I. Vision for AI Research
Artificial Intelligence, or AI, remains in an age of discovery. To create a pipeline of breakthroughs for tomorrow’s economy and security, we must deepen our commitment to AI research and engineering, as part of a reinvigoration of research and invention in the United States.
A common idea spans the work of Vannevar Bush, director of defense research and development during World War II, and decades of enterprises in Silicon Valley: science and technology can help build a better American future. In his letter to President Franklin D. Roosevelt in 1945, Bush justified this optimism: “Scientific progress is one essential key to our security as a nation, to our better health, to more jobs, to a higher standard of living, and to cultural progress.”[1]
In the decades ahead, future AI technologies have the potential to reshape the international security environment, change the character of armed conflict, and reshape the basis of our economic prosperity and military power.
Our ability to navigate this future can be increased by long-term thinking and investments today. We can anticipate future U.S. opportunities opened by today’s AI research:
- Smart cyber defense systems that imagine attacks that have not yet been seen, and adapt themselves to be resilient to them – evolving the cyber domain toward a defense-advantaged environment;
- Security and safety capabilities that can enable networked, intelligent systems to operate on an accountable, trustworthy basis;
- New sensor fusion systems that can leverage heterogeneous, proliferated sensors to provide earlier indications and warning for national and military leaders – as well as allow the U.S. and its allies to better track and defend against advanced missiles;
- Intelligent applications that assist medical doctors – accelerating diagnoses, revolutionizing emergency medical treatment in the field, and reducing medical errors; and
- Training systems that can enhance and personalize education for military officers over the span of their careers
This is not a race – races end, and we are strongest when we build on our strengths rather than focusing on narrow comparisons with others. In managing our efforts, our approach must focus on how AI technologies enhance our military, economic, and national advantages relative to authoritarian governments — not on how niche AI capabilities compare to those of others.
We must also focus on critical inputs to research and engineering competitiveness, like access to first-rate talent. Continuing to lead in scientific and technological progress requires growing the American science and technology base through STEM education programs, continuing to draw global talent to live and work in the U.S., and growing new technology collaborations with U.S. allies worldwide.
II. Our Approach
Our goals for AI research are simple:
- Future AI technologies will enhance the broad foundations of U.S. security, including our national military, economic, and scientific advantages.
- AI technologies will provide the U.S. with secure and safe tools.
We are focusing our R&D efforts into three areas:
Accelerating development of key AI applications for DOD. These applications can be specific, but collectively they span many areas, including: fusing and leveraging highly networked and heterogeneous sensors; AI-enhanced cyber and EW capabilities; multi-domain C2; AI applications for training, medicine, logistics, readiness, and anticipating and responding to major catastrophes; and AI applications to accelerate scientific discovery and invention. Our approach must manage uncertainty about how AI technologies, and choices by other governments, will shape the future of war – we will iteratively adapt our focus areas as we learn in the years ahead. In places where we already have clarity, we will pursue ambitious breakthroughs such as AI systems capable of producing Nobel-worthy scientific discoveries in biotechnology and biomedical sciences.
Creating a technological foundation for trustworthy AI technologies, especially trustworthy machine learning (ML) technologies. Modern AI and ML systems are currently brittle, can have unexpected behaviors that are unsafe, and present new software security risks. Despite these challenges, modern ML capabilities will be increasingly woven into our society, economy, and global military systems in the years ahead. We can look to past decades for a powerful lesson on what happens when security is not sufficiently prioritized: security was not a leading concern during the creation of information technology infrastructure, and today we face recurring cybersecurity challenges as a result. We must approach AI technologies differently, focusing on security early and broadly. To that end, we will grow our work substantially in AI security, safety, and resilience – and invest to mature AI engineering as a discipline. Our focus will particularly include trustworthy AI decision-making, basic research in machine understanding, secure use of AI technologies as part of larger integrated systems, and robust ML methods. This work will also provide a foundation for AI applications that continue to preserve human accountability in military operations.
Investing in infrastructure for AI research and development. The U.S. has structural advantages for AI R&D, as well as for national security application of AI research. The U.S. continues to have the world’s best research universities and research talent. The DOD has globally unparalleled access to militarily-relevant data. The U.S. has a global network of allies and partners with whom to collaborate on research initiatives; and we have the world’s best-trained military officers, who are best positioned to use AI technologies in new ways. Our efforts for the years ahead will focus on three areas first: (1) deepening our ability to leverage unique data and sensors that have been maintained by the U.S. national security community for decades – from classified datasets produced by advanced sensors to data from the Defense Health Agency that can drive new breakthroughs in American biomedical sciences, (2) Expanding AI R&D coordination and collaboration with U.S. allies, and (3) deepening our commitment to the American S&T talent base, primarily through fellowships and scholarships for U.S. STEM graduate students and early-career research faculty.
In developing the S&T base, we will prioritize growing AI engineering as a full engineering discipline with rigorous principles of analysis, design, and quality. We will also continue to improve mechanisms for the national security community to hire top talent to work on extraordinary problems for a period of time – through special appointments like Highly Qualified Experts (HQEs) and Intergovernmental Personnel Act (IPA) Assignments, and by building new bridges between fellowships, scholarships, and extraordinary Defense laboratories.
----
Dr. Matthew P. Daniels is Technical Director, Machine Learning and Artificial Intelligence, Office of the Undersecretary of Defense for Research and Engineering