Jane Vaynman
Introduction
How will emerging technology affect whether states are able to use arms control as a foreign policy tool to manage competition? Will advances in artificial intelligence (AI), for example, make it any easier for states to agree to limit conventional weapons, nuclear capabilities, or risky behaviors? Much attention has previously been paid to the role that international cooperation might play in controlling the spread of emerging technologies themselves. But irrespective of the prospects for such efforts, emerging technology is likely to also play an important role in arms control by affecting the capability of states to monitor and verify compliance. Analysts have already started to think about how technology can improve monitoring,1 leading to intuitions that having greater access to information about compliance will make states more likely to sign agreements. However, more information is not always better, and under some conditions, more effective monitoring may actually undermine cooperation efforts. In assessing the effects of emerging technology on arms control, it is important to consider the countervailing impacts of information collection on state security interests.
Arms control agreements allow states to avoid the costs of an arms race or, more generally, of a status quo in which both sides expend resources or take on risks in an effort to gain a security advantage over the other.2 As such, agreements can be beneficial even for competing states, especially in cases where arms racing essentially maintains the status quo. The most well-known “cost” of arms control — and the reason that is usually presented for why states fail to sign agreements — is the risk of cheating.3 States can face severe security threats if an adversary secretly violates a deal and gains a military advantage. In theory, the risk of cheating could be addressed through monitoring. If one side can see everything the other is doing, there is no way to gain a temporary advantage. Violations would immediately be detected and the other side could respond with its own arms build-up. The solution for getting countries to sign more agreements, it would seem, would be to increase monitoring and transparency in order to decrease the fear of cheating. However, this approach misses an often-unappreciated side effect: increased risks to a state’s security that are created through the effective implementation of a deal.
States collect information to verify agreement compliance, but the same information can also aid intelligence and espionage. The behaviors involved in observing agreement-mandated limits or restrictions — such as inspectors visiting a site where capabilities are being manufactured or destroyed — could allow the observing state to collect information about potential military vulnerabilities of its opponent.4 The collection of this additional information may be intentional, in the form of military espionage. Or, the information needed to verify the agreement may serve a dual purpose, making the collection of the additional information unavoidable. For example, to verify that a weapon is not deployed at military bases, states would have to reveal the locations of those bases, making them possible targets in a future military conflict. While the actors involved in arms control monitoring, such as government agencies or international organizations, tend to dislike their activities being referred to as intelligence collection, both practitioners and scholars recognize, at least in part, the connection between monitoring and intelligence gathering.5 To varying degrees, where monitoring compliance also reveals vulnerabilities, there is a trade-off between transparency and security, where the costs to security may well outweigh the benefits that result from a deal.6
A transparency-security trade-off is a feature of any agreement with monitoring and verification provisions. The severity of this trade-off — i.e., the degree to which transparency creates security costs — depends on a number of factors.7 For example, these factors include whether different types of capabilities are co-located and the extent of existing openness into a state’s security establishment. Another particularly important factor is the tools that are available for collecting compliance information. If compliance is verified by, for example, human eyes looking at a battleship, the state being inspected would be concerned about what else that person might see at a naval base while arriving, leaving, or even while collecting dust on their shoes.8 Changes in the way that monitoring is conducted, caused by innovation in the technology of information collection and information processing, are likely to affect the transparency-security trade-off, and in doing so, have an impact on when states come to an agreement.
This article expands on the concept of the transparency-security trade-off in arms control agreements by considering how the trade-off may change if emerging technologies such as AI and advanced satellites start to play a bigger role in security cooperation. Although there has been increasing scholarly focus on the effects of emerging technologies on other outcomes of interest in international security, such as the nature of conflict,9 strategic stability,10 coercion,11 and crisis escalation,12 far less attention has been paid to the possible effect on cooperation tools and agreement design.13 Recent scholarship that does directly explore the intersection between emerging technology and agreement verification focuses almost entirely on the benefits for transparency without considering the potential for increased security risks.14 At the same time, scholars have begun to explore how the spread and availability of technologies such as satellites have an effect on information asymmetries between the government and the public.15 Such developments have implications for informational shifts in agreement design as well. Indeed, the potential for technology to affect treaty monitoring is recognized even in treaties themselves. Treaties like Open Skies and the Comprehensive Nuclear Test Ban, which go further than most in specifying which technologies will be used for monitoring, note that new technologies may be considered by the parties in the future.16
A closer consideration of possible changes in monitoring capabilities reveals several important implications for assessing the impact of technology on prospects for agreements. First, technology that allows for improvement in unilateral information collection — collection that can be conducted effectively without consent from the monitored state — will make formal agreements more likely by creating areas of potential cooperation where the security-transparency trade-off does not apply. Second, technology that allows for both greater control over specifically what information is collected through monitoring, and greater openness in revealing monitoring methods to the adversary, will decrease the severity of the trade-off. Having demonstrable control over collected information can mitigate the risk of additional security information being collected in the course of agreement monitoring and, as a result, make it easier for states to sign agreements. Conversely, technology that allows states to expand the scope of information collection, or makes it difficult for states to reveal how information collection works, will exacerbate the trade-off. Finally, technology that enhances a state’s capability to conceal or deceive will make agreements less likely by creating a need for higher transparency to effectively monitor compliance, which will, in turn, create greater security costs.
Together, these key dimensions — unilateral monitoring, demonstrable control, and concealment — create a systematic and generalizable framework for assessing the possible effects of any new monitoring technology on arms control. This article investigates four emerging technologies that have some of the clearest applications in arms control monitoring: small satellites, drones, AI, and additive manufacturing (or “3D printing”). The analysis shows how technological changes affect both sides of the transparency-security trade-off, suggesting that agreements will be likely under some conditions and impossible in others. Perhaps counter-intuitively, more effective monitoring does not mean better prospects for arms control. Based on the assessment of specific technological developments presented in this article, the conclusion is a mixed one, with considerable pessimism. Even though capabilities such as small satellites and AI would allow for more effective verification, states may be less likely to sign agreements that utilize some emerging technology tools out of a fear that new security vulnerabilities will be revealed and exploited.
This article first builds on the theory of the transparency-security trade-off by identifying conditions under which technological change would be expected to affect the severity of the trade-off and by developing a framework for identifying implications for cooperation prospects. It then uses this framework to assess the possible effects of several emerging technologies, highlighting in particular where a technology has countervailing effects along different dimensions. The conclusion discusses how this framework can empower analysis of both future developments and other potentially relevant technologies.
Monitoring Capability and the Transparency-Security Trade-Off
Information plays a dual role in arms control agreements. It can provide assurance of compliance or indicate violations, but it can also affect the balance of power between states. Classic arms control literature focused almost exclusively on the benefits of information for detecting cheating and the need for agreements to stipulate inspections when unilateral forms of information collection were insufficient to provide transparency.17 Work that notes the role of national intelligence agencies in agreement verification has focused on the benefits of these tools for assuring compliance,18 rather than on the security costs that both an opponent’s intelligence capabilities and international monitoring agencies might create for the monitored state. More recent research has focused on the impact of agreement-related information on a state’s relative military advantage. For example, Allison Carnegie and Austin Carson show that states hesitate to reveal that they have detected agreement noncompliance, when doing so will reveal secrets about their intelligence capabilities and will potentially forfeit a future military advantage.19
In the article “Why Arms Control is So Rare,” authors Andrew Coe and Jane Vaynman focus directly on information dilemmas in agreements themselves.20 They introduce the concept of the transparency-security trade-off, whereby information needed to monitor compliance also creates costs to a state’s future security. However, Coe and Vaynman provide only preliminary ideas about the conditions under which the trade-off may be more or less severe, in terms of the degree to which marginal increases in transparency create security costs. The conditions they discussed focused on the nature of the states involved and on the characteristics of the capabilities that are being limited, but not on the different ways that monitoring itself can be carried out.
Changes in the way that monitoring is conducted, caused by innovation in the technology of information collection and information processing, are likely to affect the transparency-security trade-off, and in doing so, have an impact on when states come to an agreement.
Building on this concept, the current article identifies three monitoring-related factors that are likely to affect the transparency-security trade-off, and in doing so the potential for reaching arms control agreements. First, the degree to which a technology enhances unilateral monitoring capabilities will determine whether the trade-off applies in particular cases. Second, the degree to which a technology alters the ability of states to demonstrably control the amount and quality of information collected can affect the severity of the trade-off. Finally, technological changes may affect not only the capacity to collect information about compliance but also how states conceal their activities, which in turn affects transparency needs in an agreement.
The core logic of the transparency-security trade-off is that in order for states to prefer an arms control deal over a costly status quo — such as an arms race or conflict — an agreement needs to offer enough transparency into the behaviors of other states to be able to detect agreement violations, while maintaining enough secrecy such that their relative balance of power does not alter. These requirements are in tension, because greater transparency can undermine secrecy. How much transparency is needed to ensure compliance can depend on a number of factors, including the physical nature of the capability or action being limited, what would constitute a militarily significant violation, and the potential costs of failing to detect a violation. Relatively less transparency is needed, for example, to observe military capabilities that are physically large. Relatively more transparency would be needed if a violation — such as secretly building a nuclear weapon — would have a very large effect on the future balance of power.
When states have sufficient transparency to be able to detect violations on their own, without any additional information provided or access granted by the other side, agreements with essentially no monitoring provisions are possible. This kind of “unilateral monitoring” capability can detect military secrets even without an arms control agreement. For example, satellites already collect information on an adversary’s military bases. Using those images specifically to verify agreement compliance would not change what can be collected.21 Agreements that rely on unilateral monitoring avoid the transparency-security trade-off because they do not introduce any new threats to a state’s security.
If unilateral monitoring is insufficient to meet the transparency requirement, states will need agreements that allow them to collect additional compliance information. These kinds of agreements include extra monitoring tools that one state would not be able to use without the consent of the other state. On-site inspections are the best example of this type of tool because states need to willingly allow inspectors to access the relevant sites. Similarly, tools such as cameras or measuring equipment can only be installed with the monitored state’s participation. As intrusive monitoring increases, the risks to security also increase. But the extent of that marginal increase in security risk can vary. If small increases in the intrusiveness of monitoring expose significant vulnerabilities, the trade-off is severe. Conversely, if increases in monitoring expose few vulnerabilities then the trade-off is mild. The more severe the trade-off, the more likely it is that states will prefer the status quo over signing a deal.
Effects of Changes in Monitoring Capabilities
When considering the monitoring capabilities themselves, several challenges emerge. First, the scope of agreements that rely on unilateral capabilities is limited by what those capabilities can detect. Any changes in those detection capabilities would expand the range of agreements that can be signed. The more that states can observe unilaterally, the more willing they will be to cooperate on particular arms control limitations. If a state can only observe large armaments, such as ships or missiles, those may be the only capabilities over which it is willing to negotiate. If observation capabilities were to change to make smaller objects detectable, then agreements about smaller armaments, which still do not rely on intrusive inspections, would be more likely. This could apply both to the types of military armaments that are being limited and to the extent of the limitations.
There are several ways in which unilateral monitoring can improve. First, there could be improvements that allow for the detection of ever-smaller items or behaviors. Second, observation can become more pervasive, allowing the state to detect changes over shorter periods of time. Third, if the costs of intelligence gathering are reduced, states can increase the volume of information that they can collect. These improvements can result from the same technology. For example, a new technology might enable a state to go from being able to see an adversary’s military installation to observing vehicles next to it, and from seeing vehicles occasionally to being able to detect that they tend to arrive every Monday, and all at half the cost of the prior capability. Any technology that has applications in national intelligence would also have implications for arms control. In addition to improvements in national capabilities, advancements in monitoring tools provided by an international organization could enhance a state’s ability to observe compliance without the acquiescence of the party being observed. The International Monitoring System of the Comprehensive Nuclear Test Ban Treaty is a good example. This system includes a range of sensors — seismic, hydroacoustic, infrasound, and atmospheric radionuclide — that allow it to monitor nuclear testing around the globe. The logic of technological effects on unilateral monitoring suggests that those emerging capabilities that allow for persistent and fine-grained observation from well outside a state’s borders would make future arms control agreements more likely.
The second challenge is more complex. For agreements that need to rely on intrusive inspections to attain a sufficient level of transparency, the fundamental problem is that states will use information gathered in the process of verifying agreement compliance to then gain a military advantage. To some extent, this problem cannot be solved. When the very same information needed to verify compliance also creates vulnerabilities, the trade-off is immutable. For example, if an agreement reveals the locations of military bases for monitoring, that information itself also provides the adversary with a list of targets for a future military conflict.
However, some vulnerabilities can be mitigated by limiting what information inspections or other information-gathering tools can collect. This is the case when risk is associated with extra, non-agreement-related information that might be collected in the process of conducting agreement verification. For example, sensing equipment might be able to pick up information on capabilities beyond those being limited by a given agreement, or human inspectors might look around and observe other significant details at a site they are inspecting. Collecting extra information is of course perfectly rational from an adversary’s point of view, and there is every reason to expect that espionage will happen if it is possible. U.S. and Soviet arms control inspection teams routinely had an intelligence component, and while neither side advertised this activity, both quietly accepted it as a part of the process.
Research on how to verify limits on specific capabilities seeks to solve precisely this security problem. For example, contemporary research and development on technologies for verifying limits on nuclear warheads has focused on making determinations while protecting weapon-design information, using approaches such as information barriers aimed at removing “sensitive information while providing as much data as possible.”22 Another approach has focused on “zero-knowledge proofs,” which provide a way to confirm that two objects are the same without revealing their physical characteristics.23 Some security costs can therefore be mitigated when there are ways to narrow the scope of observation to information on agreement compliance and little else.
Technological change will play a role in this dynamic if it affects the degree of demonstrable control over the quantity or quality of the information collected in the course of intrusive monitoring. A monitoring tool that provides a high degree of control creates an ability to demarcate the line between what will be revealed and what will be kept secret at whatever point that line is negotiated. Control might be achieved by limiting the amount or scope of the information that can be collected, or through an ability to collect information at a specified quality or level of granularity. For example, a technology that allows states to build a visual measurement tool with a range of fixed resolutions suggests a higher degree of control compared to a tool that can only be built with one resolution, or a tool that does not allow for a specified level to be set for different circumstances. Greater control might also be achieved through a technical ability to determine when or how frequently information can be collected. The ability to turn data gathering on and off is one example of control with regard to frequency. If a monitoring technology provides a high degree of control, then states will have greater assurance that security information beyond the scope of the agreement will be safe while more compliance-relevant information is revealed.
Control over information is only part of the challenge, however. In order to alleviate security concerns, states need to know what the other side is capable of collecting and what they have, in fact, collected in the process of agreement monitoring. In other words, there needs to be demonstrable proof that the agreed-upon level of precision and scope in information collection is both technologically possible to attain and is being implemented as agreed. Otherwise, states would have every reason to suspect that high-precision sensing tools are actually being used to collect information not relevant to the agreement. However, to assure a monitored state that a tool is only collecting designated data, a monitoring state might have to reveal details about how the tool works, which itself may reveal sensitive information about the monitoring state’s capabilities and technological advancements. Using a well-known and even outdated technology for monitoring makes it easier for states to have common knowledge about the security risks that it might create.
Since demonstrable control is a concept being introduced here, a further illustration is helpful. Suppose the monitoring “technology” in question is a human inspector. It is possible to have a fair amount of control over where that person goes, when they arrive at a site, or how long they stay, but there is no control over that person’s ability to turn his or her head. Any information that is within eyesight can be collected. This is in contrast to something like a mounted camera, where both the resolution and the line of sight can be controlled. The capacities of human vision are of course common knowledge and so the technology is demonstrable in that sense. But it is only partially demonstrable because even if the host state confiscates all the inspector’s notes to check for espionage, there is no way to demonstrate what the inspector remembers.
Different emerging technologies may increase or decrease demonstrable control when it comes to agreement monitoring. If a new monitoring technology is itself the object of secrecy, or is poorly understood by an adversary, then demonstrable control will be low and uncertainty about possible security risks will increase. On the other hand, if either the nature of the technology or the steps taken to integrate it into monitoring efforts make it easy to demonstrate what information was collected, then uncertainty about potential intelligence vulnerabilities could be mitigated. Indeed, one interesting advantage of the monitoring systems for the Comprehensive Nuclear Test Ban Treaty — including both the global sensors and the on-site inspections — is that the collection methods and technical details of specific tools are well understood by member states, due both to widespread access to the technologies and to efforts by the Comprehensive Nuclear Test Ban Treaty Organization to communicate and even practice verification scenarios with member states.24 Any assessment of the potential impact of a new technology should therefore investigate the degree to which the technology allows for demonstrable information collection.
While improvements in a state’s ability to hide violations would make an agreement more difficult to sign, an emerging technology that improves a state’s ability to conceal sensitive information that lies beyond the scope of the agreement would improve the prospects for a negotiated deal.
The third challenge to consider is how an emerging technology affects not just monitoring capabilities but also a state’s ability to conceal information. The prior two factors — unilateral monitoring and demonstrable control — largely considered the information question from the perspective of the state doing the monitoring. For concealment, it is helpful to assume the perspective of the monitored state. This state has an interest in hiding violations if it is indeed in noncompliance. Even if it is complying, the monitored state has an interest in hiding other security-related information. If the state’s ability to hide violations improves, and both sides know this to be the case, then the monitoring state will require a higher minimum level of transparency in order to reach the same level of assurance that agreement terms are being followed. Agreements that were previously verifiable with unilateral means alone might now require intrusive inspections, and inspections might now need to be more persistent or involve greater access. Technological developments can improve the ability to hide violations by making the capability in question smaller, more portable, or more effectively shielded from observation by changing its appearance or other physical properties. Or, a technology may make it possible for violations to happen more quickly, increasing the possibility that they will go undetected.
While improvements in a state’s ability to hide violations would make an agreement more difficult to sign, an emerging technology that improves a state’s ability to conceal sensitive information that lies beyond the scope of the agreement would improve the prospects for a negotiated deal. Traditional concealment efforts have often involved physically covering some capabilities, placing them inside buildings out of sight, or moving them at night. More advanced concealment capabilities have included electronic interference, such as jamming the collection of signals intelligence. New technology tools might allow states to selectively block detection by creating information barriers, or even improved physical barriers, between weapons that are limited by the treaty and those that are not. Technology may also provide states with improved ways of creating misinformation, or creating a greater volume of it, essentially diluting the benefits that might come from intelligence gathering during agreement monitoring.
In general, improvements in concealment capabilities are a double-edged sword in terms of their effect on prospects for arms control agreements. A given technology could help to conceal violations, sensitive information, or both. Improved concealment of violations would make agreements less likely by driving up the transparency needs, while better concealment of sensitive information could make agreements easier to design by mitigating severe security trade-offs. It may be very difficult to discern which effect is greater. While it is possible that a technology will affect one kind of concealment more than the other, intuition suggests that most innovations would apply to both.
Analytic Framework
The discussion above leads to a set of implications about the effect of emerging technology on the severity of the transparency-security trade-off and the ease of creating arms control agreements. First, improvements in unilateral monitoring capabilities can increase the number of cases that avoid the transparency-security trade-off and make arms control agreements more likely. Second, monitoring technology that provides greater demonstrable control over information decreases the severity of the trade-off and will make agreements more likely. By the same token, technology that diminishes either the degree of control or the ability to demonstrate the information that was collected to other parties increases the severity of the trade-off and will make agreements less likely. Third, technology that allows the monitored state to better conceal violations increases the minimum transparency level needed for cooperation and will make agreements less likely. And finally, technology that improves the monitored state’s capacity to conceal security capabilities not limited by an agreement decreases the severity of the trade-off and will make agreements more likely.
Importantly, these implications do not operate in isolation. The same technological change that affects unilateral monitoring technology may also affect concealment capabilities. Thus, the effects of emerging technologies need to be evaluated along multiple pathways, which may have opposing effects. The analytic challenge is to estimate the magnitude of those effects and come to an overall assessment of the impact on prospects for cooperation. However, in some cases, there is simply not enough data to make a judgment on which direction will prevail. The implications outlined above provide a framework that can, even in cases of limited information, organize the assessment process by identifying the key factors to consider and the possible negative and positive effects. In this way, it allows us to avoid the mistake of only considering some impacts of a technology and not others, such as impacts on transparency but not on security. The framework also provides a means of revising assessments as more information about and experience with emerging technologies enters the picture, and it directs our attention to those elements of technological change that will matter for arms control.
Among the technologies assessed in the next section — small satellites, drones, AI, and additive manufacturing — no single definitive answer emerges as to whether the transparency-security trade-off will be mitigated or exacerbated as a consequence of new technology applications in the arms control context. At the same time, assessing new technologies using this framework does strongly indicate that while some technologies are indeed likely to improve prospects for cooperation, others may make it more difficult.
Assessing the Effects of Emerging Technologies
This section surveys several emerging technologies, identifying how they can be used to monitor arms control compliance and where that technology-enabled monitoring may create new security vulnerabilities. The particular set of technologies were selected for analysis to demonstrate the broad utility of the framework. Since not all of the key information trade-offs occur in every case, this range of technologies illustrates how each of the factors discussed above — unilateral monitoring, demonstrable control, and concealment — plays out in specific contexts.
The cases of small satellites and drones represent contemporary advancements of technologies that have been previously used for treaty monitoring and intelligence gathering and are, in that sense, the most likely to be used for similar purposes in the future. The AI case moves further afield from past treaty-monitoring experience but it engages with a technology that is at the center of extensive ongoing research on applications for information processing, including in intelligence. Finally, the additive manufacturing case shows how a technology with other primary uses may also have important implications for generating and collecting information. This case also serves to illustrate some of the paradoxes that emerge, including that a technology that seemingly makes it easier for states to build capabilities that are restricted by a given agreement also potentially enhances the ability to detect whether those violations are taking place.
These four cases notably vary in their overall implications for arms control agreements, from improving prospects for cooperation to hindering them. The analysis therefore allows us to investigate the pathways toward different possible outcomes, rather than focusing on a set of technologies that, for example, are all likely to have positive effects on agreements. While these particular cases are useful for illustrating the full range of information trade-offs and positive and negative implications, the framework is equally applicable to other cases as well. For example, robotics or blockchain technology (which is addressed briefly in the conclusion) would be useful to examine in a longer study.
Notably, the assessment of these four technologies is forward-looking. Apart from satellite technology, which has long been used for arms control monitoring, the discussion focuses on technologies that may be used for monitoring in the future, and are occasionally mentioned in that context in policy discussions or conferences, but are not currently an explicit part of arms control monitoring regimes. Table 1 provides an overview of the assessment of each technology. The sections below discuss the cases in greater detail, evaluating each technology’s impact on unilateral monitoring, demonstrable control over information collection, and concealment.
No comments:
Post a Comment