Panel Discusses Professor Deeks’s New Book, The Double Black Box
The Law School hosted a panel on Thursday, September 4 to celebrate and discuss Professor Ashley Deeks’s new book, The Double Black Box: National Security, Artificial Intelligence, and the Struggle for Democratic Accountability. The panel was introduced by Dean Leslie Kendrick ’06 and moderated by Professor Danielle Citron. The other panelists were Georgetown Professor Martin S. Lederman, Indiana University Professor Asaf Lubin, and University of Minnesota Professor Alan Rozenshtein. In addition to their scholarship, the panelists drew on extensive firsthand experience in government service.
Professor Deeks’s book examines how to ensure that the executive branch’s use of artificial intelligence (AI) is consistent with the public law values embraced by the United States and other democracies, such as legality, efficiency, and accountability. As she explained, the issue is challenging because of the intersection of the two “black boxes” alluded to in the book’s title. The first is the black box of the national security state, which is considerably insulated from scrutiny to prevent sensitive information from reaching the nation’s adversaries. The courts and legislature have difficulty checking the executive branch—which houses the intelligence and national security buraeucracy—because they lack comparable expertise and are often dependent on the executive branch for information, and therefore may afford considerable deference in this area. The second black box is AI, whose functioning and accuracy is difficult for humans to interpret. The issue will only grow as intelligence agencies, attracted by the potential of AI and concerned about lagging behind nations such as Russia and China, embrace the technology to make sensitive decisions.
Professor Deeks suggested some sources of guidance. First, several nontraditional actors can provide a degree of oversight over the executive. These include foreign allies, legal counsel within the executive branch, technology companies, journalists and whistleblowers, and the public at large. She also suggested looking to past instances where the nation confronted separation of powers concerns—such as after Watergate—and to existing statutes and oversight boards for inspiration. Nevertheless, it is clear that these solutions are partial and tentative. As Professor Deeks stated: “We have a big job ahead of us, both as lawyers and citizens, but hopefully this book at least identifies the challenge.”
The panel was full of praise for Professor Deeks’s sober analysis, clear writing, public service, and humility. A few themes emerged in the discussion that followed. One is that legal concerns depend significantly on predictions about the efficacy of future technology. At one extreme, Professor Citron seemed to feel that AI may not be more accurate than human judgement because it is trained on its own “slop.” At the other extreme, Professor Lubin, and to a lesser extent Professor Rozenshtein, argued that we will soon see artificial general intelligence, a term for a system which outperforms humans across a general range of tasks. If this is true, today’s assumptions about AI governance will cease to hold. A slightly different perspective came from Professor Lederman, who questioned whether, assuming that it is generally accurate, AI is different in kind from the techniques already used by intelligence agencies, which are highly sensitive and confidential. One serious question, then, is how nonspecialist lawyers can evaluate this technical issue in the face of so much uncertainty.
A second clear theme was the importance of developments in the human world. As Professor Lederman observed, the Trump Administration has made it the case that, “your job in the intelligence community is to give the president the intel he wants or be fired.” This prevents executive branch lawyers, traditionally a key internal check on the president, from serving this role. Professor Deeks also noted that technology companies have seemingly become less interested in the ethics of technology since she started writing the book. She therefore became more pessimistic about their potential as an external check.
Somewhat in this vein, Professor Rozenshtein gave an especially thorough set of comments about the ways that AI might concentrate executive power. For instance, AI could massively expand enforcement capacity and allow the executive to act much more quickly than the other branches can react. The president could also use AI to tailor government speech, much of which is immune from rigorous First Amendment scrutiny, to serve as highly effective propaganda. And the president could use AI to monitor workers throughout the executive branch for compliance with his or her priorities. Professor Rozenshtein noted that AI could be a “lawyer’s best friend” because of its efficiency and rationality and could create “lawful good governance of the sort that reformers have long dreamed about.” But “if we do nothing” it will “make the unitary executive true not just in theory, but in practice.”
Overall, the gap between the possible power of AI and the volatility and confusion of human actors today left many in the room pessimistic about our ability to face the challenge Professor Deeks identified. Nevertheless, it is clear that, as she observed, lawyers will be key actors in ensuring national security AI complies with our values.
