Vol. 31 No. 2 (February 2021) pp. 48-52

THE CENTAUR’S DILEMMA: NATIONAL SECURITY LAW FOR THE COMING AI REVOLUTION, by James E. Baker. Washington, Brookings Institution Press, 2021. 335 pp. Paperback $34.99. ISBN: 9780815737995.

Reviewed by Tobias T. Gibson, Department of Political Science, Westminster College (MO). Email:

Judge James E. Baker, Director of the Institute for Security Policy and Law and Professor of Law at Syracuse Law School, begins THE CENTAUR’S DILEMMA with two truisms. The first is that artificial intelligence (AI) is widespread and will only become more ubiquitous. The second is that law is rarely, and perhaps never, in line with the technology that it is supposed to govern.

As described by Baker, the approaching dilemma is that of decision making, especially in the realm of national security, that will combine human and machine—like the half human, half horse centaur of mythology. Yet, while this is imminent, and in some ways a current issue—think of your reliance on the Waze app—Baker’s stated goal of this book is to allow generalists, including policymakers, to debate and design a legal framework. The time is now because allowing time for debate, including a wide variety of stakeholders, will allow lawmakers “to make informed, purposeful, and accountable decisions about the security use of governance of AI” (pp. 5-6).

The first four chapters of the book establish the problems and provocations of artificial intelligence in a national security setting. To be sure, uses and issues related to AI will certainly arise in areas of policy spaces related to traditional national security spheres, such as military and intelligence. Fighter pilots, autonomous vehicles—whether air based “drones,” or increasingly land and water-based ones—and international surveillance tools and data collection all will be enhanced by cooperative work between human and AI. Baker’s work is focused on establishing a protocol of law and policy that will not allow the cooption of decision making by the coming, and many would argue present, AI revolution.

However, he also adds that the use of AI in everyday items, the so-called Internet of things, also needs to be governed to prevent excessive action on the part of the government and corporations that build smart cars, smart toasters, smart coffee makers, smart refrigerators, and smart phones. As Baker notes, a series of recent Supreme Court cases, discussed below, have led to bright lines in the ways that collected data is used by law enforcement. However, there is much to discern and develop, as AI technology advances beyond Facebook and Amazon algorithms and becomes far more ubiquitous.

That said, however, some of the most directly applicable and, quite frankly, most developed portions of the proposed framework are found in case law, much of which is seminal. Chapter 6, “Constitutional Law,” provides the most directly applicable chapter for scholars of law and courts. [*49]

The chapter begins by noting that domestic law is rooted in the Constitution. While self- evident, this matters when it comes to laws related to commerce, for example. Baker uses the Defense Production Act (DPA), which “is predicated on Congress’s authority ‘to regulate Commerce with foreign Nations, and among the several States’ and to a certain extent ‘to make Rules for the Government and Regulation of the land and naval forces’” (p. 95). To be sure, the DPA is a powerful tool. In the (relatively) early days of the U.S. struggles against the COVID-19 pandemic, Baker wrote in the NEW YORK TIMES that

The law is so broad in places that it is sometimes referred to as a “commandeering” authority. Lawyers prefer to say the president would act at the zenith of his authority under the paradigm presented in Justice Jackson’s concurrence in the Supreme Court’s landmark Youngstown case. But its use is not as extraordinary as some suggest, and it is not commandeering. The Defense Department alone uses the prioritization authority some 300,000 times a year, while the government uses Title III incentives 20 to 30 times per year. Although the allocation authority has not been used since the Cold War, some civilian airliners and freighters remain allocated for the Civil Reserve Air Fleet. Nor does exercise of the prioritization and allocation authority equate to state ownership. Under the act, corporations are paid fair market value for their products. Any actual “commandeering” of production would require just compensation under the Fifth Amendment’s Takings Clause. Moreover, price controls under the D.P.A. require a joint resolution of Congress signed into law by the president.

An example of a legal unknown, however, is the breadth given to Congress under the DPA that allows them to require a company or university to provide AI for the purposes of national security. Moreover, what role does the judiciary, which generally defers to the political branches, have over national security questions (Rudenstine 2016)? Baker closes the DPA and separation of powers discussion with lessons from YOUNGSTOWN, especially the key point that it is not a DPA case. The Truman administration relied on its perceived constitutional powers and did not rely on statute. And, Baker argues, one lesson is that DPA, and the Selective Service Act, are evidence that “Congress did anticipate the prospect of a labor disputes in the defense industry and gave the president statutory authority to address it” (p. 139). In short, there is an argument that these statutes, coupled with Justice Jackson’s three zones, would allow the president to act with the full power of Congress in such a contested space as AI.

Beyond separation of powers, Baker utilizes the recent FBI versus Apple row about allowing access to an iPhone used by Syed Rizwan Farook, who, with his wife, killed fourteen people and wounded another seventeen in a shooting in California. While the media rightly focused on the battle between government and corporation as a question of privacy, especially in a post-Snowden revelations political environment (Kharpal 2016), Baker spends a time discussing the legal question. [*50] Importantly, because the case involved a smart phone and data backed up on the Cloud, the issue centered around the All Writs Act, part of the Judiciary Act—of 1789. Suffice it to say that, quite reasonably, Baker has concerns about this. But, he offers a series takeaways, including (1), “where national security is concerned, the government will seek to use the law it has,” (2), “where the law is not clear, or reasonable arguments can be made on either side, courts will vary in decisional outcome,” and (3), and perhaps importantly given the overall argument of the book, “litigation is a poor substitute for legislative and executive branch policymaking” (p. 98).

While there is also much discussion on the First Amendment (who has First Amendment rights when AI base code is written by a human, but then the AI writes original work (p. 102)), Fifth Amendment procedure due process (pp. 124-126), substantive due process (pp. 126-133), and the Takings Clause, as noted above, perhaps the key discussion is that of the Fourth Amendment.

Baker offers a thorough discussion of the development of Third Party Doctrine through MILLER and SMITH, but notes how dated the doctrine is. Beyond being developed in the 1970’s, SMITH uses a metaphor regarding a telephone company knowing the number dialed versus what was said on a particular call as the difference between the address on an envelope versus the content of the letter (p. 111). He then offers thoughts on updated cases, such as JONES and CARPENTER. JONES offered an update to what a lawful search incident to arrest includes, because the amount of data available on a phone, which could be in plain sight or “pocket litter” offers police far too much information about the lives of an individual, and should be seen more as the information within one’s home. CARPENTER, though, is worth some merit, as the Court held that access to GPS data means that a suspect “… has effectively been tailed every minute of every day…” (p. 116). Moreover, CARPENTER, Baker argues “is new law” because the decision “signaled a desire to look at the privacy implications of big data and the aggregation of data on a case-by-case basis and, perhaps, on a technology-by technology basis” (p. 117). This is exceptionally important, given the impact AI and related technologies have on what constitutes a “reasonable” search. Moreover, there may be loopholes, such as the government’s willingness to purchase data rather than “search” or “seize” it (pp. 117-118).

Baker includes much on treaty law and established practices, for example, that are not part of a typical Judicial Process, Supreme Court, or Constitutional Law class. For example, the focus of Chapter 10, “The Means and Methods of War,” offers the “Ottawa Treaty,” which sets limits on landmines use, development, and production as an example of how the United States, one of thirty-six nations that is not a party to the treaty, is nonetheless largely bound to it. The U.S. adopted a policy to only use landmines on the Korean peninsula, and then only to utilize those that “disarm or self-destruct within four hours, and thus do not remain a persistent threat to either civilians or combatants” (p. 228). Baker suggests that a public relations campaign, based on international agreement(s), can be used to widely limit the use of AI-based weapons or weapons systems that may, even indirectly, target civilians and non-combatants. Indeed, as the U.S. example [*51] illustrates, while a state may not become a party to such an agreement, a country’s use of a weapon can be limited due to a widespread understanding of its deleterious and unintended consequences.

International law prevents perfidy, the “killing or wounding treacherously of individuals belonging to the hostile nation or army” (p. 230). Baker suggests that much like donning a Red Cross symbol is banned, use of AI to simulate an opponent’s weapons system, or capturing an adversary’s weapon or weapons system, and then using AI to turn that weapon on that country is the type of international law question that needs to be debated. Meaningful, widespread limits and rules must be placed on before it becomes a common international issue.

Some potential caveats for LPBR readers: the book extends outside the common teaching and research of much of the Law and Politics community; and while there is substantial discussion of constitutional law, there is at least one important and relevant case that I think Baker overlooks. As noted above, Judge Baker uses recent Supreme Court cases to help inform what could be a foundation for establishing, or at least informing, a constitutional consensus about the privacy required for consumers in an increasingly AI-based society. While Baker likely knows the law better than do I, I was surprised that he did not discuss KYLLO—especially since this case likely underscores his argument that the time is now to establish the legal and corporate limits of data collection and use. As a reminder for the reader, this case came to the Supreme Court after federal agents used warrantless thermal imaging to detect hot spots, which showed a heat pattern consistent with illegal marijuana cultivation and growing. In a 5-4 decision, the Supreme Court suggested that "[w]here, as here, the Government uses a device that is not in general public use, to explore details of the home that would previously have been unknowable without physical intrusion, the surveillance is a 'search' and is presumptively unreasonable without a warrant” (KYLLO V. UNITED STATES (2001)). Because this decision is based on a technology “not in general public use,” there was thus an element of a search. However, as Baker firmly establishes, AI and related technology is a part of much of our everyday lives. AI impacts the way we shop, listen to music, navigate our road trips, and much, much more. In other words, it is pervasive, not merely common. As such, the need to demarcate how the use of AI, and the data collected, and how it impacts privacy, both from government and corporate eyes, matters now.

At the end of the day, despite its discussions that may swing beyond much of the content of a law and courts research or teaching agenda, I highly recommend this book for four reasons. First, Baker joins a growing literature in the field of national security law. In this case, Baker, who previously served as Chair of the American Bar Association’s Standing Committee on Law and National Security, is in a unique position to advocate for a legal framework governing the use of AI. Secondly, as society becomes more reliant on AI, including autonomous cars as well as weapons, it is good to ask our current students—themselves future lawmakers, judges, CEOs, and voters—to think about the implications of an increasingly AI dependent society. Third, Baker asks his reader to apply current case law to future legal questions. This is an important exercise for any current and [*52] future law student. Finally, this is a provocative book, and one that has importance well beyond national security.


CARPENTER V. UNITED STATES, 585 U.S. ____ (2018).

KYLLO V. UNITED STATES, 533 US 27 (2001).

SMITH V. MARYLAND, 442 U.S. 735 (1979).

UNITED STATES V. JONES, 565 U.S. 373 (2014).

UNITED STATES V. MILLER, 425 U.S. 435 (1976).



Baker, James E. April 3, 2020. “It’s High Time We Fought This Virus the American Way,” New York Times,

Kharpal, Arjun. March 29, 2016. “Apple vs FBI: All you need to know.”


© Copyright 2021 by author, Tobias T. Gibson