Today’s post was written by Public Interest Declassification Board (PIDB) members Carter Burwell and Carmen Medina.
On Friday, April 19, 2024, two members of the PIDB spoke at Columbia University at the 2nd Annual Conference in Honor of Robert Jervis. The event was titled “The Dark State? Government Secrecy and American Democracy” and was hosted at Columbia’s School of International and Public Affairs (SIPA) and presented by the Arnold A. Saltzman Institute of War and Peace Studies and the Institute of Global Politics. The event grew out of the PIDB’s 2023 conference at the L.B.J. Presidential Library and the University of Texas at Austin and was spearheaded by Columbia Professor Matthew Connelly, author of The Declassification Engine: What History Reveals about America’s Top Secrets. A recording of the event is available here.
After a gracious introduction to the conference by SIPA Dean Keren Yarhi-Milo and Saltzman Institute Director Page Fortna, PIDB member Carter Burwell moderated the first panel on the “Law of Secrecy.” The panelists included Northwestern Law School Professor Heidi Kitrosser, author of the book Reclaiming Accountability: Transparency, Executive Power, and the U.S. Constitution, Columbia Law Professor David Pozen, author of the Harvard Law Review article “The Leaky Leviathan: Why the Government Condemns and Condones Unlawful Disclosures of Information,” and American Civil Liberties Union (ACLU) lawyer Ben Wizner. Among other things, the panelists discussed the state of secrecy in American government, including issues associated with executive branch primacy, legislative deference and dysfunction, and judicial acquiescence and abdication in areas such as the Freedom of Information Act and the state secrets privilege.
PIDB member Carmen Medina joined the second panel on “The Use and Misuse of Artificial Intelligence to Manage Dangerous Information.” The panel was moderated by author and journalist Sue Halpern, and fellow panelists included Camille Francois, a lecturer at SIPA, and Hanna Wallach, an AI executive at Microsoft. The panel agreed that AI had a role in a future declassification regime, but that training an AI model on the subtleties of declassification decisions would not be straightforward. Panelists agreed that a private-public partnership was essential to progress. Both the government and the American public would have to accept a certain error rate given the volume of documents and the vagaries of classification decisions (also known as “good enough for government work”). Even if a system achieved a “six sigma” level of accuracy, potentially thousands of declassification errors could occur each year. A novel application of AI might be to use an AI engine to rationalize classification and declassification guidelines. One panelist noted that she had heard anecdotally that individuals were using AI to fill in the gaps in redacted documents. Finally, panelists expressed concern that an AI declassification engine could be hacked by adversaries but that this threat was unavoidable in the use of digital technology.
Professor Elizabeth Saunders moderated a third panel on “The Crisis of Overclassification: Can Proposed Reforms Help?” Participants included Michael Collins from the National Intelligence Council, Joe Lambert, former Director of Information Management Services, and Tim Naftali, a senior research scholar and adjunct professor at SIPA. Panelists discussed the DNI’s commitment to greater transparency and ongoing efforts to declassify materials, including the priorities and efforts of Professor Jervis at the CIA. The conference then closed with Professor Connelly hosting a panel of the moderators with discussion about various paths forward on classification reform.