Regulating the Future

April 2019 - TradeSecure

Concerns over Defining “Emerging Technologies”

On 19 November 2018, BIS published an Advanced Notice on Proposed Rulemaking (ANPRM) seeking public comment on criteria for identifying “emerging technologies” that are essential to U.S. national security in order to develop export controls on presently uncontrolled technologies. This process is being undertaken pursuant to Section 1758 of the Export Control Reform Act of 2018 (ECRA), which mandates the establishment of a regular multi-agency process for identifying appropriate controls on emerging and foundational technologies that are “essential to the national security of the United States” and that are currently subject to no or very limited controls under other existing U.S. export control regimes.[1] This is exercise also supports requirements in the Foreign Investment Risk Review Modernization Act (FIRRMA), which was enacted with ECRA under the National Defense Authorization Act (NDAA) of 2018.

The ANPRM included fourteen broad representative categories of technology from which BIS seeks to determine whether if and which emerging technologies are important to U.S. national security for which effective export controls should be implemented:

  1. Biotechnology
  2. Artificial intelligence (AI) and machine learning
  3. Position, Navigation, and Timing technology
  4. Microprocessor technology
  5. Advanced computing technology
  6. Data analytics technology
  7. Quantum information and sensing technology
  8. Logistics technology
  9. Additive manufacturing (e.g., 3D printing)
  10. Robotics
  11. Brain-computer interfaces
  12. Hypersonics
  13. Advanced materials
  14. Advanced surveillance technologies

The ANPRM also includes a list of representative examples of such technologies for each of the above categories (e.g., computer vision and national language processing within the AI and machine learning category).[2]

As an unfolding process, there is a dearth of working definitions, sufficient to inform policy in a salutary fashion, of emerging technologies – let alone how to control them. We were curious about the collective thinking on this complex topic. As such, we tried to distill the central themes and concerns across the ANPRM comments. What follows is a basic text analysis thereof.

Figure 1. Comment Contributor by Type

Contributors: The comment submission period ended in January 2019, with BIS publishing a final list of comments in early March. Overall, 232 individuals (51% or 118) and organizations (49% or 114) submitted comments (see Figure 1). Of the organizations, industry/trade associations, technology companies (e.g., semi-conductor and ICT) and universities contributed the majority of comments (~44%). Two foreign governments (France and Japan) also submitted comments.

Figure 2: Country References

Topics: Of the 232 comments submitted, the vast majority (~75%) addressed artificial intelligence (AI), followed by additive manufacturing, robotics, quantum computing, and biotechnology.

China was the only country mentioned to a statistically significant degree (China = 647 references, see Figure 2). The next most cited country was Japan (130), followed by Germany (62), Israel (56), and Russia (53).

In terms of the subject technologies listed in the ANPRM, the leading category was AI, at >45% of the references. Additive manufacturing, robotics, quantum computing, biotechnology, and data analytics represented nearly 80% of the ANPRM list technology references. One category, “Brain-Computer Interfaces” was conflated with AI and Machine Learning. On a related note, most references to “neural networks” – included as subcategories to both AI and Brain-Computer Interfaces – referred to AI as networks and not as “neural controlled.”

Beyond the technology topics, most submission focused on the following major themes:

  • Research (>1,250 contextual references);
  • Foreign availability (-305 contextual references); and
  • Criteria (-300 contextual references)

In both instances, sentiment analysis grouped both themes against concerns about overly restrictive (i.e., “broad” or “general”) controls. In this regard, over ~85% of the comments clustered around “showing concern” about the proposed list-building initiative. “Criteria” was heavily correlated with foreign availability and multilateral control parity. Indeed, sentiment analysis for “criteria” evinced a clear consensus on the exercise of caution in establishing control parameters given the dynamic nature of the technologies addressed.

Figure 3: ANPRM Technology Category Reference

Further text analysis highlighted a number of relationships between terms. For example, a collocation analysis revealed that ‘risk’ was highly collocated with ‘unilateral,’ ‘economic,’ and ‘research.’[3] (See Figure 4).[4] China, as country most cited, was highly collocated with ‘AI,’ ‘Japan,’ ‘Russia,’ and ‘research.’ ‘Collaboration,’ mentioned 243 times, was collocated with ‘development,’ ‘AI,’ and ‘research.’

Figure 4: Collocates of “Risk”

Conclusion: Since the inception of modern export controls, the fundamental question vexing regulators and industry concerns control criteria, as recently noted in ECRA’s admonition on stifling circumscriptions: “avoid negatively impacting U.S. leadership in the science, technology, engineering, and manufacturing sectors.” Predicated upon weapons of mass destruction (WMD) technologies and materials, the known technology universe was well understood and, as a technology class, evolved slowly. While there are civilian applications for maraging steel (e.g., exotic guitar strings), further applications will not revolutionize the economy in any way similar to AI’s impact. The emerging technology phyla, however, includes a range of exponentially developing technologies, defined in the business literature as “disruptive.” The “emerging” set is changing even as while we field comments for definitions. True to form, the dual-use aspect of any control list technology is always in its application as opposed to its constitutive nature. Unlike most of the current Commerce Control List (CCL) entries, the BIS starter kit of 14 technologies are wildly unrestrained in their applicability to current and new economic products and services, as well as to current and future weapons systems.

Apart from the above textual analysis, which merely conveys statistical portraits of clustered themes and word counts, a brute force reading of the comments does provide a more visceral listing of anxieties, complaints, and recommendations. The primary anxiety expressed in the comments concerns the negative economic and security impact of either overly restrictive, and/or futile controls. For example, overly restrictive controls could result in innovation off-shoring (or “innovation arbitrage”)[5], declining technology and ethical leadership, technology islanding or autarky, and a diminished technology reservoir from which the military could draw.

Complaints varied on a continuum regarding the USG approach to defining emerging technologies, arguing that the government should have started with very specific technologies rather than working from general categories. In other words, the onus should be on the government to establish why and how a technology is a national security threat a priori, not the other way around. One commentator succinctly captured this dilemma in the ANPRM process:

The ANPRM notes that, “Certain technologies, however, may not yet be listed on the CCL or controlled multilaterally because they are emerging technologies. As such, they have not yet been evaluated for their national security impacts.” These two sentences are at the heart of the problem of defining emerging technology within an export control framework. The uncertainties and ambiguities around emerging technology make them difficult if not impossible to govern from an export control perspective, and yet this is exactly what the process to be established through this ANPRM is tasked to do.

In terms of practical recommendations, the comments were generally of the catch-all variety, mentioning the use of the EAR’s pre-existing tool kit of non-listed and end-use and user controls: catch-all, Entity List, and 0Y521. The comments also concluded that the listing exercise would be utterly meaningless in the absence of multilateral consensus. There was almost no mention of specific control examples (e.g., “here’s how you can export control AI….”).

More so than in the recent export control past, the ANPRM has captured the fundamental anxiety of our age: technology governance. The earlier operating system of controls was constructed around a well-established threat, WMD, while emerging technology is rapidly and fundamentally rewriting our security and economic narratives in such ways that the narratives cannot practically congeal around a least common denominator that will define the constituent elements of “essential to the national security of the United States.”

[1] BIS also announced that it will issue a separate ANPRM in the future regarding the identification of “foundational technologies” that may be important to U.S. national security.
[2] Presumably, BIS developed the ANPRM representational list from Congressional testimony and other USG research. While the civilian applications for emerging technologies are relatively well established in the business literature (e.g., “disruptive technology”), there is decidedly les information about prospective “weapon” systems. For an illustrative example of the former, see McKinsey Global Institute, “Disruptive technologies: Advances that will transform life, business, and the global economy,” May 2013 < https://mck.co/2KqDe0z> and World Economic Forum Anja Kapersen, “8 emerging technologies transforming international security,” September 2015
[3] Collocation analysis examines sequences of words or terms which co-occur more often than would be expected by chance, or terms that appear more frequently in proximity to keywords across an entire corpus.
[4] Collates connected to risk are arrayed by distance, with close proximity indicating a high degree of collocation, likewise decreasing with distance.
[5] See, Adam Thierer, “Innovation Arbitrage, Technological Civil Disobedience & Spontaneous Deregulation,” 5 December 2016 < https://techliberation.com/2016/12/05/innovation-arbitrage-technological-civil-disobedience-spontaneous-deregulation/>.