Tom Stefanick
Brookings InstitutionZoom link: https://mit.zoom.us/j/96339087990
Abstract: The surprising capabilities of deep neural network-based algorithms operating on shared internet data, scientific data, and other collaborative endeavors have been revealed in milestones since around 2012, and are now merging with internet business models. There are also expectations that these computing capabilities will transform warfighting, which is a large part of the motivation for US efforts to choke off China’s high-end semiconductor development. This talk will address some of the reasons why a broad-based transformation of military targeting and weaponry based on these deep learning algorithms is unlikely. These reasons fall into two categories: information limits on detection and estimation in adversarial data environments, and the availability of existing approaches that have been in use and are well-suited to the battlespace environment. I will identify some applications for deep learning that are likely to be useful for US military and intelligence programs.
BIO: Tom Stefanick is a Visiting Fellow in the Strobe Talbott Center on Security, Strategy, and Technology at the Brookings Institution, working on military targeting and maritime security. Prior to this position, he was a senior vice president at Metron, Inc. and oversaw business units producing algorithms, autonomous systems, and simulation models for the Department of Defense. He worked in the House Armed Services Committee in 1987 and 1988 on Soviet submarines and strategic antisubmarine warfare after completing a book on that topic.