NITRD Releases a 2020-2024 AI R&D Progress Report

Earlier this week, the Artificial Intelligence (AI) Research and Development (R&D) Interagency Working Group, under the Networking & information Technology Research & Development (NITRD) program produced a progress report on federal efforts surrounding AI R&D in the last four years. The report paints an overall picture of the investments and progress made in the federal AI R&D community.

The new document is centered on agency programs that coincide with the nine objectives listed in the National AI R&D Strategic Plan: 2023.

Prior to covering some of those agency initiatives, let’s first look at the federal AI R&D budget since FY 2018. Note that Core AI are those programs with a primary focus on AI R&D, whereas AI Crosscut are those with a primary focus on areas other than AI but with which AI is involved.

 

Source: NITRD

Note: FY 2018-2022 are Actual figures

Despite an ongoing increase in the AI R&D budget from FY 2018-2022, particularly propelled by the pandemic and AI being a common priority over multiple Administrations, budgets have remained somewhat stagnant for AI R&D in the last three years. This is especially peculiar with the onset of generative AI within this time frame. Looking at the FY 2024 request, NIH represented 30% of Core AI with $926M, followed by the NSF at 25% and $757M, and DARPA at 10% with $322M.

NITRD does not provide an explanation for the recently stalled AI R&D budget figures. Taking a look at the program’s AI Dashboard, it does not appear that any particular agency faced major “swings” or tradeoffs in budgetary numbers to cause the standing totals. Presumably, it may be due to the ongoing testing R&D initiatives not reaching or able to scale yet. It remains to be seen what the FY 2025 budget request numbers, expected in the Fall time frame based on NITRD budget supplement releases from last several years.  

Back to the progress report. Below are select agency investments under each strategic objective to provide an overview of the federal AI R&D space.

Strategy 1: Make Long-Term Investments in Fundamental and Responsible AI Research

• Census Bureau: The agency’s Data Science Training Program is testing the use of automation to compare satellite imagery from two points in time to identify changes in house units in preparation of the 2030 Census.
• DARPA: Under the DAPRA Air Combat Evolution program, the agency is testing software in F-16 aircraft to demonstrate AI agent control of a full-scale fighter jet.
• FBI: The agency is using AI algorithms in the Threat Intake Processing System (TIPS) Database to identity, prioritize and process actionable tips.

Strategy 2: Develop Effective Methods for Human-AI Collaboration

• DOD: The Autonomous Air Combat Operation program is testing AI-driven autonomy for airborne tactical platforms to ultimately develop an advanced AI-driven autopilot with aviation and navigation functionalities.
• NNSA: The agency is piloting a real-time AI-assisted defect screening capability for material and part certification of weapons systems.
• USDA: The department’s AI Institute, Agricultural AI for Transforming Workforce and Decision Support (AgAID) is testing predictive AI models to assist farmers in predicting and responding to extreme weather.

Strategy 3: Understand and Address the Ethical, Legal, and Societal Implications of AI

• DOD: The department developed the Responsible AI Toolkit to identify, track and improve the alignment of AI projects by using tailored assessments and tools throughout the AI product lifecycle.
• DHS: The Science and Technology Directorate is exploring an understanding of how bias enters AI models, models’ decay, and countermeasures to harmful synthetic media.
• NASA: The agency is developing explainable machine learning algorithms to explain why anomalies and their predecessors in aviation data are identified.

Strategy 4: Ensure the Safety and Security of AI Systems

• DOD: The Systemic Testing of AI Image Recognition (STAR) program develops technologies to improve user trust in autonomous system test and operations. For example, an image classifier tool to assist testers to prove reliability of AI-based imagers.
• NIST: the U.S. AI Safety Institute at the agency supports tools to measure and improve AI safety and trustworthiness.
• NSF: Under the Secure and Trustworthy Cyberspace (SaTC) program, the agency supports AI system safety including detecting deepfakes and making AI systems less prone to hallucinations.

Strategy 5: Develop Shared Public Datasets and Environments for AI Training and Testing

• Energy: The Office of Science is using high-performance computing and large datasets from its laboratories to advance AI R&D.
• DHS: The Science and Technology Directorate is researching environments where users and access multiple clouds to train models to enable large-scale data analytics and AI on cybersecurity data.
• NIH: The Bridge to AI Program develops AI-ready data sets ranging from biomedical to behavioral research, setting the stage for AI in medicine.

Strategy 6: Measure and Evaluate AI Technologies Through Standards and Benchmarks

• NIJ: Justice’s research agency is exploring machine learning in its Review and Revalidation of the First Step Act Risk Assessment Tools to improve equitability and efficiency in the system.

Strategy 7: Better Understand the National AI R&D Workforce Needs

• VA: The department created the All Services Personnel and Institutional Readiness Engine (ASPIRE) program to assess and report on technical skills, particularly in AI R&D.

Strategy 8: Expand Public-Private Partnerships to Accelerate Advances in AI

• DOD: Through the SBIR and CRADA acquisition approaches, the department is pursuing partnerships to develop AI science and technology solutions for space.
• Energy: The department is working with private industry, academia and research institutes to create trustworthy generative AI models for science under the Trillion Parameter Consortium (TPC).
• USPTO: The AI and Emerging Technologies Partnership is a cooperative agreement between the agency and independent inventors, small businesses, nonprofits and others to promote awareness and inclusivity of AI efforts.

Strategy 9: Establish a Principled and Coordinated Approach to International Collaboration in AI Research

• DOD: The DOD is prioritizing ethical considerations and collaboration with international partners in its approach to developing and fielding military applications of AI through the Australia, United Kingdom, and United States (AUKUS) Partnership and the  Australia, United Kingdom, and United States (AUKUS) Partnership program.

Looking ahead, the report states that agencies with AI R&D investments, “…aim to advance the strategic priorities laid out in the 2023 Strategic Plan update, by fostering collaboration between federal agencies, industry and international partners, and the research community to ensure a seamless exchange of knowledge and expertise and translation of research breakthroughs into practical applications.” Moreover, nurturing an AI-skilled federal workforce and education and training will continue to be key focal areas to advance AI capabilities.