May 1, 2010  

Unblinking eyes

Managing the strategic consequences of persistent surveillance

The tactical use of a new set of intelligence collection systems known as persistent surveillance (PS) is well-known. Sometimes colloquially referred to an “unblinking eye,” a PS system is designed to maintain a constant watch over a target of interest. The primary beneficiaries of PS systems have been U.S. armed forces waging counterinsurgency warfare in Iraq and Afghanistan, where long-duration tracking of individuals has enabled many capture/kill operations.

Few, however, have sought to analyze the strategic implications of PS technologies. Yet new surveillance systems that combine PS with large-scale data retention and advanced algorithms have strong potential for social disruption. Recent technological trends indicate that, within the next five to 10 years, states will begin to field capabilities to monitor their citizens — and the citizens of other states — with sufficient detail to empower new forms of coercion. The potential created by these new technologies will lead states to compete over a redefined notion of influence in international politics. The U.S. military could gain significant first-mover advantages by hastening development of these technologies while simultaneously working to defend the populations of the U.S. and its allies from the information-gathering efforts of other, authoritarian, states.

PS systems are designed to maintain a constant watch over a target. Tactically, targets are usually individuals; at the strategic level, PS may be applied to a large physical space, such as an urban area. In the long run, as systems improve and come to include many sensors, one may consider strategic PS referring to an intelligence agency monitoring an entire state. High-resolution imagery is the technology most often associated with PS: It allows a surveillant to focus on objects of interest from within a very large viewing frame. For instance, with commercially available techniques such as gigapixel photography, a surveillant may take a digital photograph of an entire city skyline and then focus on a target as small as an apartment window.

Imagery is but one possible sensory input to a PS system; others include signals, such as radar or digital data from information networks. Examples of such networks are financial or banking systems. PS may combine several kinds of sensors for the best possible understanding of target behavior. In the case of an individual, this may constitute visual observation of the target from afar with high-resolution imagery, audio surveillance through distributed microphone arrays, transactional observation by tracking credit card charges, and analysis of the individual’s social network by recording time-stamp information on incoming and outgoing phone calls and e-mail.

Large-scale data retention is the ability to store extremely large quantities of digital information. The growth of this technology sector is readily apparent; as consumers, we notice increases in the size of hard drives over time as their cost declines. PS may be combined with large-scale data retention systems, thereby creating a TiVo-like capability to replay long periods of recorded social behavior. Such a system would allow surveillants to develop models of an individual’s behavior — How often does he attend religious services? Does he tend to speak with the same person before or after going? For how long? — and to validate new leads in a law enforcement or intelligence investigation.

For example, if new information suggests that the individual under surveillance conducts meetings about illicit activity at a particular place, the TiVo-like archive may be used to determine how far back in time those meetings had taken place. Such information could be used to develop new investigatory leads for prosecution or intelligence collection, particularly with sufficiently high resolution to identify individuals attending those meetings going back weeks or months. For a PS system in which the target is a public area, rather than an individual, large-scale data retention would allow for tracing the steps leading to an event in that area.

A sequence with clear national security implications would be the persistent recording of a high-threat public place. Large-scale digital retention would allow for very-high-quality video to be stored for long periods of time. Should an event such as a terrorist attack take place, the archival imagery of the public space could be reviewed to determine details of high consequence, such as the moment a bomb was placed or a terrorist casing the targeted area. With sufficiently high-resolution imagery, a law enforcement or intelligence user could then zoom in on an individual face in a heavily populated urban environment, thus identifying the attacker. Unfortunately, these capabilities apply equally to authoritarian regimes, which may use forensic capabilities to crack down on dissidents following public protests.

Algorithms are the programmed instructions that tell a computer what to do under specified conditions. For instance, the most basic algorithms use Boolean logic: “if condition x holds, then execute function y.” Computer systems throughout time have employed algorithms, but within the past couple of decades, algorithms have become far more sophisticated. More advanced algorithms offer the potential to automate tasks that normally would require some kind of human user judgment and input. An example might be algorithms capable of target discrimination; instead of having an operator tell a computer whether an object on a screen is a tank or rock, an advanced algorithm might be able to process sensory input to separate tanks on a battlefield from surrounding objects.

New algorithms have potential to revolutionize surveillance technologies in several key ways. The first is the aforementioned discrimination between objects of interest and the surrounding environment. A closely related capability, and one already demonstrated with reasonable levels of success, is the ability to identify unique targets within a large viewing frame. In particular, algorithms have been written that can draw on video footage to identify individuals by their face or gait. Second, algorithms may be used to develop models of social behavior at a level of complexity (referring to the number of variables and interactions) beyond the capability of the human brain. Third, advanced algorithms can make surveillance systems “smarter” by coordinating various sensor inputs.


This last function, sensor coordination, is an understated yet highly consequential application of advanced algorithms. A surveillance system must know how to maintain an “eye” on a target of interest as it moves between sensors, how to combine various sensory inputs into a meaningful aggregate target picture, and when to alert system users to noteworthy activity. So-called genetic algorithms, such as those that constantly seek optimization through random variation within user-defined parameters, may be one of the most consequential recent breakthroughs in software development. As computational power increases, so does the potential for software to manage systems of enormous complexity.

Each of the three aforementioned technologies may independently increase the power and scope of surveillance efforts. Combining the technologies, however, portends truly revolutionary advances. Such an integrated surveillance system would allow a surveillant to observe an extremely large number of individuals indefinitely (PS), store those data for study at a later time (large-scale data retention), and automate functions previously performed by surveillance personnel, such as using gait or facial recognition techniques, to identify and track individuals of interest (advanced algorithms). Current technology suggests the potential for surveillants to develop some of the following applications during the next five to 10 years:

• Governments will be able to police public spaces by developing models of “normal” behavior that the system may use to identify suspicious activity.

• Governments will have an unprecedented ability to locate and track specific persons of interest in urban environments.

• Corporations will be able to develop radically new models of consumers by monitoring many transactions at once.

• Providers of public infrastructure will gain historically unprecedented levels of detail on the use of their services, such as by monitoring traffic flows.

• Intelligence agencies may have the power to drastically improve their awareness of a wide range of behaviors of and in foreign countries.


For decades, academics and policymakers have debated the question to what extent may complex social systems be understood and modeled? To those skeptical of the ability of behavioral scientists to capture the many variables that make up a social system, the technologies described here may sound like fantastic notions from the realm of science fiction. Many of these technologies, however, are already being applied to national security problems. In a recent article in the journal Joint Force Quarterly, for example, three U.S. military officers described the use of PS technologies to establish “pattern of life analysis” against enemy targets; these officers wrote that 24/7 multisensor observation has led to “hundreds of examples of successful raids.” Similarly, in an interview with Air and Space Power Journal, a retired British Royal Air Force officer noted the ability to manipulate opponent behavior through constant surveillance by overhead assets.

Academic publications demonstrate the ability of advanced algorithms to discriminate between and track targets in video footage. In 2001, researchers at Carnegie Mellon University demonstrated algorithms that automatically detect people and vehicles, determine their 3-D locations, and control a multisensor surveillance system such that a target of interest is tracked as it moves between sensors (e.g. from one camera’s field of view to another). A point worth noting about these experiments is that the video used is of far inferior quality to already available high-resolution imagery such as that discussed in the context of PS.

A final thought for technology skeptics to consider is the potential for incremental improvements to surveillance systems over long periods of time. Those who point out the flaws of attempts to model complex social systems often fail to recognize the cumulative long-term gains in the fielding of new technologies. Incremental improvements that drastically increase the complexity of information systems are visible everywhere. Consider, for example, the development of UPS or FedEx package-tracking services. At the first move from paper to digital tracking, it might have seemed implausible that, in the foreseeable future, consumers would have the ability to automatically receive a text message on their cellular phones alerting them of a package delivery. Incremental improvements in package-tracking networks, such as the placement of sensors first at distribution centers and later in the hands of individual deliverymen, along with many software upgrades, have produced a massive system of data retention and management.

Regarding our argument, an integrated surveillance system such as that described above may be initially fielded with only one sensor (e.g., a camera), an ability to track only one kind of target (e.g., a person of interest), and function in only a small area (e.g., 100 yards). With an endless time horizon, however, one may envision a large number of small improvements that could eventually take the system to the level of fidelity described in our above example. Surveillants may begin by slowly increasing the range of the system (e.g., from 100 to 200 to 300 yards and so on). The surveillants may then upgrade the cameras to improve the quality of imagery, which in turn enables a greater range (fewer cameras are required to cover the same amount of physical space). Subsequent steps may include adding new sensors capable of measuring and recording other variables of social behavior, storing data for longer periods of time by adding more digital memory, and the always-present potential for new algorithms to redefine the use of existing data.


The U.S. is imperiled by other states’ use of such technologies against Americans and against their own populations. The former is characteristic of competitive strategies; the U.S. seeks to improve its informational advantage while defending itself from other states’ prying eyes. The core idea of interest is that, by dramatically increasing knowledge of an adversary’s behavior, one may reach a better approximation of that adversary’s incentive structure, in turn offering opportunities to effect policies that alter those incentives. Integrated surveillance systems, therefore, may be viewed as the technological precursor to a new form of influence in international politics. By recognizing this long-term threat — which will take decades to fully unfold — the U.S. stands to gain from first-mover advantages. By developing extensive PS-empowered intelligence collection systems on hostile foreign regimes, the U.S. would be better positioned to prevent those regimes’ technological advancements in turn.

The second threat category is the strong potential for nondemocratic states to use new surveillance technologies to perpetuate their regimes. As it is the stated long-term strategy of the U.S. to reduce the incidence of conflict by promoting stable democracies, the diffusion of surveillance IT necessitates a re-examination of how to promote democratization.


On the other hand, the U.S. seeks to gain leverage over adversarial regimes by gaining a superior understanding of their public social behavior. For instance, one may take an existing concept, such as “relative deprivation” (how a person’s material welfare compares with that of others in their society), and design surveillance systems to improve quantitative measures of that concept. In the most ambitious vision of such systems, looking ahead even further than 10 years, you begin to foresee the unraveling of traditional assumptions (e.g., economic utility functions) regarding social behavior in favor of increasingly thick description. This kind of information can help the U.S. to identify allies of the West within societies controlled by autocrats and better predict the onset of social revolution or other disturbances. This, in turn, would allow for targeted assistance and better-timed intervention in favor of democratization and spreading liberty abroad.

Social scientists and the federal government alike have yet to offer sufficient groundwork on integrated surveillance systems to inform a proper long-term competitive strategy. The Defense Department and military bureaucracies could (and should) begin to secure the U.S.’ information advantage by undertaking a series of targeted initiatives:

• Adopt the view that revolutionary surveillance capabilities are an emerging security topic of extraordinary strategic significance. Such a view is in marked contrast to the current DoD and military approach to using advanced surveillance techniques, which is to develop niche capabilities for specific applications at the tactical and operational levels of warfare. DoD and the services should be using the lessons learned from these more limited “unblinking eye” applications to choose new, large-scale investments in PS conceptualization, testing, and deployment.

• The social and legal implications of the diffusion of surveillance technologies is woefully understudied. While a small academic community on surveillance exists, these authors have focused almost exclusively on limited technologies such as closed-circuit television cameras in urban areas. Few have considered the potential for markedly more sophisticated technologies — although such technologies are on the market — and for their combination. DoD should draw on its existing partnerships with academia, such as through the defense secretary’s Minerva Initiative, to create an academic “center of excellence” to further assess the social implications of advanced surveillance and to propose best practices for deploying such technologies in accordance with the letter and spirit of the U.S. Constitution.

• DoD and the services currently have no framework for assessing the threat of advanced surveillance technologies to the U.S., as it is widely assumed that the United States has an advantage in the field. In our view, the U.S.’s relative advantage in surveillance should be a variable to be measured, not an ordained constant, in future operating environments. The U.S. should, therefore, establish a defense intelligence cell that is dedicated to analyzing foreign intelligence surveillance system achievements, and the Office of Net Assessment should produce periodic analyses of the relative U.S. advantage.

TIMOTHY J. JUNIO is a doctoral candidate in political science at the University of Pennsylvania. He has worked as a consultant to the Office of Secretary of Defense on long-term planning and strategy issues. JONATHAN PROTZ is an assistant professor of mechanical engineering and materials science at Duke University. Junio and Protz worked together on the defense secretary’s policy planning staff in 2008 and 2009.