Upon his confirmation as the Assistant Secretary of the Air Force (Acquisition, Technology & Logistics), Dr. Will Roper issued a memorandum to the acquisition workforce, proclaiming, “artificial intelligence (AI) will revolutionize warfare.”1 He stressed the importance of networking, data, and software in pioneering a new warfighting domain.2 While the efforts of the Air Force in air, space, and cyberspace have tightened the observe–orient–decide–act (OODA) loop, efforts in the new domain of AI “will likely draw this loop into a knot of unprecedented decision speed.”3
Doctor Roper is not alone in his vision of the near future of the military and the starring role AI will play. Artificial intelligence has been described by science and strategy experts as a revolutionary technology, changing the way wars are fought.4 The stand-up of the Army Futures Command, the announcement of the Joint Artificial Intelligence Center, the 2018 National Defense Strategy, and the Fiscal Year (FY) 2019 National Defense Authorization Act (NDAA) all demonstrate that the Department of Defense (DoD) is ready to enter into this new domain. It can be said, however, that the DoD’s realization of the importance of AI is late; the U.S.’s near-peer competitors have elevated AI to strategic priorities, with China and Russia as leaders in the field.5 Consequently, AI, along with autonomous weapons and robotics, is the focus on the DoD’s Third Offset Strategy to maintain a technological advantage over military capabilities of near-peer competitors.6 However, while the intent to build national security capabilities in AI is clear, questions remain as to how the DoD will meet its strategic objectives through its acquisition efforts. Acquisition attorneys can provide value to requiring activities and their contracting office by understanding the technical possibilities of AI, considering the ethical and legal implications of such acquisition, and knowing the acquisition tools available to meet these challenges.
What is AI?
Despite the recent attention and the DoD’s embrace of AI, the technology and application remains shrouded in misunderstanding and vague notions of HAL 9000 and the Terminator. There is no universally accepted definition of AI, though that is not for lack of trying.7 In the FY2019 NDAA, Congress tasked the Secretary of Defense to delineate a definition of the term “artificial intelligence” for use within the department.8 Congress provided working definitions of various forms of AI:
- Any artificial system that performs tasks under varying and unpredictable circumstances without significant human oversight, or that can learn from experience and improve performance when exposed to data sets.
- An artificial system developed in computer software, physical hardware, or other context that solves tasks requiring human-like perception, cognition, planning, learning, communication, or physical action.
- An artificial system designed to think or act like a human, including cognitive architectures and neural networks.
- A set of techniques, including machine learning, that is designed to approximate a cognitive task.
- An artificial system designed to act rationally, including an intelligent software agent or embodied robot that achieves goals using perception, planning, reasoning, learning, communicating, decision making, and acting.9
This broad definition covers many current research and development (R&D) projects underway throughout the DoD and other federal agencies. With many applications of AI ranging from the already ubiquitous Siri or Alexa, to predicting maintenance requirements of vehicles, to the revolutionary uses of autonomous weapon systems, AI—as we know it and as we foresee it developing —will likely permeate our everyday lives, and fundamentally change how the DoD operates. Artificial intelligence-enabled software is expected to be particularly helpful in intelligence, processing large amounts of data, such as video footage from a remotely piloted aircraft (RPA) to free human analysts to make decisions based on the data.10 Such work was the focus of the Algorithmic Warfare Cross Functional Team, known as Project Maven, a DoD partnership with Google’s AI team. This contract resulted in protests by Google employees, who opposed the use of its technology for war-fighting efforts, and ultimately led Google to decide not to renew the contract.11 This episode illustrates the uphill battle the DoD faces in leveraging the commercial sector—especially Silicon Valley technology firms that do not typically compete for government contracts—in its pursuit of keeping up with Russia and China. Compounding the problem is the relative lack of a coherent AI acquisition and adoption strategy when juxtaposed to its competitors.
Our “Sputnik Moment”
China and Russia have both articulated their plans for developing AI. Vladimir Putin stated that AI leadership was a means to become the leader of the world.12 China has estimated that they can boost economic growth with AI by twenty-six percent by 2030.13 Both countries are striving to be the dominant power in AI and are utilizing the blurred line between private and public industry in their countries, which is very different from the commercial sector and government procurement system in the U.S. In both China and Russia, there is little, if any, distinction between defense and commercial sectors.14 Additionally, China is acquiring AI expertise from the U.S., funding over $1 billion in venture capital in U.S.-based tech firms since 2010.15
Because of the capabilities of AI, and the anticipated uses of such capabilities by China and Russia, experts have sounded the alarm, including former Deputy Secretary of Defense Bob Work, who claimed AI is in a “Sputnik Moment” akin to the Cold War’s space race.16 Like the space race, supremacy in the AI domain cannot be won by the DoD alone—the defense industrial base, augmented by non-traditional contractors such as technology firms, must be leveraged by the government.
Obstacles to AI Acquisition
There are several challenges the DoD must overcome to remain competitive in the AI domain. The first is understanding the potential of AI and determining the legal and ethical restrictions on the use of such technology. In order to do so, requiring activities must agree on concrete definitions of terms such as AI and autonomy prior to drafting requirements.17 Beyond communicating with potential offerors what the requirements are, definitions are necessary to conduct a legal review of any new weapon system. A legal review of the intended acquisition or procurement of weapons or weapon systems is required by the DoD Directive 5000.01, The Defense Acquisition System.18 Such review must ensure compliance with the laws of war.19 Understanding the capabilities of AI, particularly autonomous systems, is challenging to experts in the field; lawyers may require training or consultation with such experts to ensure that the review is sufficient.
Perhaps the most challenging sector of AI in terms of ensuring legal compliance with the laws of war is that which includes lethal autonomous weapons systems (LAWS). Defined as “AI systems capable of independently identifying a target and employing an onboard weapon system to engage and destroy it with no human interaction,” LAWS are one special class of AI that poses a host of legal questions.20 Elon Musk has warned that LAWS will “permit armed conflict to be fought at a scale greater than ever, and at timescales faster than humans comprehend.”21 While the United Nations struggles to define, let alone establish restrictions on the development of LAWS,22 the U.S. must consider its position on how much control it is willing to cede to AI.23 While some may argue that the U.S. will be left behind in the AI arms race if it limits its use of AI by keeping humans “in the loop” (requiring human approval prior to the system carrying out an action),24 such decisions have yet to be made by the international community.
While the legal and ethical dilemmas posed by LAWS is worth considering, many AI applications short of LAWS are being procured by the federal government already. The fundamental question facing current AI acquisition is whether the acquisition system can keep up. Due to the pace of innovation in AI and the rate technology diffuses across international boundaries, any advance in technology resulting in fielding new military capabilities is likely to be short-lived; it is unlikely that any technology advantage in AI will last more than two to five years given the level of competition.25 Not coincidentally, the topic that has taken up more time and energy within Congress and the Pentagon than AI in recent years is acquisition reform.26 Much of the reform that has taken place has been for speeding up the lethargic pace of defense acquisitions and removing regulatory and bureaucratic roadblocks that hinder the DoD’s access to the commercial sector, particularly in Third Offset technologies like AI. Unlike most major defense-related technologies in the past, AI development is led by civilian companies; thus, to achieve success in AI acquisition, these reforms are necessary. Because private sector funding dwarfs current government R&D,27 the DoD must leverage the commercial sector to achieve its AI and wider strategic goals. Attorneys should understand the potential obstacles to engaging with the commercial sector. The highest hurdles are attracting industry to participate with the DoD and then overcoming the traditional acquisition system’s red tape. These are not mutually exclusive; much of the recent acquisition reform is focused on providing the acquisition corps tools to bypass lengthy, costly, and burdensome procurement laws and regulations in order to become a more attractive customer.
By knowing the tools and innovative business practices available to the DoD to attract the commercial sector and develop, acquire, and field new AI applications at high speed, attorneys can help shape the acquisition strategy to align with the National Defense Strategy. Under the traditional procurement system, governed by a system of statutes and regulations, such as the Federal Acquisition Regulation (FAR) and its supplements, common complaints from industry are that government acquisition is too slow, rigid, overbearing, and expensive.28 One potential tool to address these complaints is the DoD’s other transaction authority (OTA) under 10 U.S.C. § 2371b. While OTAs are not new,29 the authority has been expanded several times since the FY2016 NDAA,30 and OTAs are now experiencing a renaissance within the DoD. Because OTAs are not encumbered by many procurement statutes or the FAR, the DoD can enter into agreements with the commercial sector in much the same way as a commercial buyer. The agreements can be negotiated and tailored to the requirements, while bypassing restrictive compliance regulations and expensive accounting systems. Importantly, the intellectual property (IP) requirements under the Bayh-Dole Act and Defense Federal Acquisition Regulation Supplement (DFARS) Part 227 that force technology firms to hand over IP to the government do not apply under section 2371b. For technology firms developing certain AI applications, IP could be that firm’s only asset—agreeing to provide unlimited rights to the government could result in corporate suicide. The freedom to deviate from the rigid IP rules under traditional contracting procedures could be the difference between a firm choosing to enter into an agreement with the DoD or turning to purely commercial pursuits.
Additionally, because the Competition in Contracting Act (CICA) does not apply to OTAs, the timeline between the request for proposal (or other solicitation method) and award can be significantly reduced.31 While section 2371b authorizes the DoD to enter into an agreement for a prototype that enhances mission effectiveness, the term prototype is not defined.32 The DoD has interpreted prototype broadly to include hardware, software, and even business practices adapted to military use. Moreover, section 2371b(f) provides for the option to award a follow-on OTA or contract for production of the prototype without competition, provided the original OTA was competed and the agreement provided for a follow-on option, and the prototype project was successfully completed.33 Given the possibility to bypass many barriers to commercial participation and a clear path to developing and fielding emerging technology through use of section 2371b, the DoD should continue to embrace and expand its use of OTAs for AI acquisition. An attorney well-versed in this alternative acquisition method, in both its possibilities and its pitfalls, will be able to effectively advise requiring activities, program managers, and contract officers in AI procurements.
Other potential acquisition tools to consider for AI acquisition are section 804 authority—which permits rapid acquisition and rapid fielding for middle tier programs intended to be completed in two to five years34—and section 806 authority—which allows the Secretary of Defense to, under certain circumstances, waive any provision of acquisition law or regulation if the acquisition of the capability is in the vital national security interest of the U.S.35 These authorities help bypass parts or all of the programmatic requirements under DoD Directive 5000.02, leading to faster acquisition timelines. When contracting with nontraditional technology firms that are performing AI R&D, rapid acquisition is critical. Obtaining funding is necessary for start-ups to survive. Long acquisition lead times in the range of years can limit competition as cash-starved start-ups lack the capital to stay in business throughout the source selection process of many traditional procurements. For more established tech firms with no funding concerns, the fear becomes that they will look outside the federal government for business, potentially to near-peer competitors. Limiting programmatic requirements can help ensure funding goes to the most innovative solutions, rather than simply the biggest contractor.
However, even if the DoD fully embraces the tools made available to it through recent acquisition reform, it still has to attract businesses like Google to develop technology such as Project Maven. Overcoming public perception and employee protests will have to be a part of the DoD’s overall AI acquisition strategy. To help address that issue, the DoD has stood up several organizations that focus on building relationships with non-traditional defense contractors from Silicon Valley, Austin, Boston, and other tech hotbeds. The Defense Innovation Unit (DIU) was a pioneer in this field, and the Army is embracing the concept with its newly stood-up Army Futures Command in a skyscraper in downtown Austin, Texas, rather than inside the wire of an Army post. Another organization within the Pentagon that utilizes these new acquisition tools to fast-track the development of military applications of commercially available technologies is the Strategic Capabilities Office (SCO).36 While much of the SCO’s portfolio is classified, it is known as the initial phase of the Third Offset Strategy.37
Recent successes in attracting commercial start-ups by organizations such as the DIU has come from competitions where the DoD can evaluate multiple prototypes from industry, and the firm can claim primacy in that particular market area.38 For start-ups, recognition as the standard-bearer is “a more valuable incentive than return on investment, which is why competitors in the DARPA robotics challenge were willing to spend a collective $85 million to a win $1 million prize.”39 These contests can be carried out under a simple OTA to increase interest in the commercial sector in working with the DoD, and they can provide the DoD with an opportunity to see what advancements the commercial sector has made in AI that would be worthwhile to pursue and adapt to military purposes.
Forming an AI Acquisition Strategy
While the previously discussed acquisition tools will assist the DoD in procuring discrete AI applications, it is important for the DoD to develop an overarching AI acquisition strategy. To start, requiring activities should procure AI applications with a purpose. Before acquisition of new AI capabilities, the requiring activities should understand what the AI will do and how those it will be incorporated into doctrine, as well as ensure interoperability with existing systems. The key to leveraging AI to meet the National Defense Strategy is not to simply acquire AI and then learn what it does and field it in the future. Adopting and fielding AI faster than competitors is essential to maintaining a technological advantage, however incremental and temporary that advantage may be.
To provide sound counsel, acquisition attorneys should become conversant with the legal and ethical issues posed by the advancement of AI technology. From the initial drafting of requirements through award, attorneys can help navigate the various issues that face the program manager and contracting officer in procuring AI. Mastering the acquisition tools available to meet this national security priority is critical to maintaining a technological advantage in this new arms race. TAL
1. Memorandum from William B. Roper, Jr., subject: Greetings to the Acquisition Workforce (Mar. 13, 2018) (on file with author).
2. See id.
4. Paul Scharre, Army of None 5 (2018).
5. See Daniel S. Hoadley & Nathan J. Lucas, Cong. Research Serv., R45178, Artificial Intelligence and
National Security 17-21 (2018).
6. Jesse Ellman, Lisa Samp & Gabriel Coll, Ctr. for Strategic & Int’l Studies, Assessing the Third Offset Strategy 3 (2017). The “Third Offset Strategy” is the DoD’s focused efforts to ensure “the ability of the United States to project military power in the face of an emergent suite of advanced military capabilities” such as AI. Id. at 1.
7. See Hoadley & Lucas, supra note 5, at 1, 23.
8. Nat’l Def. Authorization Act for Fiscal Year 2019, Pub. L. No. 115-232, § 238, 132 Stat. 1636 (2018) [hereinafter FY2019 NDAA].
10. Hoadley & Lucas, supra note 5, at 9.
11. Palmer Lucky & Trae Stevens, Opinion, Silicon Valley should Stop Ostracizing the Military, Wash. Post (Aug. 8, 2018), https://www.washingtonpost.com/opinions/silicon-valley-should-stop-ostracizing-the-military/2018/08/08/7a7e0658-974f-11e8-80e1-00e80e1fdf43_story.html?noredirect=on&utm_term=.eb54ad0fbb2a.
13. Colin Clark, Our Artificial Intelligence ‘Sputnik Moment’ is Now: Eric Schmidt & Bob Work, in Falling Behind, DoD Scrambles to Buy Tech Faster 14 (2018).
14. Andrew Hunter, A Strategic Approach to Defense Investment, Ctr. for Strategic & Int’l Studies (Mar. 26, 2018).
15. Hoadley & Lucas, supra note 5, at 19.
16. Clark, supra note 13, at 14.
17. Connie Lee, Army to Pursue ‘With Urgency’ Autonomous Systems Strategy, Nat’l Def. (June 1, 2018), https://www.roboticresearch.com/news/army-to-pursue-with-urgency-autonomous-systems-strategy/.
18. DoD Law of War Manual 337 (June 2015, updated Dec. 2016); U.S. Dep’t of Def., Dir. 5000.01, The Defense Acquisition System, para. E1.1.15 (12 May 2003, certified as current as of Nov. 20, 2007) [hereinafter DoDD 5000.01].
19. DoDD 5000.01.
20. Hoadley & Lucas, supra note 5, at 12.
21. Id. at 1.
22. See id. at 23.
23. The current U.S. policy is found in DoD Directive 3000.09, Autonomy in Weapon Systems (Nov. 21, 2012). The Directive states that it is DoD policy that “[a]utonomous and semi-autonomous weapon systems shall be designed to allow commanders and operators to exercise appropriate levels of human judgment over the use of force.” Id. at para. 4a.
24. See Scharre, supra note 4, at 29.
25. See Ellman et al., supra note 6, at 1.
26. From Fiscal Years 2016–2018, “National Defense Authorization Act (NDAA) titles specifically related to acquisition reform contained an average of 82 provisions (247 in total), compared to an average of 47 such provisions (466 in total) in the NDAAs for the preceding 10 years.” Moshe Schwartz & Heidi Peters, Cong. Research Serv., R45068, Acquisition Reform in the FY2016-2018 National Defense Authorization Acts (NDAAs) 1-2 (2018).
27. “There are multiple Silicon Valley and Chinese companies who each spend more annually on AI R&D than the entire United States government does on R&D for all of mathematics and computer science combined.” Gregory C. Allen & Taniel Chan, Artificial Intelligence and National Security, Bulletin of the Atomic Scientists (Feb. 21, 2018).
28. See U.S. Gov’t Accountability Office, GAO-17-644, Military Acquisitions, DOD is Taking Step to Address Challenges Faced by Certain Companies, (2017), at 9.
29. Other Transaction Authority was first granted to the National Aeronautics and Space Administration (NASA) in 1958 to permit the new agency to enter into agreements with commercial entities outside of the procurement system in order to respond to the Soviet launch of Sputnik. Space Act, Pub. L. No. 85-568, 72 Stat. 426 (1958).
30. See Nat’l Def. Authorization Act for Fiscal Year 2016, Pub. L. No. 114–92, § 815, 129 Stat. 726, 893 (2015) [hereinafter FY2016 NDAA]; Nat’l Def. Authorization Act for Fiscal Year 2018, Pub. L. No. 115-91, 131 Stat. 1283 (2017) [hereinafter FY2018 NDAA].
31. The Competition in Contract Act (CICA), 41 U.S.C. § 253, sets out timelines for publication, solicitation, and award; requires full and open competition unless an exception applies and written justification is obtained; and provides options to protest the specifications and source selection.
32. 10 U.S.C. § 2371b(a).
33. 10 U.S.C. § 2371b(f); see also Oracle America, Inc., Comp. Gen. B-416061, 2018 CPD ¶ _ (Comp. Gen. May 31, 2018) at 18–19. Oracle was a protest testing the limits of Section 2371b and its follow-on authority. While GAO ultimately sustained the protest, it affirmed DoD’s interpretation of the term prototype and made clear the path to successful follow-on activities.
34. FY2016 NDAA, Pub. L. No. 114–92, § 804, 129 Stat. 726, 882 (2015).
35. FY2016 NDAA, § 806 at 885.
36. Cheryl Pellerin, DOD Strategic Capabilities Office is Near-Term Part of Third Offset, DOD News (Nov. 3, 2016), https://www.defense.gov/News/Article/Article/995438/dodstrategic-capabilities-office-is-near-term-part-of-third-offset/.
37. Larry Lewis, Insights for the Third Offset: Addressing Challenges of Autonomy and Artificial Intelligence in Military Operations, CNA (Sept. 2017), https://www.cna.org/CNA_files/PDF/DRM-2017-U-016281-Final.pdf.
38. Ellman et al., supra note 6, at 8.