U.S. agencies and policymakers lack mechanisms to assess their strategic weaknesses and opportunities in technologies that are critical to national security (DARPA and DOD at NASEM, 2019) and competitiveness. The U.S. government also lacks data on domestic and international production capabilities, including the long chain of base and intermediate suppliers supporting the production of final goods. This affects the resilience of our supply chains and contributes to our reliance on other nations for products, increasing the vulnerability of that reliance. Finally, central to comparative technology and production capabilities is an understanding of a nation’s human capital, and its growth, or lack thereof.
Understanding of human capital contributes to more comprehensive insights into how knowledge and production capabilities can transform, pivot, and recombine. Without domestic information on knowledge, human capital, and production capabilities, it is difficult to evaluate whether private incentives are sufficiently misaligned with public needs to justify interventions, or what types of investments might transform those dynamics.
We seek to demonstrate and enhance timely situational awareness of global technology capabilities, countries’ relative position therein, the distribution of domestic technology capabilities, and the national technology funding portfolio as it relates thereto. We will achieve this by leveraging existing research activities and datasets dispersed across the nation, and utilizing the analytic advantages offered by machine learning and natural language processing.
Global situational awareness
Experts: James Evans (U Chicago), Dashung Wang (Northwestern), YY Ahn (Indiana U.), and Britta Glennon (U. Penn)
The effort on global situational awareness of knowledge, production, and human capital capabilities brings together leading academic researchers in the areas of science and network science, with leading academics on global invention. In the first year, the academic component of the network will focus on mapping comparative advantage in national scientific production to funding structures, and on how U.S. multinationals can play a role in global invention and knowledge transfer. This area of research effort will be supported by Dewey Murdick of Georgetown University’s Center for Security and Emerging Technology (CSET), with an emphasis on timely global situational awareness of critical technologies and their applications and methods to accelerate processes for turning ideas into capability-transforming applications. With its ongoing analysis, CSET will explore collaboration opportunities and solicit feedback on public-facing data interactives to help critical technology network members better understand the global emerging technology landscape. Matching the network, CSET’s focus will be on AI/ML and synthetic biology (see letter of collaboration and the Other Resources section). A key assessment outcome of the one-year pilot will be identifying novel insights possible for U.S. competitiveness and needs for investment. We will compare and merge datasets from existing efforts across the network on global situational awareness of science, invention, human capital, and funding.
Global situational awareness of science
In their recent work, Miao, Murray, Jung, Larivière, Sugimoto, and Ahn have created a map of revealed comparative advantages of national scientific production that underpins national scientific development (Miao et al., 2022). In the one-year pilot, Ahn and Sugimoto propose to build a much higher-resolution (at the level of individual papers rather than disciplines) map of comparative advantage of knowledge by leveraging the "Science Genome” neural representation learning framework for mapping the space of knowledge that they and their co-authors have been developing.
They will utilize large-scale publication and funding acknowledgment records from each country. The successful one-year research will produce a preliminary "heat-map" of knowledge where each country's revealed comparative advantage in publication and funding can be visualized and quantitatively analyzed. Glennon and colleagues add a causal component to the work by Miao et al. Glennon, Murciano-Goroff, and Xiao will investigate:
- global heterogeneity in the significance of, and the areas of science supported by, foreign funding
- the causal impact on science when foreign funding is interrupted
Tensions between the U.S. and China throughout the 2010s supply an empirical context for observing how interruptions to critical flows of foreign funding impact science more broadly. In addition to outcomes, the two research efforts will learn from each other through differences in datasets, results, methods, and report lessons.
Evans will build directly on the work of Ahn and Sugimoto with a predictive model of how different national and global research funding landscapes impact the emergence and trajectory of novel research ideas (Rzhetsky et al., 2015; Shi, Foster, and Evans, 2015; Foster, Rzhetsky, and Evans, 2015). Evans will assess the direction and the risk distribution of national research portfolios, both from funding announcements, funding attributed by papers, and papers themselves. He will draw on computational approaches to manifold learning and formalisms from the field of statistical mechanics to construct an analogous “energy landscape” over the global space of scientific ideas, specifically ideas supported within national funding portfolios (Ziembowicz, 2013).
In this framing, research with “high energy” is improbable, novel, interdisciplinary, and potentially disruptive. By constructing an energy landscape for the global system of scientific ideas, his model will correlate with existing measures of novelty, interdisciplinarity, and disruption, helping to predictively model where new papers can be expected to emerge in the space of ideas and what their reception, downstream relevance, and multi-dimensional impact will be (e.g., Xu, Wu, and Evans,2022; Wu, Wang, and Evans, 2019).
Evans will suggest policy and portfolio interventions to help diversify U.S. productive scientific exploration. He will also use these insights to evaluate the portfolio strategy of nations, including patterns of risk, innovative return, focus, and direction in the scientific space. These evaluations will enable prediction of future national research attention and advances.
Situational awareness of invention
In recent decades, leading multinational firms have dramatically expanded the number of locations in which they do R&D (Branstetter et al., 2018). This presents a puzzle: R&D is a knowledge-intensive activity, but local knowledge sources in non-traditional locations are often far from the technological frontier. How are multinational firms able to effectively leverage human capital in these expansive R&D networks?
Branstetter, Glennon, and Jensen find that foreign affiliate research teams have become much more integrated with the multinational R&D network. In addition, there has been a pronounced shift away from the primacy of dyadic patterns linking foreign affiliates with the U.S. home base and toward a true network structure of co-invention, in which foreign affiliates collaborate and co-invent with one another. They also find that R&D-performing foreign affiliates are more likely to produce impactful research when they are more central within the MNC’s larger network. They argue that these results represent a strategic and dynamic process providing foreign inventors with complementary knowledge that enhances their inventive productivity and enables them to become active contributors to the multinational’s global innovation effort.
Glennon, Branstetter and Jensen define the local knowledge stocks of a country as a USPTO patent stock—with depreciation—in particular technology areas, and use this construction to compare over time (1) different countries in the same technology space, and (2) different technologies in the same country. Graphs which label technologies based on patent classes can be easily updated each year, and could be modified to look at specific technology areas. Already it is possible to see that countries like the UK have fallen behind in a number of areas in which they previously led, while countries like China have moved to the technological frontier in some areas (but not all). (For example, 1980s-1990s, the UK, Canada, France, Germany, and Japan were in the 90th percentile in electricity patent stocks; by 2010, China, Taiwan, Korea, Germany, and Japan were the countries in the 90th percent in terms of electricity patent stocks.)
In the upcoming year, they will deepen their analysis of country comparative knowledge stocks in different patent classes, particularly in the four deep dives (AI, semiconductors, biotech, and energy storage) being pursued by the network. They will expand our analysis to include country comparisons of knowledge stocks in other patent systems, in particular, the European Patent Office, to expanding this analysis to include the U.S. in the country comparisons. (The current analysis focuses on knowledge transfer from U.S. multinationals, but due to the U.S. naturally being the largest creator of patents in the USPTO, the team needs to use other patent systems to assess U.S. comparative advantage versus these countries.) They will also begin to explore algorithms to timely and dynamically update these classifications.
Finally, they will produce a set of statistics and graphics on what, where, and who foreign governments fund. As the research in the network proceeds, the team will attempt to learn across situational awareness in science and in invention to understand opportunities to accelerate the emergence and diffusion of disruptive technologies, and what barriers might be. For example, connected to the research effort by Evans on composition and risk of national research portfolios, Bloom, Hassan, Kalyani, Lerner, and Tahoun (Bloom et al., 2021) have employed textual analysis of patents, job postings, and earnings calls to document the diffusion of disruptive technologies across U.S. firms and labor markets. In later years, the network will attempt to exploit these synergies to advance our understanding of the transition from scientific discovery to commercialization of new technologies, and the bottlenecks thereto.
Similarly, our national network will leverage and learn from Stanford's AI Index as a possible motif for merging data across science and technology indicators in other critical technologies. The Stanford AI Index conducts novel research, measuring and evaluating the rapid rate of AI advancement (via a cross-sector lens) from research and development to technical performance, ethics, economy, education, and governance. Stanford’s annual report is widely read by academic researchers, industry leaders, and policymakers across the globe.
For this proposal, the Stanford AI index will continue and scale up this work, allowing the index to broaden the landscape of assessing U.S. AI competitiveness and raising the situational awareness of this critical technology. In particular, this proposal would support additional research on measuring public investment and funding gaps in AI across the globe, as well as quantifying the geopolitical competition of developing foundation models (e.g. technical performance, talent capacity) to evaluate U.S. competitiveness.
Domestic situational awareness
In addition to lacking timely situational awareness of global knowledge, human capital, and production capabilities and the U.S. standing therein, the U.S. government also lacks situational awareness of its public and private funding portfolio with sufficient technology and geographic specificity to inform science and technology investments. Situational awareness with sufficient technological specificity and dynamism may provide valuable insights into funding gaps and inequalities (whether geographic or race), in participation in science and technology activities and funding. As explored in the global situational awareness activities, analysis of the funding portfolio offers new insights into how and to what extent portfolio redundancies enhance the rate, direction, novelty, and diffusion of innovation, and where there may be diminishing returns or even negative effects.
S&T activities in terms of people and data, places and institutions
Expert: Julia Lane
The Democratizing Data platform will build on existing work (Romer and Lane, 2022; Lane et al., 2022; Chang et al., 2022, The Coleridge Initiative, 2021; NCSES & The Coleridge Initiative, 2021) that characterizes fields by the interconnections of written work (full text publications), datasets, and the people and networks engaged in the production of science and technology.
For the purposes of the network, the Democratizing Data team will focus on applying the approach to characterize S&T research in semiconductors (and possibly AI) and expanding the focus to include the funding source, the institution, and the state of the research producers. The team will draw on the expertise of other network members for publication information (Sugimoto, Wang, and Murdick), and for domain expertise (Fuchs for semiconductors and Branstetter and Brynjolfson for AI). Data would be provided to the NCSES so that they can calibrate the results with existing information at NCSES. It would be structured so that future links to UMETRICS could be used to understand the “missing millions” (Ross et al., 2022).
The goal will be to present, at the end of one year, the domestic distribution of people and institutions engaged in semiconductor research, and how U.S. investments may be able to accelerate the engagement of more individuals in these fields, increase their geographical, institutional and demographic diversity, and advance science and technology in critical fields.
Private funding of science and technology
Experts: Josh Lerner (Harvard) and Joshua Graff-Zivin (UCSD)
There is a long history of efforts attempting to bring together the U.S. national funding portfolio across agencies. Researchers in this network hope to improve the technological specificity and dynamism of technology classifications, and to contribute to the statistical effort to link datasets across agencies in a way that informs the nation of its national funding portfolio. While government funding and national science and innovation ecosystems are critical to the generation and commercialization of ideas, less is known about the direction and role of science and technology funding by private actors.
The network will focus activities on two sources of private funding: philanthropic entities (which primarily fund science) and venture capital (which primarily fund technology development and commercialization). These sources of funding differ importantly in the stages of research that they typically support and in their mechanisms of allocation, making analyses of lessons across the two types of private funding activities particularly valuable.