Our aim in this essay is to consider how policymakers make decisions about government surveillance in what we might call the machine learning state –– a nation-state equipped with sufficient bureaucratic and technological capacity to rely extensively on machine learning techniques for surveillance, law enforcement, and national security. We focus particularly on the question of how the state’s political economy influences its decision to adopt privacy-relevant machine learning technologies. Since machine learning tools can also be deployed in many ways that are not pertinent to privacy, our focus therefore is on a specific subset of state uses of such technology –– to engage in surveillance of the public and its activities. In order to lay the groundwork for nuanced engagement with the legal and policy trade-offs in this domain, we aim here map the main technological and institutional forces shaping a state’s deployment of new machine learning capabilities that can affect privacy, and then to explore, more tentatively, their likely effects on technological uptake.
We propose that state adoption of machine learning instruments for surveillance occurs in a variation on what Robert Putnam famously characterized as a “two-level game.” The state is operating simultaneously in a domestic political environment populated by institutions mediating conflicts involving civil society and firms competing to expand and monetize machine learning capacities, and also in an international environment in which it is competing with other sovereign nations that are cultivating and deploying similar capacities for geostrategic ends. How and to what end machine learning instruments are deployed depends on the strategic choices that the national government makes in these two overlapping yet distinct contexts. We emphasize that it would be a mistake to analyze these choices purely in terms of responses to domestic considerations, because states may be willing to shoulder the costs of domestic backlash so they can further geostrategic goals. We also elucidate the way the net vector of domestic and international pressures is likely to affect legal and policy interventions in this space. Based on this exercise, we identify concerns and trade-offs relevant to the possible reforms; and explore both the limitations of existing Fourth Amendment doctrine, and the potential of state and federal legislative or regulatory alternatives as reform instruments.
Cuéllar, Mariano-Florentino and Huq, Aziz Z., "Privacy’s Political Economy and the State of Machine Learning" (2019). Public Law and Legal Theory Working Papers. 787.