standards development

NIST’s Center for AI Standards and Innovation also plays a significant role in the plan, including by supporting the development of the science of measuring and evaluating AI models with DOE and NSF, convening meetings for federal agencies and the research community to share best practices on AI evaluation, and publishing guidelines and resources for federal agencies to conduct evaluations of AI systems that align with their mission. (The U.S. AI Safety Institute, established in 2023, was recently renamed as CAISI at Trump’s direction.)

“Understanding how to measure and evaluate models is still an open scientific question,” Kratsios said at an event discussing the action plan, hosted by the Center for Strategic and International Studies. “Currently, the state-of-the-art is literally having a biologist ask questions to the model and see what comes back… but is that scalable? Is that something you could standardize across the entire world and create some sort of coherent, multi-country understanding of what the best practices are for doing evaluations? All those are open metrology questions.”

Kratsios later added that Congress now has the opportunity “to think about how to legislate on the standards institute and give it, sort of, statutory cover for some of the actions that we want to be doing long term.”

Workforce programs

The action plan also recommends NSF and DOE support workforce development in AI, including by expanding hands-on research training and development opportunities for undergraduate and graduate students, creating industry-driven training programs to address workforce needs, and coordinating an AI hackathon “to solicit the best and brightest from U.S. academia to test AI systems for transparency, effectiveness, use control, and security vulnerabilities.”