PART II: : WINNING THE TECHNOLOGY COMPETITION
Chapter 11: Accelerating AI Innovation
Following is a summary of Part II, Chapter 11 of the National Security Commission on Artificial Intelligence's final report. Use the below links to access a PDF version of the Chapter, Blueprint for Action, and the Commission's Full Report.
To remain the world’s leader in artificial intelligence (AI), the U.S. government must renew its commitment to investing in America’s national strength: innovation.
This will require making substantial new investments in AI R&D and establishing a national AI research infrastructure that democratizes access to the resources that fuel AI. Members of Congress must come to terms with the reality that tens of billions of dollars will be needed over the next several years. The return on these investments will transform America’s economy, society, and national security.
A lack of national urgency is dangerous at a time when underlying weaknesses have emerged in our AI ecosystem that impair innovation and when viewed against the backdrop of China’s state directed AI progress.
The development of AI in the United States is concentrated in fewer organizations in fewer geographic regions pursuing fewer research pathways. Commercial agendas are dictating the future of AI and concentrating heavily in one discipline: machine learning (ML).1 Despite promising moves, government funding has lagged behind the transformative potential of the field, limiting its ability to shape research toward the public good and support progress across a range of AI disciplines.2 As a result, the AI innovation environment rests on a narrowing foundation.
Ingenuity, Not Access, Should Be the Key to AI Innovation in America
The consolidation of the AI industry threatens U.S. technological competitiveness in five ways.
The growing divide between “haves” and “have nots” in AI will exacerbate the well-documented lack of diversity in the field,4 limiting the field’s collective ability to build equitable, inclusive systems.
The rising cost of developing cutting-edge ML models and high likelihood of acquisition by leading technology companies means AI startups have narrowing paths to growth in the United States.5 Lack of competition undermines the industry’s ability to innovate and be globally competitive in the research and development of AI.
More than 90% of U.S. innovation sector job creation occurred in just five major coastal cities between 2005 and 2017.6 This divergence concentrates gains from technological progress in just a few regions and misses out on latent innovation potential in the rest of the country.
The federal government holds the responsibility to reverse these trends. It must step in and step up to provide strategic direction and sustained resources, as both a funder and consumer of technology.7 It must break the mold of standard scientific research funding. The outcomes of technology innovation, which generate greatest value when translated into fieldable solutions, are driven by multi-sector contributions and a culture of risk acceptance. The status quo at federal agencies and research entities is insufficient to make these big bets and propel promising technology concepts from laboratory to field.
A passive national approach that relies too heavily on the private sector to drive innovation and determine research agendas—and that presumes commercial innovation can simply “spin-in” to become government applications—will not win this strategic competition, nor will it fully capitalize on the transformative potential of AI. The United States—through government leadership in partnership with industry and academia—must increase the diversity, competitiveness, and accessibility of its AI innovation environment. That begins with a substantial infusion of new R&D dollars.
The United States Should
Scale and coordinate federal AI R&D funding.
A bold, integrated push for long-term investments in AI R&D will foster nationwide AI innovation and drive breakthroughs. An infusion of sustained resources, guided by a comprehensive strategy and distributed through a diversity of mechanisms, will enable U.S. researchers to push the boundaries of the field by supporting a wide range of AI approaches and novel applications of AI to other fields.
Establish a National Technology Foundation (NTF).
A new, independent organization would complement successful existing organizations, such as the National Science Foundation (NSF) and The Defense Advanced Research Projects Agency (DARPA), by providing the means to more aggressively move science into engineering.
The NTF would drive technology progress at a national level by focusing on generating value at intermediate levels of technical maturity, prioritizing use-inspired concepts, establishing infrastructure for experimentation and testing, and supporting commercialization of successful outcomes. This requires an organization that is structured to accept higher levels of risk and empowered to make big bets on innovative ideas and people.
Increase federal funding for non-defense AI R&D at compounding levels, doubling annually to reach $32 billion per year by Fiscal Year 2026.
This would bring AI spending to a level near to federal spending on biomedical research.8 Overall, the government should spend at least 1% of GDP on R&D to reinforce a base of innovation across scientific fields.9 Additional funding should strengthen basic and applied research, expand fellowship programs, support research infrastructure, and cover several agencies, with an emphasis on:
- National Technology Foundation (proposed)
- Department of Energy
- National Science Foundation
- National Institutes of Health
- National Institute of Standards and Technology
- National Aeronautics and Space Administration
Prioritize funding for key areas of AI R&D.
Amplified federal funding should prioritize AI R&D investments in areas critical to advance technology that will underpin future national security and economic stability, supporting areas that may not receive significant private-sector investment. Coordinated through the newly established National AI Initiative,10 investments should reflect a portfolio approach, focused on advancing basic science, solving specific challenge problems, and facilitating commercialization breakthroughs.
Triple the number of National AI Research Institutes.
The government should triple the current number of federally funded national AI research institutes across a range of regions and research areas.11 This would increase training and research opportunities for students and academic faculty, national lab researchers, and non-profit research organizations.
Invest in talent that will transform the field.
In parallel, NSF or the proposed NTF should invest in top AI researchers and interdisciplinary teams, launching grant awards that make big bets on the people and the out-of-the-box ideas that could lead to unexpected breakthroughs.
Expand access to AI resources through a National AI Research Infrastructure.
Democratized access to compute environments, data, and testing facilities will provide researchers beyond leading industry players and elite universities the ability to pursue progress on the cutting edge of AI. It will strengthen the foundation of American AI innovation by supporting more equitable growth of the field, expanding AI expertise across the country, and applying AI to a broader range of fields.
This national infrastructure should have these main elements.
Leverage both sides of the public-private partnership.
U.S. leadership in technologies like AI depends upon closer public-private collaboration and a shared sense of responsibility for U.S. global competitiveness.
The government should:
Create markets for AI and other strategic technologies. The application of AI across government agencies can save taxpayer dollars and improve the quality of public services. Some applications can be adopted directly from the private sector, while others are unique to the government mission. By accelerating AI adoption across federal agencies, the government can drive additional commercial investment in AI applications that benefit national security and the public good.12
Form a network of regional innovation clusters focused on strategic emerging technologies. The government should designate regional innovation clusters focused on strategic emerging technologies like AI to foster the growth of small companies in sectors that are critical to overall U.S. competitiveness. Established through a competitive bid process, the clusters would offer participants from industry and academia tax incentives, research grants, and access to federal R&D resources.
The private sector should:
Privately fund an AI competitiveness consortium. Private firms should establish a non-profit organization with $1 billion in funding over five years to broaden AI research opportunities and support AI skills and education. This donation-funded organization would focus on fostering economic opportunity through resources for AI research and AI skills training. Such corporate social responsibility spending to promote AI education and entrepreneurship would help bridge the gap between digital “haves” and “have nots.”
Tackle some of humanity’s biggest challenges.
“By focusing on solving real human problems that impact the lives of millions of people, we can build a new raison d’etre for the triangular alliance of government, academia, and industry ...”
1 A 2020 analysis of arXiv papers on AI found private-sector basic AI research to be thematically narrower than the broader corpus of AI publications, focusing on deep learning and computational infrastructure to support deep learning. Furthermore, the study found that elite academic institutions that collaborate more closely with industry had a similar narrowing of thematic concentration, leading to a tilting of the U.S. AI research environment away from the diversity still preserved in other countries. Joel Klinger, et al., A Narrowing of AI Research?, ArXiv (Nov. 18, 2020), https://arxiv.org/pdf/2009.10385.pdf. Increasing specialization of hardware achieved through industry investments has further prioritized commercial use cases, making it costly to pursue approaches outside the mainstream. Sara Hooker, The Hardware Lottery, ArXiv (Sept. 21, 2020), https://arxiv.org/pdf/2009.06489.pdf. 2 The Trump Administration’s proposed budget for non-defense AI R&D in Fiscal Year 2021 was $1.5 billion, a growth from the just under $1 billion spent in Fiscal Year 2020. The Networking & Information Technology Research & Development Program, Supplement to The President’s FY2021 Budget, National Science & Technology Council at 4, 12 (Aug. 14, 2020), https://www.nitrd.gov/pubs/FY2021-NITRD-Supplement.pdf. The National AI Initiative Act of 2020 included in the National Defense Authorization Act for 2021 included authorization for additional investments in AI R&D at the National Science Foundation (NSF), Department of Energy (DOE), National Institute of Standards and Technology (NIST), and the National Oceanic and Atmospheric Administration (NOAA). See Pub. L. 116-283, William M. (Mac) Thornberry National Defense Authorization Act for Fiscal Year 2021, 134 Stat. 3388 (2021).
3 A recent study found that from 2004 to 2018, 131 AI professors left universities for industry and 90 adopted a dual affiliation while maintaining part-time positions at a university. The study also documented the adverse effect that these departures had on AI startups of students from these universities. Michael Gofman & Zhao Jin, Artificial Intelligence, Education, and Entrepreneurship, SSRN at 2 (Oct. 26, 2020), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3449440. High salaries in the commercial sector pull researchers from academic tracks—in 2019, 57% of AI/ML PhD graduates in North America went to industry versus staying in academia for post-doc, research, or faculty appointments. Stuart Zweben & Betsy Bizot, 2019 Taulbee Survey, Computing Research Association at 11 (May 2020), https://cra.org/wp-content/uploads/2020/05/2019-Taulbee-Survey.pdf.
4 The annual Taulbee study that tracks the field of computer science (CS) found that women make up 21.0% of CS bachelor degree graduates and 20.3% of CS doctoral graduates, and domestic underrepresented minorities make up 14.7% of CS bachelor degree graduates and only 3.1% of doctoral graduates. Stuart Zweben & Betsy Bizot, 2019 Taulbee Survey, Computing Research Association at 4, 5, 22 (May 2020), https://cra.org/wp-content/uploads/2020/05/2019-Taulbee-Survey.pdf. A trend toward narrowing participation in the field holds the potential to worsen this state. See Nur Ahmed & Muntasir Wahed, The De-democratization of AI: Deep Learning and the Compute Divide in Artificial Intelligence Research, ArXiv (Oct. 22, 2020), https://arxiv.org/abs/2010.15581.
5 For example, non-elite universities and AI startups have difficulty affording the cost of compute resources and data for training sophisticated ML models. Nur Ahmed & Muntasir Wahed, The De-democratization of AI: Deep Learning and the Compute Divide in Artificial Intelligence Research, arXiv (Oct. 22, 2020), https://arxiv.org/abs/2010.15581. Ninety percent of Silicon Valley AI startups were purchased by large technology companies between 2013 and 2018. Ryan Kottenstette, Silicon Valley Companies Are Undermining the Impact of Artificial Intelligence, TechCrunch (March 15, 2018), https://techcrunch.com/2018/03/15/silicon-valley-companies-are-undermining-the-impact-of-artificial-intelligence/. These same companies dominate U.S. patent lists, excluding adoption patents. Al AuYeung, Who is Winning the AI Race?, IPWatchdog (Feb. 1, 2020), https://www.ipwatchdog.com/2020/02/01/winning-ai-race/id=118431.
6 Specifically, Seattle, Boston, San Francisco, San Diego, and San Jose. Robert D. Atkinson, et al., The Case for Growth Centers: How to Spread Tech Innovation Across America, Brookings (Dec. 9, 2019), https://www.brookings.edu/research/growth-centers-how-to-spread-tech-innovation-across-america/.
7 The NSF and other government agencies are doing admirable work, with the resources available, to encourage diverse research and create economies of scale for AI innovation, but they will not produce a strategic effect at the current level of effort, which is set against the backdrop of an overall decline in federal investment in R&D. Other notable recent federal initiatives include DARPA’s Artificial Intelligence Exploration Program, which fast tracks funding for awards up to $1 million to explore feasibility of new AI concepts within an 18-month timeframe; and NSF’s National AI Research Institutes effort, which in 2020 funded seven multi-institution, university-based research institutes at $4 million per year for five years and plans to launch another eight in 2021. Artificial Intelligence at NSF, NSF (Aug. 26, 2020), https://www.nsf.gov/cise/ai.jsp. 8 Funding for the National Institutes of Health (NIH) has grown from $30 billion in 2010 to $41 billion in 2020. NIH Budget History: NIH Budget Mechanism Detail, NIH Data Book (Oct. 2019), https://report.nih.gov/nihdatabook/category/1; Budget, NIH (June 29, 2020), https://www.nih.gov/about-nih/what-we-do/budget.
9 In 1953, the U.S. spent 0.72% of its GDP on R&D. In 1957, when the then-Soviet Union launched Sputnik, it had grown to 1.3%. R&D spending peaked at 1.86% in 1964. In 2017, it declined below 1953 levels to 0.61%. Federal R&D Budget Dashboard, American Association for the Advancement of Science (last accessed Jan. 14, 2021), https://www.aaas.org/programs/r-d-budget-and-policy/federal-rd-budget-dashboard.
10 The National AI Initiative Act of 2020 included in the National Defense Authorization Act for Fiscal Year 2021 creates a structure for a more strategic approach to harnessing AI through establishment of a National AI Initiative Office within the Office of Science and Technology Policy and associated advisory group and interagency construct. See Pub. L. 116-283, William M. (Mac) Thornberry National Defense Authorization Act for Fiscal Year 2021, 134 Stat. 3388 (2021).
11 The NSF awarded grants for the first national AI research institutes in 2020, supporting seven university-based, multi-institution consortia organized around fundamental and applied areas of AI research, and plans to fund a second round of institutes in 2021, coordinating support not only with interagency partners but also with private sector stakeholders to launch eight additional institutes. Artificial Intelligence at NSF, NSF (Aug. 26, 2020), https://www.nsf.gov/cise/ai.jsp. 12 Congress took an important step in the Consolidated Appropriations Act, 2021 by calling on the General Services Administration to create a five-year program to be known as the ‘‘AI Center of Excellence’’ (AI CoE) to “facilitate the adoption of artificial intelligence technologies in the Federal Government,” among other duties. The AI CoE can help bridge discrete efforts across federal agencies to create a sizable market for government-specific AI applications. See Rules Committee Print 116-68, Text of the House Amendment to Senate Amendment to H.R. 133, U.S. House Committee on Rules at 378-81 (Dec. 11, 2020), https://rules.house.gov/sites/democrats.rules.house.gov/files/BILLS-116HR133SA-RCP-116-68.pdf (referring specifically to section 103 of the Consolidated Appropriations Act, 2021). In addition, the Defense Innovation Unit (DIU) is playing a role in creating markets at the intersection of AI and other strategic technologies through its project-based approach. Focus areas include AI applications for space systems, advanced diagnostics, semiconductors/advanced hardware, and other critical technologies identified by NSCAI in Chapter 16 of this report. DIU’s experience indicates that creating a market for strategic technologies begins with the Department of Defense (DoD) and other government agencies pursuing an approach that is (a) contractually flexible, (b) aligned with firms’ technological development plans, and (c) generating financial incentives through opportunities to scale production. DIU Making Transformative Impact Five Years In, DoD (Aug. 27, 2020), https://www.defense.gov/Explore/News/Article/Article/2327021/diu-making-transformative-impact-five-years-in/.