Using Artificial Intelligence to Analyze Government Grant Data and Reveal Science Frontiers
President Trump challenged the Director of the Office of Science and Technology Policy (OSTP), Michael Kratsios, to ensure that scientific progress and technological innovation fuel economic growth and improve the lives of Americans. Much of this progress comes from federal research grants, which include detailed plans for pioneering scientific research. These applications outline hypotheses, data collection methods, experiments, and approaches that lead to discoveries, patents, data, and advances. Collectively, they form a blueprint for future innovation.
Artificial Intelligence (AI) now enables the creation of powerful tools to refine how research funding is awarded. AI can provide new insights into future discoveries and research needs, influencing both public and private investment decisions and accelerating the application of federally funded research. We recommend that OSTP lead a multiagency effort to apply AI analysis to grant applications. This would help predict scientific trends, improve peer review, and guide investment decisions. Agencies involved should include all members of the National Science and Technology Council (NSTC).
Challenge and Opportunity
The federal government funds about 100,000 research awards annually, covering all science fields. The volume is so large that reviewing these in depth is almost impossible. For example, spending just 10 minutes per award would require over eight years of full-time work to review just one year's funded projects. Additionally, for each funded award, 4–12 applications are rejected, adding to the data volume. Analyzing all this by human effort alone is unfeasible.
Fortunately, emerging AI can process these documents at scale, maintaining confidentiality and protecting trade secrets while generating useful summaries. These AI tools could also support the grant review process. The NIH's Research, Condition, and Disease Categorization (RCDC) system, launched in 2009, is a precedent. It automatically assigns NIH-funded projects to spending categories, replacing manual categorization and saving about $30 million annually in staff time.
The NIH developed digital “fingerprints” of grant applications by text-mining titles, abstracts, and specific aims. These fingerprints match scientific terms to assign categories. NIH staff then expanded this to track research products and match reviewers to applications by expertise. NIH’s RePORTER website offers the Matchmaker tool, which helps applicants find similar funded projects and relevant review panels.
We suggest that all federal agencies collaborate to broaden and deepen AI use, leveraging all available data for more insightful analyses.
Use Cases
Use Case 1: Funder Support
Federal staff can use AI analytics to identify research opportunities and guide funding priorities. Agencies must consider not only an application's merit but also how it fits existing portfolios and goals. Scientific questions often span multiple portfolios—for example, a protein affecting different organ systems. AI can help reveal such connections and ensure balanced investments across research stages. It can also highlight underserved areas, support deliberate replication, and reduce unintentional duplication.
Use Case 2: Reviewer Support
AI can provide application reviewers with objective context about how a proposal relates to currently funded projects. Reviewers often rely on memory to compare applications, but AI can offer consistent, data-driven insights to better inform funding decisions.
Use Case 3: Grant Applicant Support
Applicants could receive AI-generated context comparing their ideas to both funded and rejected applications, while protecting confidentiality. NIH’s Matchmaker tool is a good start but only covers funded projects. Enhanced AI tools could summarize failed or in-progress applications, helping applicants understand funding likelihood and avoid repeated mistakes.
Use Case 4: Trend Mapping
AI can help scientists, biotech, pharma, and investors monitor emerging funding trends in near real-time. Current public summaries of federal awards are helpful but quickly outdated due to volume. AI can identify overlaps, gaps, and new opportunities, including unfunded projects, offering a clearer picture of research directions and potential underinvestment.
Use Case 5: Results Prediction Tools
Analytical AI could forecast the timing and topics of future research outcomes. For example, pharmaceutical development predicts clinical trial timelines using public data. AI could scale similar predictions across scientific fields, such as materials for solar cells or diagnostic technologies. Including data from failed applications could highlight where science is struggling or where investments are lacking.
Plan of Action and Leadership
OSTP should lead a multiagency development effort to apply AI analysis comprehensively to grant applications. This will improve science forecasting, peer review, and research funding decisions. Agencies should include all NSTC members. Collaboration with private sector AI experts and a broad range of stakeholders is essential since the data is government-owned and the benefits span public and private sectors. We foresee four stages in this effort.
Recommendation 1: Agency Development Pilots
Each agency should pilot select use cases to test AI training and output tools with their data. Agencies vary in how they store and handle application data and have distinct scientific focuses. Piloting will help optimize AI models and assess their impact, especially on peer review. Trend mapping and prediction tools can be validated using historical data subsets.
Agencies may need to upgrade data systems and adopt standardized formats so data can be combined across fields. Application sections used for AI should be machine-readable to maximize effectiveness.
Recommendation 2: Prizes and Public–Private Partnerships
OSTP should coordinate with private sector organizations to clarify the impact of opening both funded and rejected applications to AI analysis. Collaboration could include prizes for developing accurate tools and user-friendly interfaces. These challenges would use test data from existing applications and their research outputs, with winners selected by transparent criteria.
Conclusion
Research grant applications are a largely untapped resource. They detail work plans linked to specific research outputs, many of which are already indexed and machine-readable. Using these data alongside AI promises to improve funding decisions and reveal emerging scientific trends with accuracy approaching expert human foresight. However, the data is currently closed to AI innovation. With coordinated leadership and resources, federal science agencies can unlock the full potential of these applications, benefiting both public and private research sectors.
Your membership also unlocks:
