Sensing Weapons

International law seeks to regulate the movement of weapons, and material that can be used to make military grade weapons, in a range of ways broadly premised on the detection or attestation of movement. Legal attention is directed, for example, to the development, manufacture (or potential manufacture), testing, retention, sale, dissemination, and transportation of weapons of mass destruction (WMD) (a category typically reserved for nuclear, biological, and chemical weapons), as well as their deployment or threatened deployment.Footnote ⁴⁴ Also of international legal concern—and the focus of distinct legal measures and doctrines—are the environmental effects of such weapons’ usage and associated movements of hazardous wastes and toxic substances, as well as the prospect of weapons’ and related materials’ seizure by, or unauthorized rerouting to, nonstate actors.Footnote ⁴⁵ International law is directed, too, toward the movement of other categories of weapons and related materials, including mines and small arms.Footnote ⁴⁶

Rather than survey the range of international laws touching upon the question of the movement of weapons and related materials, during warfare and otherwise, this section will focus on a single instrument illustrative of the kind of sensorium that international law seeks to produce around, and anchor in, such objects: the 1968 Treaty on the Non-Proliferation of Nuclear Weapons (NPT).Footnote ⁴⁷

The Nuclear Non-Proliferation Treaty

The NPT was a product of the Cold War and its fragmentation: a period during which politico-legal perception tended to be oriented around the state and characterized by suspicion and, during the 1960s, growing skepticism (including toward state authority).Footnote ⁴⁸ Conceived as a containment device, in anticipation of states concluding a general disarmament agreement, the NPT was designed to offset small states’ “temptation[s] to exploit the enormous temporary advantage derived from the possession of [nuclear] weapons.”Footnote ⁴⁹ This would, it was hoped, avert the prospect of a “small State or a revolutionary group … set[ting] off a world-wide nuclear war” while those “few highly developed States” possessed of both nuclear weapons, and “a sense of deep responsibility” associated with those, were engaged in negotiations toward general disarmament.Footnote ⁵⁰ Reflecting this predisposition for entrusting small groups of “responsible” states and state representatives with the capacity to sense and act on behalf of many, a Geneva-based multilateral forum known as the Conference of the Eighteen-Nation Committee on Disarmament was charged by the United Nations General Assembly with negotiating the treaty, in the course of which they “took wide soundings from other nations.”Footnote ⁵¹ The text of the treaty, so negotiated and opened for signature in 1968 (simultaneously in Washington, D.C., London, and Moscow), codified a “quid pro quo relationship between two classes of states parties, each class having differing rights and obligations accorded them under the treaty”: nuclear weapon states and non-nuclear-weapon states.Footnote ⁵² The NPT entered into force in March 1970 with forty-three parties, including three of the five states defined as nuclear-weapon states according to the treaty's criteria.Footnote ⁵³ As of March 2017, the NPT had 191 states parties.Footnote ⁵⁴ Only five UN member states have declined to become parties to the NPT: India, Israel, Pakistan, South Sudan, and the Democratic People's Republic of Korea (which withdrew from the treaty in 2003).Footnote ⁵⁵

Characteristic of the NPT is its combination of nuclear nonproliferation goals with the promotion of “peaceful” uses of nuclear technology. Under Article I, nuclear-weapon states parties commit “not to transfer … nuclear weapons or other nuclear explosive devices or control over such weapons or explosive devices,” nor to “assist, encourage, or induce” non-nuclear-weapon states to manufacture or acquire them.Footnote ⁵⁶ No prohibition is thereby imposed on cooperative nuclear weapons development among nuclear-weapon states, nor on transfers by nuclear-weapon states of components of, or information and materials related to, such weapons and devices.Footnote ⁵⁷ Non-nuclear-weapon states parties, on the other hand, are bound by Article II not to receive transfers of, manufacture, or otherwise acquire, seek, or receive assistance in the manufacture of such weapons or devices.Footnote ⁵⁸ Pursuant to this undertaking, non-nuclear-weapon states parties are required by Article III to conclude safeguards agreements with the IAEA to verify fulfillment of their NPT obligations—transfers of “source or special fissionable material” or related equipment to such states can only occur under the rubric of such safeguards.Footnote ⁵⁹ Also provided by the NPT are state commitments to: facilitate exchange in “equipment, materials and scientific and technological information for the peaceful uses of nuclear energy”; and make available the “potential benefits” from peaceful applications of nuclear explosions to non-nuclear-weapon states on a nondiscriminatory basis and “under appropriate international observation.”Footnote ⁶⁰ Further, parties to the NPT also undertake to pursue good faith negotiations toward “a treaty on general and complete disarmament under strict and effective international control.”Footnote ⁶¹

The sensory economy contemplated by the NPT is marked by at least five characteristics. First, this economy is in large part an ocular one. The NPT prioritizes “observation”: a priority also apparent in the system of safeguards prescribed in INCIRC/153, the document adopted in 1972 which set out “The Structure and Content of Agreements between the [International Atomic Energy] Agency and States Required in Connection with the Treaty on the Non-Proliferation of Nuclear Weapons.”Footnote ⁶² The two main “verification” mechanisms which INCIRC/153 provides are: (1) recordkeeping and reporting on the part of non-nuclear-weapon states, on the one hand (including maintenance of a “system of accounting … and control”); and (2) the conducting of “observations” and “examination” by the IAEA through the medium of “Agency inspectors,” on the other.Footnote ⁶³

Even so, this ocular economy is not uniform or comprehensive in its sweep. Rather, it is telescopic; it is predicated on the privileging of, and attribution of depth and meaning to, certain sites and scenes, and the assumption of a partially obscured, yet penetrating, confidential viewpoint on those. Access to these sites is conditional on adherence to a carefully moderated style of address or exchange between inspector and inspectee. This is comparable to the generation of a perspectival illusion of depth on a two-dimensional surface through the guided interaction of painter and viewer, or (in the case of the written word) writer and reader, and the “parceling out of the visible and the invisible” effected thereby.Footnote ⁶⁴ In both settings, the propriety and impropriety of certain forms of conduct and human-nonhuman interaction are matters to which particular attention is devoted in order to produce the desired perceptive effect. Consider, for instance paragraph 87 of INFCIRC/153, which provides that inspectors “shall carry out their activities in a manner designed to avoid hampering or delaying the construction, commissioning or operation of facilities.”Footnote ⁶⁵

By this means (as a second noteworthy feature), the NPT constitutes a class of officials charged with registering worldly phenomena and maintaining the global sensorium on the behalf of all. Throughout the fin de siècle of the 20th and the early 21st century, such figures were popularly and credibly embodied by the directors general of the IAEA: Dr. Hans Blix (who headed the IAEA from 1981 to 1997 and later led the UN Monitoring, Verification and Inspection Commission (UNMOVIC) in charge of monitoring Iraq); and Dr. Mohamed ElBaradei (who headed the IAEA between 1997 and 2009).Footnote ⁶⁶ Blix was at the center of public controversy preceding the 2003 invasion of Iraq.Footnote ⁶⁷ Later, he was honored with a range of awards attesting to the stance he took at that time as well as his broader service, including the Sydney Peace Prize in 2007, and a Fulbright Prize in 2014.Footnote ⁶⁸ ElBaradei was a recipient of the 2005 Nobel Peace Prize (together with the IAEA itself) and much celebrated as a truth-teller in an “age of deception.”Footnote ⁶⁹

The world of the NPT is to be made so through the movement and sensory intake of specially anointed bodies such as Blix's, ElBaradei's, and those of the teams of inspectors that they have led. To the task of detecting worldly phenomena, these figures are expected to bring a particular set of dispositions, inculcated by training. Hans Blix is reported to have told trainees in February 2003 that an inspector should be “driving and dynamic—but not angry and aggressive”; “ingenious—but not deceptive”; capable of “keeping some distance—but not arrogant or pompous”; and able to maintain a facility for the “craft and tools of inspection.”Footnote ⁷⁰ This is a role to which IAEA inspectors are also expected to bring formal qualifications—credentials that must be made known to the state in question and may inform that state's acceptance or not of the inspector's designation.Footnote ⁷¹ Even so, their capacity to draw on banks of expert knowledge does not sustain IAEA inspectors’ authority under the NPT; this is not a case of some epistemic community bearing down on the international legal order by force of scientific expertise.Footnote ⁷² The state may decline to accept an inspector's designation and the IAEA may withdraw that designation (at the request of the state of inspection or at its own initiative) at any time without explanation.Footnote ⁷³ Rather, it is through the generation, recording, processing, and transmission of sense data in a manner and style appropriate to their office that the currency of IAEA inspectors’ authority may be kept in circulation.Footnote ⁷⁴

Of particular significance to the NPT's operation in this respect is the passage of texts through the hands of such official sensors. The NPT is, in this sense, a manual arrangement—a third noteworthy aspect of its sensory economy. It contemplates the physical handling and examination of descriptions, inventories, documents, reports, files, and copies of files.Footnote ⁷⁵ And their analysis, for the most part, is made contingent on the human faculties and intuitions of those inspectors proposed by the director general of the IAEA (and accepted by the state in question) as they engage in the ritualized “performance of inspections.”Footnote ⁷⁶

A fourth characteristic of the sensory economy of the NPT is that it presumes that the small is revelatory of the large. This is apparent in the focus on sampling evident in both the NPT and associated texts: INFCIRC/153, discussed above; and the later Model Additional Protocol, INFCIRC/540, adopted in 1997 in the aftermath of the First Gulf War.Footnote ⁷⁷ The latter was designed to bolster the powers of IAEA inspectors and remedy the weakness of the safeguards regime that had become apparent in the unearthing of a significant, clandestine nuclear weapon program in Iraq, maintained despite it being a non-nuclear-weapon state party to the NPT.Footnote ⁷⁸ “In order to ensure optimum cost effectiveness,” INFCIRC/153 provides that use should be made of “statistical techniques and random sampling to evaluate the flow of nuclear material.”Footnote ⁷⁹ Similarly, INFCIRC/540 provides for the “collection of environmental samples” (both location-specific and wide-area samples) alongside visual observation, the use of radiation detection devices, and “other objective measures which have been demonstrated to be technically feasible.”Footnote ⁸⁰

These international legal instruments thus rely on that “economical metonymy … by which a tiny part allows the grasping of the immense whole.”Footnote ⁸¹ Through the collection, transfer, analysis, and description of samples, the NPT and associated agreements provide for the specifics of the localities on which they focus to expand and change. In this way, samples yield far more sense data than they might otherwise and circulate along a far more expansive route than they might traverse by other means.Footnote ⁸² In particular, they come to speak to the question of a state's adherence to, or diversion from, the pathway of nuclear energy's peaceful uses.Footnote ⁸³

Fifth, and finally, the NPT and related protocols establish a proprioceptive sensory scheme: that is, they generate information about bodies’ and objects’ positions in and movements through space, and direct such movements accordingly. They do this in a narrative register: providing for the creation and circulation of “descriptions” of the “general arrangement,” “layout,” and “features” of facilities relevant to the safeguarding of nuclear material and of the uses and locations of nuclear material, for instance, and of “the sequence of the actions taken” to prepare certain inventories.Footnote ⁸⁴ They do this also through measurement and physical containment measures (such as “seals and other … tamper indicating devices”) and reporting of transfers of nuclear material.Footnote ⁸⁵ Proprioceptive information is gathered and organized, too, through provision for IAEA inspectors’ access to certain locations, without “undue concentration on particular facilities.”Footnote ⁸⁶ Again, bodies and information must circulate along specified paths in order to generate the references and assurances demanded of them by the NPT and related agreements.

Within and around the sensorium just described, there are also broader and narrower configurations of sense and capacity related to it. There is the broader distribution of “responsibility” alluded to above, whereby nuclear-weapon states parties to the NPT are vested with certain dispositions and freedoms (individual and collaborative), while non-nuclear-weapon states parties are “safeguarded” and shepherded along a path of “peaceful use,” risk-laden and volatile. There is also a narrower arrangement of confidences nestled within this sensorium, whereby “confidential information” must be subject to a “stringent regime to ensure effective protection against disclosure … .”Footnote ⁸⁷ Within the sensorium of the NPT, the incidence of “proprietary or commercially sensitive information or design information” elicits a tailored sensory approach, overlaid with imperatives associated with the market.Footnote ⁸⁸

Far from setting all things and people on a single, common plane, or purporting to gaze upon them from some higher vantage point, the NPT and related legal instruments follow an array of interlocking sensory routes and set in motion a variety of sensory practices and interactions, always anticipating the prospect of failure, omission, obstruction, and friction. The IAEA “shall not mechanistically or systematically seek to verify the information” reported to it, INFCIRC/540 stipulates, despite that instrument's apparent effort “to transform IAEA inspectors from accountants to detectives.”Footnote ⁸⁹ Rather, the tenor of those inspectors’ sensory work is to remain intensely human, guided by the intuitions, stylized dispositions, and multisensory capacities of the body.

Relatively few dimensions of this sensory work will be familiar to those to whom it is broadly of concern. Not many people understand the specialized vocabulary, equipment, or protocols of inspection under the NPT. Opportunities for public witnessing of this work are extremely limited. Nonetheless, the operation of the NPT and related instruments depends on the continuous, ritualized circulation of sense data and the disposition and status of the bodies, objects, and procedures through which that data are understood to pass. Over and above any question of the juridical soundness of its principles, their representation of social habit, or their defense of elite interests, it is this sensory economy which sustains what one scholar of the NPT has referred to as the “normative and legal weight of the [NPT] regime.”Footnote ⁹⁰

The manual reports and telescopic inspections, for which the NPT and related instruments provide, anticipate objects and machines playing significant roles; they occasion a wide range of human-nonhuman alliances. These involve documents, files, transportation technologies, “instruments and other measuring and control equipment,” “tamper-indicating devices,” surveillance technology, “radiation detection and measurement devices,” communication systems, including satellite systems, and the many different types of technology described in the Annexes to INFCIRC/540.Footnote ⁹¹ Thus, the heightened automation of sense—and expectation of its further automation—which we will turn in Part IV of this article does not represent abandonment or replacement of the sensorium installed by the NPT and related instruments so much as its extension and reorientation, with which significant redistributions of competence and authority are associated. Before turning to that redistribution, however, let us examine a further dimension of the sensory economy elicited and maintained by international law and policy—that surrounding the detection of refugee migration.

Sensing Mass Migration

Mass movement of persons is not per se a phenomenon of which public international law doctrine and international institutions seek to keep track. However, when persons moving in significant numbers are identified—or self-identify—as refugees in large part, their movement is taken to warrant international legal notice, monitoring, and response. The large-scale claiming or determination of refugee status is symptomatic for international lawyers of “breaches of human rights standards or … the collapse of an existing social order in the wake of revolution, civil strife, or aggression … ” and so of the prospect or actuality of politico-legal disorder.Footnote ⁹² Such phenomena trigger international legal obligations on the part of states to take measures to try to bring those migrating within the protective embrace of a sovereign state and thereby restore a sense of global order premised on the coexistence of such politico-territorial units. They also trigger authority and entitlement on the part of international institutions, especially the UNHCR .Footnote ⁹³

Like the sensory economy manifest in international law and policy regarding nuclear nonproliferation, that surrounding refugees is designed to be sensitive to, and (ideally) preventative of, threats of violence and disorder. Once again, however, this article offers no more than a snapshot of that sensory economy. It does so through a focus on selected instruments representative of international law and policy in this field from a certain period. In this case, the instruments in question are two UNHCR policy manuals from the mid-1990s. The first, published by the UNHCR in January 1994, is entitled Registration: A Practical Guide for Field Staff (1994 UNHCR Guide).Footnote ⁹⁴ This is designed to guide UNHCR field workers in grappling “with the problem of collecting data about refugees who need protection or assistance,” primarily through “‘registration’ or recording information about individuals or families that will serve identification and programming purposes.”Footnote ⁹⁵ A related manual, published in June 1997, is concerned with the design and implementation of systems for the distribution of resources to those seeking humanitarian assistance from the UNHCR: the Commodity Distribution Guide for Field Staff (1997 UNHCR Guide).Footnote ⁹⁶

In contrast to the status and aims of the NPT, both of the aforementioned manuals present practical knowledge drawn from the experience of UNHCR field workers and extract from that “current best practice which must be adapted to the particular circumstances of your operation” as of the mid-1990s, or “a kind of ‘toolbox’ … [to] be applied or adapted to various situations … .”Footnote ⁹⁷ As such, they represent an instance of what international lawyers commonly characterize as “soft law”: norms that tend to guide behavior pedagogically and demonstratively, rather than prescriptively or by recourse to formal sanction.Footnote ⁹⁸

Sensing Refugees in the Mid-1990s

Six features of the sensory economy advanced in the 1994 and 1997 UNHCR Guides merit highlighting alongside the practices described above and below in Part IV.

First, both the 1994 UNHCR Guide and the 1997 UNHCR Guide emphasize the direct participation, of those whose susceptibility to UNHCR authority and eligibility for UNHCR assistance are being assessed, in the economy of their detection and evaluation. The process of detecting refugees’ presence and ascertaining their needs is cast as an interactive and, at least potentially, reflexive process. The objects or targets of sense are expected to become sensing subjects themselves, cultivating and contributing expertise on their own condition.Footnote ⁹⁹ The “we” to be interpellated by the 1994 UNHCR Guide and the 1997 UNHCR Guide, and associated circulation of data, is thus configured quite differently to the “we” called forth by the NPT, which tends to be assembled among, around, or in opposition to states.

Accordingly, the 1994 UNHCR Guide advocates “actively promot[ing] community understanding and responsibility in registration activities … .”Footnote ¹⁰⁰ It emphasizes the “usefulness” of “working with the refugee population over a prolonged period” and “promoting community responsibility and participation in all stages of the process”—both for the task of registration and in the forging of “durable solutions for the population concerned.”Footnote ¹⁰¹ UNHCR field workers are advised to “first … consider whether, given the necessary support and tools, [the target population] could organize a satisfactory registration system for themselves” and where this is not possible, to retain the local population's involvement.Footnote ¹⁰² “[E]ven in the most uncooperative situation,” the 1994 UNHCR Guide insists, “there should be involvement of both parties [that is, the UNHCR and the target population] in the planning and information phase” of refugee registration.Footnote ¹⁰³ “It is vital,” the 1994 UNHCR Guide continues, “to convince the population that it is in their own interests to cooperate with a registration exercise.”Footnote ¹⁰⁴

Similarly, one of the 1997 UNHCR Guide's “key points” is that those responsible for distributing assistance must involve and inform the refugees in question in the process of determining and addressing their needs:

In contrast to the NPT's reliance on state reporting, sensory inputs from refugees are not to be mediated through the leadership of sovereign states, nor even through their own leadership; would-be leaders within refugee communities attract circumspection on the part of the UNHCR. “Any well-designed distribution system must provide a way to keep refugees continuously and directly informed (i.e., not just through the refugee leadership),” the 1997 UNHCR Guide insists.Footnote ¹⁰⁶

Second, the sensory economy that the 1994 UNHCR Guide seeks to establish and maintain is not primarily an ocular economy premised on observation. Instead, it is in large part a tactile economy, premised on the marking, mingling, touching, and handling of bodies. For purposes of “fixing” a refugee population, for instance, or “defining the target group for [initial] registration,” the following options are presented in the 1994 UNHCR Guide: affixing wristbands, handing out fixing tokens, temporary cards, or marking hands with gentian violet antiseptic dye—either by moving “house to house” among a refugee population, or by having that population pass through designated registration points. A further alternative entails requiring refugees to enter an enclosure for a specified period for purposes of registration staff, bearing identifying armbands, issuing fixing tokens to them “simultaneously.”Footnote ¹⁰⁷ For purposes of estimating the size of a refugee population, the 1994 UNHCR Guide recommends “organiz[ing] a headcount in the camp or centre” or “during an influx at entry points, bridges or transportation points.”Footnote ¹⁰⁸ Aerial photography is recommended as a way of gaining “a ‘bird's eye view’ of the camp,” to “prepare maps and to help estimate the population.”Footnote ¹⁰⁹ Nonetheless, those who count are expected to move among the bodies that they are seeking to record—touching and marking them, adorning them, handing them tokens.

Third, the 1994 UNHCR Guide urges those working in the field not to individualize the targets of their sensory work. Rather, individuals are to be assembled in groups and recorded, addressed, and supported in some collective configuration—as a population, especially—for purposes of their engagement with the international legal system in question. In the 1994 UNHCR Guide, those planning to register refugees are advised to “consider … the characteristics of the population,” including “[n]ationality,” “ethnic origin(s),” and the “[p]revious refugee experience of [the] population” as a whole.Footnote ¹¹⁰

In detecting a population as such, the 1994 UNHCR Guide also provides for the distribution of capacity unevenly across it. It is “usually necessary,” the Guide stipulates, “to identify those who are vulnerable, who are often the least likely to come forward and make their needs known”; likewise, the UNHCR must “[d]efine the number of people … who have special needs.”Footnote ¹¹¹ The input of the self-sensing refugee alluded to above is, accordingly, to be distrusted; his or her sensory capacity is conditioned and must be supplemented and verified, especially when combined with vulnerability. In the 1994 UNHCR Guide, scarcity, sufficiency, incapacity, and capacity are to be distributed across a population through the mechanism of registration, although the basis for doing so is the “family registration form.”Footnote ¹¹²

In the 1997 UNHCR Guide, the primary groupings into which refugees are corralled for recording purposes are those of family and nationality, the latter of which is determined by reference to citizenship, and other “group[ings] of beneficiaries based on their original administrative or social structure.”Footnote ¹¹³ According to the 1997 UNHCR Guide, “the distribution system should aim to support the family … . Only in exceptional circumstances, and for short periods, should the focus of assistance move from the family level to individuals within the family.”Footnote ¹¹⁴ In the case of “large populations,” the 1997 UNHCR Guide recommends “it may be useful to divide the camp into zones (or blocks),” but these are envisaged as being comprised of “groups of families of the same size.”Footnote ¹¹⁵

A fourth noteworthy feature of both the 1994 and 1997 UNHCR Guides is their framing of the economy of sense in transactional terms: as a series of “deals” whereby participation in “fixing, documentation and verification,” and associated data yield, are traded for certain “benefits” flowing to the “target population.”Footnote ¹¹⁶ More than a matter of establishing reliable “ground truth” data, the 1994 UNHCR Guide envisages the determination of refugee numbers and needs as a matter of ongoing negotiation around “interests” and “investment[s],” approached with a view to the “honour[ing]” of “promises.”Footnote ¹¹⁷ These deals in data depend upon persuasion, the 1994 UNHCR Guide makes clear, for which purpose the prospect of greater fairness for the “vast majority of people” must be “publicize[d]” in order to “outweigh initial resistance from a minority with vested interests.”Footnote ¹¹⁸

The 1997 UNHCR Guide counsels fieldworkers in comparable terms as follows: “[t]he needs, and your assessment of them, will change with time, but at any given time it is assumed that you will have reached a consensus with key partners as to what they are.”Footnote ¹¹⁹ This also entails reaching agreement around clearly defined roles for “key partners”: “It is important to know and appreciate the roles and responsibilities of the main actors involved at various stages of commodity distribution,” the 1997 UNHCR Guide stresses, as well as to monitor and report the equity of distributions made.Footnote ¹²⁰ In the 1994 UNHCR Guide, those with recognized stakes in the “data deals” for which it provides include: “donors”; “the public”; “the host government”; the UNHCR; the World Food Program; and nongovernmental organizations, “in their capacity as operational partners.”Footnote ¹²¹

Comparing this to the NPT, the latter does provide for safeguards agreements to be “negotiated and concluded” between the IAEA and non-nuclear-weapon state parties.Footnote ¹²² Once acting under the rubric of such an agreement, however, the IAEA is invested with “jurisdiction” and capacity for “control” so long as that is exercised “for the exclusive purpose of verifying” the nondiversion of “source or special fissionable material” from “peaceful nuclear activities within the territory of the State.”Footnote ¹²³ The NPT thus contemplates a far more durable, infrastructural, and concentrated distribution of sensory capacity than that envisaged by the 1994 and 1997 UNHCR Guides.

A fifth feature of the sensory economy fostered by the 1994 and 1997 UNHCR Guides is its emphasis on prior, accumulated knowledge, demanding that those charged with distributive evaluation “know the refugees” with whom they are called upon to deal, over and above any sensory data that they might draw directly from them.Footnote ¹²⁴ In this version of a sensory economy, those wielding greatest authority according to law or policy must exercise not just their sensory faculties, but also their good sense or common sense. “Population estimation requires experience, prudence and good judgment,” counsels the 1994 UNHCR Guide.Footnote ¹²⁵ “Experience shows,” the 1997 UNHCR Guide instructs, “that it is essential to find out specific information concerning every distinct refugee group in order to provide appropriate services efficiently.”Footnote ¹²⁶ For this reason, decision makers “must have specific information on the refugees, including on the population composition, gender and age, division of labour, access to and control of resources.”Footnote ¹²⁷ Likewise, they must “know the policy of the host government regarding refugees,” including “land use, type of settlement, local economy, [and] job opportunities.”Footnote ¹²⁸

In this characterization, sensory processing must be iterative, synthetic, and multisource, if it is to be authoritative. “It should be a rule of thumb in registration,” directs the 1994 UNHCR Guide, “that it is preferable to be receptive to a variety of indicators, allowing them to confirm or contradict each other, rather than taking a single indicator, such as a one-time ‘snap shot’ of physical presence, in isolation.”Footnote ¹²⁹ “This approach,” the 1994 UNHCR Guide continues, “encourages openness on the part of respondents, and responsiveness on the part of the registration system itself … .”Footnote ¹³⁰ Similarly, decisions regarding the “[t]argeting of assistance” must “reconcile a mix of objectives,” according to the 1997 UNHCR Guide.Footnote ¹³¹

Sampling makes up a part of this mix, recalling the economical metonymy highlighted above with reference to the NPT.Footnote ¹³² However, the sample is not envisaged as quite as rich and revelatory as in the nuclear nonproliferation context; rather, sampling is “a technique” among many, the results of which should be “fed back into the analysis … [to] help in elaborating a more detailed set of localized estimates.”Footnote ¹³³ In other words, the small (or local) is expected to be revelatory, for the most part, of the small (or local).

Further, both the 1994 and 1997 UNHCR Guides contemplate the use of computers in sensory processing.Footnote ¹³⁴ They do so, however, in limited and mostly instrumental terms, much as one might deploy a reliable underling to undertake a straightforward task. They envisage, for instance, the computerized generation of lists and cards and the processing of “very simple forms of documentation.”Footnote ¹³⁵ As the 1994 UNHCR Guide puts it, quaintly: “[i]t is useful to ask the computer to provide you with a list of all records where there are considerable similarities [among registrants] ([e.g.,] 75 per cent of information is the same).”Footnote ¹³⁶ In this sense, the sensory economy of the 1994 and 1997 UNHCR Guides is mostly manual.

In these respects, both the 1994 and 1997 UNHCR Guides cast sensory determination of worldly phenomena as a complex, creative, and intensely human process. According to this conception, the sensory “task [of the UNHCR] becomes that of the pragmatic manager of conflicting interests. Everything is dependent on … professional ability and good sense … [and] ability to manage the world order by equitable compromises. …”Footnote ¹³⁷ As in the NPT and related practice, the sensory economy of international law and policy on refugees depends on the prevalence of faith in a class of officials, armed with accumulated “best practice” and “good sense,” who are charged with sensing the world, in all its multiplicity, on behalf of others. In the 1994 and 1997 UNHCR Guides, however, these officials are to remain in continuous negotiation with other sensory sources, including refugee bodies; no such input is contemplated by the NPT except through mechanisms of state reporting and intelligence cooperation.

Sixth, and finally, the 1994 and 1997 UNHCR Guides invite a particular relationship to uncertainty in the detection of worldly phenomena—that is, an approach attentive to, yet undeterred by, the ubiquity of falsehood. “In most situations, but particularly in emergencies, the exact number of persons of concern will evolve day by day” and “[t]here is no system capable of establishing daily numbers with complete accuracy,” observes the 1994 UNHCR Guide.Footnote ¹³⁸ “Incorrect registration and cheating occurs in all situations and is impossible to stop completely,” that Guide continues.Footnote ¹³⁹ In similar terms, the 1997 UNHCR Guide emphasizes that “[r]efugee situations evolve” and that “there is usually great uncertainty about numbers.”Footnote ¹⁴⁰ Moreover, the “issue of population numbers” is recognized as “one linked to power and perspective.”Footnote ¹⁴¹ Decision makers “may well have to organise distribution in … situation[s] of uncertainty,” the 1997 UNHCR Guide counsels.Footnote ¹⁴² Uncertainty is, in this account, something with which field workers must continually wrestle, yet by which they need not be paralyzed. Rather, UNHCR fieldworkers are encouraged to “live with uncertainty.”Footnote ¹⁴³ Grappling with uncertainty, and with associated differences of perception, is an indispensable part of maintaining a viable sensory economy on the global plane, according to the 1994 and 1997 UNHCR Guides.

From the NPT and related instruments, and the 1994 and 1997 UNHCR Guides, we can discern how international legal authority is sustained (in both “soft” and “hard” forms) through the continuous, ritualized circulation and translation of sensory data. Far from a common scheme aspiring to uniformity and universality, such as international law is often presumed to install, the sensory economy just described appears more a collection of traveling fairs, following divergent routes and relying on differing resources. In the foregoing account, clusters of some-time official sensors traffic the distinct sensory capacities with which they have been entrusted by international law and policy, struggling to sustain, through their combined endeavors and accumulated authority, the durable prospect of a sensory commons. This is no “catallaxy,” or “spontaneous order of the market,” animated by free choice and marked by competition; this sensory economy both demands and consolidates investments, collaborations, entrenchments, hierarchies, and accumulations of many kinds.Footnote ¹⁴⁴ International legal infrastructure both sustains and is sustained by the uneven circulation of data and the variable distribution of sensory capacity.

Sensing bodies do not predate the operation of this juridical sensorium, existing in some preverbal, prelegal condition of authenticity.Footnote ¹⁴⁵ Rather, sensing bodies are vital to the maintenance of legal authority on the global plane—hence the preoccupation of international legal instruments and institutions with distributing them and vesting them with distinct competencies and powers of attestation. International legal authority thus depends not so much on the disciplining of sensing bodies through forces of reason, abstraction, or temperance as on practices of enabling, enrolling, and mobilizing sensing bodies, setting them on winding and interlocking paths of inquiry, dissemination, and distribution.Footnote ¹⁴⁶ It is in this way that, for all the divergences just described, international legal work strives to make possible experiences of collectively inhabiting a common sense-world or global sensorium and by this means, however contentiously, sustains the authority of international law and policy to act upon that world.

Remote Sensing Weapons

Despite the elaborate normative and sensory infrastructure described in Part III, perhaps nowhere are the limits of human perception on the global plane more acutely felt than in the field of nuclear nonproliferation. This awareness may be all the more painful because of its coexistence with a sense of humans’ sublime destructive potential. Humans have demonstrated seemingly infinite capacity to devise ways to kill one another and destroy nonhuman forms of life—continuing innovation surrounding nuclear weaponry and related technologies being one expression of that. Yet human faculties have proven consistently wanting in detecting nuclear proliferation efforts secreted by those responsible for them. The IAEA's gross underestimation of the state of Iraq's uranium enrichment program that became apparent after the end of the First Gulf War in 1991, noted above, despite years of apparent compliance by Iraq with its NPT obligations, was the subject of public derision and IAEA self-criticism.Footnote ¹⁴⁸ As television viewers around the world watched a standoff between Iraqi police and soldiers and some forty-four IAEA inspectors in a Baghdad car park in September 1991, IAEA inspection appeared more a matter of smash-and-grab spectacle (featuring a miscast group of global extras) than the overseeing of a decorous, reliable, and measurable flow of data.Footnote ¹⁴⁹ To the extent that clandestine uranium enrichment efforts have become known, this has often seemed to depend more on chance than treaty compliance. Insight into the North Korean nuclear program, for example, emerged from the “weird access” afforded Stanford Professor Siegfried Hecker and academic colleagues during visits to that country, most notably in 2010.Footnote ¹⁵⁰

These sensory failings, combined with other factors, have informed a growing conviction that the future of nuclear nonproliferation lies, at least in part, with automation and computation. The range and complexity of technologies deployed, or projected to be deployed, in the course of monitoring and verification under the NPT (and otherwise by the IAEA and comparable bodies) are too great to survey here. A January 2014 workshop convened by the IAEA entitled “Scanning the Horizon: Novel Techniques and Methods for Safeguards” discussed, for example, statistical methods, robotics, and machine-aided techniques for analyzing material of unknown composition, among them: active neutron interrogation (direct measurement of the fissile content of irradiated fuel using a large neutron source to induce fission); and x-ray fluorescence (measurement of the frequency, wavelength, energy, and intensity of electromagnetic radiation—that is, x-rays—emitted by a material that has been bombarded with x-rays or gamma rays).Footnote ¹⁵¹ The range of technologies discussed at an October 2014 IAEA Symposium on International Safeguards included: satellite imaging and image processing capabilities; developments in spectrometry and spectroscopy; surveillance technologies, remote monitoring instruments, electronic seals, and associated data systems; techniques and software for quantification of detection probabilities and error rates (including open source data analysis); training applications of virtual reality software; modeling and simulation technologies; communication systems; materials accountancy software; sampling and containment technologies; information management, data mining, and pattern recognition software; and more.Footnote ¹⁵²

Technologies for detecting, analyzing, and cross-correlating seismo-acoustic signals, infrasound, and radioactive particles are also projected to play a significant role in nuclear nonproliferation and related endeavors. The Comprehensive Test-Ban-Treaty Organization (CTBTO) is said to operate “170 seismic stations worldwide, 11 under hydroacoustic centres detecting sound waves in the oceans, 60 listening stations for atmospheric infrasound (low-frequency acoustic waves that can travel long distances) and 96 labs and radionuclide-sampling facilities.”Footnote ¹⁵³ This reportedly represents “the optimal number [of autonomous sensors] for global coverage” so that it is “now impossible to test even a small nuclear weapon in secret anywhere on earth.”Footnote ¹⁵⁴ Even so, the CTBTO is committed to installing an even more comprehensive network of 337 seismic, hydroacoustic, infrasound, and radionuclide detecting stations.Footnote ¹⁵⁵

In the nuclear nonproliferation context, these technologies seem, to many, to hold out the prospect of the world's unconcealment: that is, a renewal of the hope that we (at least a “we” that has access to the relevant data or representations of those data) might more fully inhabit a common sensory domain on the global plane; that all might come to know and grapple with more or less the same basic facts about the world, at more or less the same time.Footnote ¹⁵⁶ “‘What we're really aiming for is continuous high sensitivity of the entire network’, so no corner of the globe ever goes unmonitored” one commentator recounts, quoting W. Randy Bell, Director of the CTBTO’s International Data Center.Footnote ¹⁵⁷ Yet the prospect of this unconcealment is to come at a price; no longer are the publics to whom this work may be of concern likely to appreciate—that is, maintain some rudimentary, “common sense” grasp of—the sensory work ongoing in this context. W. Randy Bell does not explain exactly how, or by whom, this monitoring will be conducted, nor express any compulsion to so explain.

This prospect is not without its benefits. Technologies in use or under review for nuclear safeguards work offer the prospect of relief from the psychic, political, and economic costs of struggling to verify the unknown, amid more widespread use of nuclear energy, associated demand for monitoring, and a record of recurrent failure. As an indication of the strain upon the IAEA, one specialist publication observed that the “entire analysis section” within the IAEA's Department of Safeguards (“encompassing open-source analysts, satellite imagery analysts, and trade analysts”) numbered approximately sixty people as of 2015, and that the verification of Iranian nuclear activities alone would require dedication of approximately one third of that section.Footnote ¹⁵⁸ At the October 2014 IAEA Symposium on International Safeguards, the IAEA Director General, Yukiya Amano, lamented that

To this end, Amano remarked in an address at the Brookings Institution later that same month, the IAEA “make[s] increasing use of modern technology such as remote monitoring and satellite imagery.”Footnote ¹⁶⁰

Far from attempting to track the redistribution of sensory competences that may be associated with all these technologies, this section will focus only on increasing reliance placed (both in actuality and as a matter of aspiration) on satellite image data and their automated analysis in NPT verification and monitoring work. This section explores the implications of this for the sensory economy surrounding the NPT.

Satellites and Citizen Sensors

Advanced satellite imaging technology has been an “area of development” in IAEA safeguards work since the mid-1990s.Footnote ¹⁶¹ Since that period, the IAEA has used analysis of satellite imagery to assist in its evaluation of site declarations and reports submitted by state parties, in its detection and assessment of undeclared nuclear facilities, and in preparing for (or sometimes prompting the initiation of) safeguards inspections or related visits.Footnote ¹⁶² Except to the extent that satellite image data are included in information provided by NPT states parties to the IAEA, the data in question are sourced from commercial satellites.Footnote ¹⁶³

Alongside satellite imagery, emphasis has more recently been placed on the potential for IAEA safeguards personnel to make use of geospatial images and information otherwise made available, including by analyzing open source data.Footnote ¹⁶⁴ As one group of experts working for the U.S. Department of Energy observed:

Before exploring how the mediation of sense through automated analysis of geospatial image data—specifically, satellite imagery—might change the distribution of capacity described in the previous section, it is important to note another shift in the approach to NPT verification and monitoring occurring alongside this change in technical practice, and related to it. As in previous instances, this shift in practice followed from an experience of “crisis” for the IAEA: namely, the discovery of previous undeclared nuclear material and activities in Iran in early 2003, and the revelation by Libya later that same year that it had been developing uranium enrichment capability for more than a decade (both states being parties to the NPT; neither state having a safeguards agreement in place at the time under the 1997 Model Additional Protocol known as INFCIRC/540).Footnote ¹⁶⁶

In the wake of this experience, the IAEA consolidated a more tailored, state-specific approach to verification and monitoring, building on practices of state-level implementation developed throughout the 1990s: an approach that became known as the “state-level concept.”Footnote ¹⁶⁷ Rather than evaluating the implementation of safeguards on a facility-by-facility basis, and being guided by states’ declarations in doing so, the IAEA resolved to take into account the characteristics of each state as a whole and to develop a customized approach to verifying compliance by that state, including whether there may be undeclared material or activities anywhere within its territory. This prompted debate about the ambit of IAEA authority and allegations of overreach, to which the IAEA responded with reassurances and affirmations of continuity.Footnote ¹⁶⁸ Recourse to analysis of satellite imagery is consistent with this effort by the IAEA to gain a more fulsome, yet finely grained overview of activities and materials located in any one state's territory (including those outside state control) and to do so as nonintrusively as possible (in light of tension surrounding its adoption of the state-level concept).

Let us now revisit the five characteristics of the sensory economy established by the NPT and related safeguards protocols—highlighted in Part III above—and consider the potential effect of recourse to the automated analysis of satellite image data upon each of these.

First, the use of satellite imagery represents a departure from the ocular, telescopic focus of the NPT and related instruments. Despite many commentators’ analyses of satellite images’ use in terms of photographic representation, computerized analysis of satellite image data is not an endeavor comparable to NPT inspectors’ visual examination of nuclear facilities and reports. The fact that satellite and geospatial imaging may yield photographs for human scrutiny—in some instances, very powerful ones—is but one dimension of the analytical process in question, and arguably quite a peripheral one.Footnote ¹⁶⁹ Satellite image data of the kind analyzed for the implementation of nuclear safeguards are not amenable to being encompassed by a single gaze or human imagination in a way comparable to, say, the satellite photographs of earth hanging in space first circulated in the 1960s.Footnote ¹⁷⁰

Satellite image data do not “appropriate the thing photographed” in the mode of the photograph, creating “miniatures of reality that anyone can make or acquire.”Footnote ¹⁷¹ Famously, for example, IAEA inspectors did not know what they were looking at when first scrutinizing satellite imagery indicating the movement of “large disc-shaped objects” away from the Tuwaitha facility in Iraq. Only later, in the aftermath of the First Gulf War, were these understood to be calutrons: a mass spectrometer developed during World War II to separate isotopes of uranium and, by the late twentieth century, a dated (and hence disregarded) type of technology used for uranium enrichment.Footnote ¹⁷² Rather, for purposes of safeguards analysis, satellite image data comprise large quantities of unstructured data: they have no predefined organization and often combine many different data forms. Given their condition as such, “automated pre-processing, information extraction, and image information mining” are considered “essential” in order “to detect, analyze and manage significant features of interest” within those data.Footnote ¹⁷³

Computerization has not, of course, dispensed with human involvement in the technical practice of satellite image analysis; this practice entails a plethora of human-nonhuman interactions.Footnote ¹⁷⁴ Nonetheless, it is computer algorithms that are primarily responsible for extracting “interesting patterns and knowledge” from such data.Footnote ¹⁷⁵ Algorithms are “well-defined computational procedure[s] that take[] some value, or set of values, as input and produce[] some value, or set of values, as output.”Footnote ¹⁷⁶ Machine learning algorithms are those with capacity to modify their processing operations autonomously on the basis of newly acquired information.Footnote ¹⁷⁷ Employing such algorithms, information extraction from satellite imagery typically entails computerized analysis of a “whole range of [data, including] … spatial, spectral (reflective, emissive), polarisation, temporal and semantic properties of image pixels or image objects.”Footnote ¹⁷⁸ This commonly concerns data that are decentralized or “distributed”—that is, gleaned from a number of different, uncoordinated sites and sources.Footnote ¹⁷⁹ These must then be linked and combined for purposes of analysis—a process that itself demands “a set of coupled algorithms, along with their respective input parameter files” capable of generating “computational ontolog[ies]” or models of locations and/or events.Footnote ¹⁸⁰

The scope of what might be potentially “interesting” in data so processed—commonly defined in computer science literature as that which is “non-trivial, implicit, previously unknown and potentially useful”—need not be determined a priori; what is known as unsupervised or semisupervised data mining may itself generate a sense of what merits interest, as described below.Footnote ¹⁸¹ This may involve, among other processes, the automated “aggregation of similar neighboring pixels into homogenous objects, also referred to as segmentation … based on different criteria … .”Footnote ¹⁸² In other words, even the determination of what constitutes an “object” for purposes of analysis and change detection is automated to a significant degree.

The prospect of inspectors “observ[ing]” phenomena within this data, as INFCIRC/153 contemplates, seems at considerable remove from these practices of analysis.Footnote ¹⁸³ This may be especially the case when data mining of satellite imagery is unsupervised, or semi-supervised, as alluded to above.Footnote ¹⁸⁴ Supervised data mining proceeds from a training set of data known to have certain features: a record of past successes and failures, or preidentified instances of the type of norm-deviating event of interest to the human (or nonhuman) supervisor(s). The goal is for data mining software to learn the signature, or generate a number of possible signatures, of points of interest in the training data and classify unlabeled data by employing those signature(s). Unsupervised data mining, on the other hand, commences without an initial model, hypothesis, or norm from which deviation must be sought. The aim is to: generate and explore regularities and anomalies; infer the properties of some function capable of predicting phenomena in the data; create a model on that basis; and continuously refine those inferences and the ensuing model.Footnote ¹⁸⁵ Supervised mining offers a clear measure of success and failure (or degree of error) and a basis for redressing the latter; learning takes place through the detection and correction of errors. Unsupervised mining offers no ready way of evaluating the validity or usefulness of inferences generated; part of the process is continually revisiting and discarding hypotheses which the data mining practice itself will have generated.Footnote ¹⁸⁶

Information gleaned from, or produced through, satellite image data analysis does, nevertheless, come from somewhere. There are particular groups of people and institutions that participate in this work, based at international organizations (such as the IAEA), national scientific and research institutions, universities, and corporations, in settings public, private, or hybrids of both.Footnote ¹⁸⁷ Yet, with reference to the second feature of the sensory economy surrounding the NPT highlighted above, both the composition and role of a class of officials (charged with sensing for others) have changed as the expertise demanded of them has become both more intensely specialized and more disbursed. The authority of international law and policy continues to depend on the charismatic eliciting, intuitive interpretation, and measured circulation of sensory data by a class of professionals. However, that class can no longer be encapsulated by the figure of an IAEA director general or, for that matter, an IAEA inspector. Perhaps that figure might be better represented by a lead researcher in the safeguards group at the Forschungszentrum Jülich in Germany (a not-for-profit interdisciplinary research center—one of the largest in Europe—founded in 1956 by the German state of North Rhine-Westphalia and funded through a combination of state and federal government subsidies and income from consultancy or collaboration with industry): a figure unlikely to be publicly known.Footnote ¹⁸⁸

The redistribution of sensory capacity associated with reliance on satellite imagery for nuclear safeguards work is not just a matter of change in the types of people and institutions involved. It also entails transformation in the human-nonhuman alliances and sensory faculties in question. In particular, it involves a shift in emphasis away from the manual sense economy—the focus on handling things, documents, and instruments—emphasized above (the third of the features of the sensory economy surrounding the NPT remarked on earlier). IAEA inspectors continue, of course, to handle documents, reports, and instruments, and to be scrutinized publicly in their so doing. Questions of the authenticity of particular documents, and the sharing of documents, or failure to do so, remain at issue in public discourse surrounding the IAEA and the law and policy of nuclear nonproliferation.Footnote ¹⁸⁹ Yet the credibility of the IAEA has come to rest far more on the public sense of their proximity to, and affinity with, the “latest kit”—that is, with “technologies to unearth work on clandestine nuclear weapons … more diverse and more powerful” than ever before—than on their mastery of documentation or manual instrumentation at particular sites.Footnote ¹⁹⁰ In this way, the manual economy of nuclear nonproliferation earlier described seems to fold into or under a contemporary economy of “innovation” and “disruption” in which credibility, status, value, and success are evaluated and expressed in quite distinct ways (the broader dynamics of which lie beyond the reach of this article).Footnote ¹⁹¹

A further shift discernible in the reliance on satellite imagery in nuclear safeguards work is a tendency to emphasize the revelatory potential of large-scale datasets, rather than the fecundity of the small sample, the latter being a fourth feature of the sensory economy earlier described as surrounding the NPT. Inspectors continue to take samples or “‘swipes’ from surfaces of equipment and buildings” and to make use of new technologies for their on-site and off-site analysis.Footnote ¹⁹² Nevertheless, it is high-volume, large-scale data, and data drawn from a large number of sites and agents, that seem to be invested with greatest hope in contemporary discourse and practice surrounding nuclear safeguards. This was apparent in many of the papers given at the October 2014 IAEA Symposium on International Safeguards, where presenter after presenter emphasized the potential of “crowd-sourc[ed]” data to “expand the existing nuclear non-proliferation knowledge base,” “enlarge the view of the analyst on a universe of data,” and render the IAEA's accounts of phenomena of concern far more credible than preexisting alternatives.Footnote ¹⁹³ This reorientation toward the large-scale raises, in turn, the specter of a technology-driven loss of control—long a theme of politico-legal thought.Footnote ¹⁹⁴

Fifth, and finally, the proprioceptive dimensions of the sensory economy surrounding nuclear nonproliferation remain significant, but have taken on new forms. The NPT, and related instruments and practices, continue to direct movement and organize peoples and things in space in an effort to “close[] off the easier roads to proliferation” and direct proliferators toward “more difficult and time-consuming paths.”Footnote ¹⁹⁵ Placement in the NPT's binary categories of nuclear-weapon states and non-nuclear-weapon states remains important for determining the extent of treaty obligations and IAEA authority. This categorization may, however, hold somewhat declining significance in light of the emergence of the state-level concept and the related notion that IAEA authority should “expand to verification of nuclear material cycles [in] states possessing nuclear weapons.”Footnote ¹⁹⁶

Overlaying such ongoing geopolitical distributions in space and time, however, is the growing importance of the placement of phenomena (or of computational simulations of the same) in datasets derived from satellite imagery and other geospatial data drawn from a mostly undifferentiated pool of “citizen sensors.”Footnote ¹⁹⁷ This is a form of placement that cannot be witnessed or experienced in conventional terms (much as citizenship of a state—and relationship to most of one's co-citizens—cannot be experienced except in mediated representations of the same). Indeed, insofar as it arises in the course of unsupervised or semisupervised data mining, this proprioceptive proximity may be immune to human perception altogether unless and until it is deemed actionable and invoked as the basis for legal or policy action (such as an IAEA complementary access inspection, under an Additional Protocol corresponding to INFCIRC/153, to assure the absence of undeclared nuclear material and activities).Footnote ¹⁹⁸ Even then, the relationship between a certain action and a particular assemblage of data will typically not be made apparent in any public arena.

To the extent that the sensory economy surrounding the law and policy of nuclear nonproliferation is becoming more reliant on this form of virtual or simulated distribution in space and time—one registered and expressed only algorithmically—this may be indicative, on the global plane, of the further deauthorization of a politico-legal order premised on “attestive-analytical vision” capable of being experienced in common, along the lines recounted by Yaron Ezrahi.Footnote ¹⁹⁹ This article has earlier shown that, at least on the global plane, politico-legal authority's justification has always involved mobilization of a greater array of sensory faculties than just the visual.Footnote ²⁰⁰ Nonetheless, this article tells somewhat of a similar story to Ezrahi's insofar as it suggests that the redirection of sensory data, and its production and processing in new ways, may signal the decline of structures and rituals of global public sense, with associated challenges for international law and policy. It remains to be seen whether the promise of “crowd-sourcing,” and the translation and circulation of crowd-sourced sensory data, will prove capable of backfilling the foundations of a global sensorium that international law and policy have struggled so long and hard to maintain.

Refugees and the Redistribution of Sense

In contrast to the stories of crisis and institutional self-flagellation recounted above (with reference to the IAEA's limited, flawed, or misdirected sensory capacity), the UNHCR had by the mid-1990s developed quite an extensive, agonistic vocabulary and set of “best practices” surrounding its recording of the incidence and distribution of refugees. This will have become apparent from the account of the 1994 and 1997 UNHCR Guides presented above. Unlike the IAEA and the law and policy of nuclear nonproliferation, the authority of the UNHCR, and associated law and policy, has never seemed to hinge on the institutional ability to make preemptive revelations, nor even to ensure the global circulation of accurate data, so long as the UNHCR remains responsive to manifest human need and may be seen to be so. Accordingly, the “framework of attestive [sensory] culture” in which a “skeptical yet discerning” global audience may “presume to witness [or otherwise perceive] the actions of political agents” and international organizations might seem more likely to remain intact in the law and policy on refugees, notwithstanding the technical transformations just described. The sensory economy surrounding the work of the UNHCR since the mid-1990s is, however, also undergoing transformation, as the next section will indicate.

Biometric Sense

In the early years of this century, the UNHCR initiated a program (the UNHCR Program) for biometric registration and deduplication of Afghan refugees living in camps in Pakistan and applying for humanitarian assistance for repatriation.Footnote ²⁰¹ The UNHCR Program was initiated in the context of the mass-repatriation of Afghan nationals from refugee camps in Pakistan back to Afghanistan, after the fall of the Taliban in 2001. From 1988 onward, the UNHCR concluded a series of agreements with Pakistan, and later with Afghanistan too, providing for the voluntary repatriation of Afghan refugees located in camps in Pakistan with screening, supervision, and material assistance to be provided by the UNHCR; these agreements afforded the legal basis for the UNHCR Program.Footnote ²⁰²

Between 2001 and 2005, the UNHCR facilitated the return of over three million refugees to Afghanistan.Footnote ²⁰³ As part of this process, the UNHCR determined that every returnee should receive “transport assistance ranging from $5 to $30—depending on his [or her] final destination—a UNHCR family kit with plastic tarpaulin, soap, and hygiene items, as well as wheat flour from the World Food Programme.”Footnote ²⁰⁴ In distributing this humanitarian aid, the UNHCR used traditional identification methods to try to distinguish “genuine” first-time claimants from “recyclers” claiming multiple assistance packages, but found these methods wanting.Footnote ²⁰⁵ At the UNHCR’s request, commercial technology vendor BioID Technologies (BioID), in cooperation with Iridian Technologies, developed a biometric registration facility and mobile registration units for the organization's deployment of preexisting iris recognition technology, the operation of which was described as follows:

The mandatory participation of all returning Afghans seeking assistance from the UNHCR (from the age of six) in this biometric screening process thereafter came to be enshrined in Article 15(2) of the Agreement Between Pakistan, Afghanistan and the UNHCR Governing the Repatriation of Afghan Citizens Living in Pakistan 2007 (extended in 2010 and 2013) (Tripartite Agreement).Footnote ²⁰⁷ This agreement provides that: “Iris recognition processing will be mandatory for all returning Afghans wishing to obtain travel and reintegration assistance … .”Footnote ²⁰⁸ In contrast, the predecessor to this Tripartite Agreement, concluded in 2003, provided only for verification of identity documents.Footnote ²⁰⁹

As with satellite imagery analysis for nuclear safeguards work, data mining is central to the sensory mechanics of the UNHCR Program. The data assembled for this purpose does “not correspond to any sampling plan or experimental design.”Footnote ²¹⁰ Rather, data collection is conducted at multiple sites, under often challenging and highly variable conditions, at rates contingent on refugee flows. In the UNHCR Program, the particular data mining technique used by BioID to classify phase information (that is, a numeric expression—in the form of a “bit stream”—of a pattern extracted from a set of iris images) has not been described publicly by either BioID or the UNHCR .Footnote ²¹¹ Nonetheless, published descriptions of iris recognition techniques suggest that this may involve a type of data mining model known as a neural network, employing machine learning.Footnote ²¹² While neural networks vary widely, they are all predicated on the processing of numeric input through a series of interconnected nodes (some of which are hidden) and the attribution to connections among those nodes of associated weightings, with each layer of these nodes being comprised of the weighted sum of values in the preceding layer. In many instances, the weighting attributed to nodal connections is learned through the processing of, and verification of performance against, a training set of input data.Footnote ²¹³ Alternatively, it may be that this iris recognition is carried out using a decision tree: another type of predictive data mining model used for classification, again employing machine learning.Footnote ²¹⁴ Decision trees are “[t]ree-shaped structures” that represent sets of binary tests on the basis of which data is divided and classified at each “branch.” After training and validation of outputs, the tree can be used to “generate rules for the classification of a dataset” without supervision.Footnote ²¹⁵

After a year of this system's operation, and the processing of just over 200,000 refugees, the UNHCR reported that approximately 1,000 people trying to claim multiple assistance packages had been detected, “in addition to more than 70,000 families … rejected [during the same period] … under other screening methods … .”Footnote ²¹⁶ Those other screening methods—maintained alongside iris recognition—included “interviewing potential returnees and examining their family photos.”Footnote ²¹⁷ The iris recognition system was said to have performed “flawlessly,” despite the risk of data corruption posed by “the heat and dust of Pakistan's border territories with Afghanistan,” and without reference to error rates associated with factors such as: image compression; contact lens use; pupil dilation; and corneal bleaching, scarring, inflammation, and other pathologies.Footnote ²¹⁸ According to the UNHCR, concerns that use of the technology might intimidate, raise traditional objections to women being photographed, or compromise privacy also proved unfounded: “only the eye is seen onscreen”; “[t]ests on women and children are done by female refugee agency workers”; and “the code describing the iris has no link to the name, age, destination or anything else about the refugee.”Footnote ²¹⁹

Let us return to the six features of the sensory economy earlier described, surrounding the work of the UNHCR and related law and policy, to explore how these may have been affected by the introduction of the UNHCR Program.

First, recall that the 1994 and 1997 UNHCR Guides sought to enroll refugees in a reflexive sensory practice, seeking their long-term participation in the processes of registration and distribution for which those manuals provided. The shift in sensory practice discernible in the UNHCR Program likely makes it difficult, however, to sustain this commitment to addressee involvement. Emphasis is placed, in public explanations of the UNHCR Program, on its speed and definitiveness. According to BioID's account, “[t]he whole process … takes less than 20 seconds.”Footnote ²²⁰ Refugee participation is envisaged, accordingly, in terms of delivery and acceptance; the body is to be delivered up as a site of data collection and then quickly set aside, to make way for the next. In place of active participation, passive incuriosity seems to be the disposition encouraged by the UNHCR Program: “How can they argue now, the machine can't make a mistake,” a UNHCR staff person, Rifaat Tajik, stated with apparent relief.Footnote ²²¹

The tactile dimensions of the sensory economy of UNHCR work and policy comprised the second feature highlighted above. The UNHCR Program seems, however, to dispense with most, if not all, recourse to touch. Photographic representations of the process of registration featured in publications of the UN and the UNHCR show enrollees sitting before cameras, alone, with camera operators at some distance from them, while “only the eye is seen onscreen.”Footnote ²²² In place of the contacts for which the 1994 and 1997 UNHCR Guides provide—marking bodies with dye, adorning bodies with wristbands, passing tokens from palm to palm, estimating and counting in sometimes tightly packed conditions—the UNHCR Program insists on distance in the course of detecting and registering a refugee body. This likely helps to reinforce assurances of privacy and physical safety: the process “involves no risk to the eye” and “the digital code for each iris is stored without any personal information,” the UNHCR insists.Footnote ²²³ Capturing the iris image does require those seeking assistance to remain briefly in a small space together with “[a]id workers operating the machines,” but any contact in the course of this is fleeting and unremarkable according to UNHCR reports.Footnote ²²⁴ “Refugees … have been very co-operative and have adapted to the new verification technology without causing a fuss,” one UNHCR account states.Footnote ²²⁵

Third among the features of refugee law and policy's sensory economy mentioned above was a tendency to elicit, and apportion capacity and incapacity among, group configurations, “populations” especially. In the UNHCR Program, however, this orientation toward collective agency and representation breaks down. Those enrolled in the UNHCR Program continue to be identified as Afghans, and sometimes as Pashtun or Shinwari.Footnote ²²⁶ Nonetheless, the aim of the UNHCR Program seems to be to cut through the murkiness and inflexibility of group relations as repositories of sensory data and experience. Much is made, for example, by the UNHCR of the Program's capacity to liberate people from the burden of family obligation, as in the following report:

Indeed, automated detection of a body and its evaluation as eligible or ineligible for assistance (on the basis of matching IrisCode) need not engage a subject as such, let alone a subject's relations with other subjects, in order to yield a definitive, actionable predicate. The unit of analysis with which data mining is concerned may not be recognizable as a place, person, or thing that registers as familiar.Footnote ²²⁸ Data searches typically address a much lower or more fragmentary level of analysis, namely mathematical correlations between bits of data. In order for these mathematical correlations to be made useful and actionable for policy purposes—or potentially so—no occasion need arise for a legal subject as such to be “hailed.”Footnote ²²⁹ That is, there is no need, as part of such processes, to invite some groups, entities, or beings to experience themselves and others as embodied legal persons, with all that that implies (usually, possession of a more or less coherent will as well as a relatively stable identity and specifiable location in space and time). The IrisCode generated and stored under the rubric of the UNHCR Program does not call forth a subject as such, let alone a subject with any particular qualities, capacities, vulnerabilities, or propensities. It does not signify flawed or virtuous character or capture any particular circumstances that might explain why an applicant might seek multiple assistance packages. All it demands and makes actionable is a mathematical relationship (meeting certain specified thresholds) between data gleaned from one set of processed images of an iris and data gleaned from a second set. The precise way in which IrisCode matching intersects with “other screening methods” attentive to group dynamics (interviews and family photos’ scrutiny, for instance) is unclear, but it appears to draw the latter to a perfunctory close.Footnote ²³⁰ According to one UNHCR report: “if the test reveals that the refugee has been enrolled before—and only about half of one per cent are found to be ‘recyclers’—the person is refused assistance.”Footnote ²³¹

As well as eliciting and being oriented around collective identities (with which the UNHCR Program more or less dispenses), the sensory economy that the 1994 and 1997 UNHCR Guides sought to maintain was shown above to be transactional or organized around negotiated deals in data; this was the fourth of its features highlighted in Part III. This is not, however, a feature of the UNHCR Program, at least according to published reports of its operation (and recognizing that it still operates within broader parameters of UNHCR policy and practice, of which this may still be a feature). UNHCR publications reporting on the UNHCR Program certainly convey a sense that refugee applicants for assistance must be persuaded of the Program's virtues, if their continued participation is to be secured. Nevertheless, the embeddedness and inscrutability of the parameters on the basis of which applicants for humanitarian assistance may be disentitled make the prospect of factoring negotiation into this encounter difficult to realize. Recall the UNHCR staff person's remark, quoted above: “How can they argue now, the machine can't make a mistake.”Footnote ²³² That person continued as follows:

The emphasis in the UNHCR Program thus seems to rest less on negotiation and more on definitive—albeit mostly unexplained—outcomes. Even though the algorithms in question will likely have been trained using a set of correct and false matches between batches of IrisCode (verified as such by nonautomated means), the precise algorithmic derivations of that training set used to identify a match will have been generated through machine learning. As such, these are the outcomes of automated inquiry rather than its predetermined starting points. They will have emerged, and will continue to emerge, in ways that may not always be explicable, even to those well versed in relevant data-mining techniques. Research and practice in data mining tends to make far more of search efficiencies realized than explaining, and opening to general scrutiny, the precise features and components of different data-mining algorithms:

For these and other reasons, it is often “hard to explain how the system reached a decision” with respect to any particular association or match detected.Footnote ²³⁵ Accordingly, scenarios in which these sensory practices—questions of their fairness in particular—are made negotiable are hard to envisage.

As a fifth noteworthy feature, the sensory economy of the 1994 and 1997 UNHCR Guides has involved trade in perceptions of, and faith in, “good sense,” much as the NPT and related instruments have done. Expectations have been sustained by international law and policy that the work of detecting changing worldly phenomena would or should be informed by accumulated human knowledge and experience, both empirical and intuitive. In the case of the UNHCR, this has elicited sustained, ongoing effort on the part of the institution and its workers to “know refugees” in their particularities, in ways not confined to what may be extracted from them by way of verifiable data.

In contrast, the “sense” of an applicant for humanitarian assistance likely to be generated by the UNHCR Program is oriented around a singular, brief encounter and thereafter memorialized only in digitized, anonymized form. At least as far as the UNHCR Program itself is concerned, the “knowledge” that this program yields arises from momentary co-occurrence between data so recorded, the broader significance of which will often be unclear. The finding of a match between X and Y in IrisCode does not import presumptions of: cause and effect; common purpose or normative commitment; proximity in time or space; identitarian similarity; commonality of predicament, experience, or history; or affective or communicative connection as between code sequence X and code sequence Y or the humanoid sources of those sequences. Their association is based only on analysis of a large data set having generated, for the time being, a greater probabilistic affinity (or lower dissimilarity measure) between code sequence X and code sequence Y, or particular subsets of the same, than between code sequence X and all other sequences that may be assembled from the database at the relevant time (and that affinity otherwise satisfying applicable thresholds).Footnote ²³⁶ As additions and eliminations are made to and from the database, the relevant thresholds may be modified.

In addition, many of the patterns or relationships produced in data mining will be presumed superfluous and subject to winnowing. In the course of mining data, there are often many meaningless associations discovered.Footnote ²³⁷ Algorithms designed to find relationships in data will tend to do so regardless of whether there are any significant relationships “naturally” present in the data.Footnote ²³⁸ This subordinates the process of gaining “good sense” to an expectation of continual, ubiquitous tinkering and shedding of data.Footnote ²³⁹ What may be deemed useful and usable, from a mass of relationships represented, will depend to a significant degree upon the happenstance of design features’ confluence and interaction with data.Footnote ²⁴⁰

The sorts of fleeting associations foregrounded in data mining, and the relationships of proto-people and bit streams that they draw together, are not configurations to which international law and policy have been well attuned to date. Certainly, international lawyers have been concerned with practices of counting, enumeration, and measurement in international law, and justifiably so.Footnote ²⁴¹ Nonetheless, international law and policy have mainly traded in global associations presumed much thicker, more meaningful, and more durable than those instantiated in data in the ways described above. Patterns appearing momentarily in data have not typically been a basis for action or inference in international legal affairs, but they are becoming more so. It is for this reason that the redistribution of the sensible described above signals a potential undermining of international legal authority. International legal order's perceived correlations with good sense and good judgment have depended, in large part, on its organization around, and articulation through, relationships that many people find meaningful and comprehensible. To the extent that its operation becomes premised on relationships less accessible to experience, its capacity to inform global common sense (for good and for ill) may come under threat.

Sixth, and finally, the 1994 and 1997 UNHCR Guides discussed above advanced a particular relationship to uncertainty in the work of detecting worldly phenomena—as a condition to be “lived with” collaboratively and agonistically. The UNHCR Program, however, fosters a very different disposition toward uncertainty. Uncertainty endemic to iris recognition (reflected in false rejection rates, for example) cannot be eliminated and tends to be exacerbated by a range of factors mentioned above (including image compression and corneal bleaching).Footnote ²⁴² Nonetheless, once quantified and cabined within biometric registration infrastructure and practice, this uncertainty comes to appear relatively untroubling. Recall that, after operating for some time, the UNHCR Program technology was said to have “performed flawlessly.”Footnote ²⁴³ To the extent that uncertainty (expressed as a false match rate or equivalents) is acknowledged, it seems most likely to be addressed according to a logic of technical “optimiz[ation] … for best performance” in the UNHCR Program, rather than through the thrashing out of differences of power and perspective contemplated by the 1994 and 1997 UNHCR Guides.Footnote ²⁴⁴

The introduction of the UNHCR Program thus suggests that significant shifts in the sensory economy surrounding the policy and practice of refugee registration may be underway. Yet its effects are not, by any means, wholly negative. Perhaps, as UNHCR publications suggest, the Program will help ensure that scarce resources are distributed more equitably, with greater efficiency, so that “assistance is reaching those who really need the help of UNHCR to repatriate.”Footnote ²⁴⁵ The concern of this article is not to evaluate the UNHCR Program. Rather, what is of interest here is how the UNHCR Program seeks to generate and sustain a sense that all (or some population that serves as a proxy for “all” in the UNHCR’s daily practice) are confronted by and working with the same basic facts about the world—how it seeks to sustain, in other words, a global sensorium. The techniques that the UNHCR Program employs for so doing differ quite markedly from those that the 1994 and 1997 UNHCR Guides sought to foster. In place of a population, across which vulnerability, capacity, and need were to be distributed iteratively and through constant negotiation (according to the aforementioned Guides), the UNHCR Program produces and acts upon a computational ontology of interlocking datasets, the human-nonhuman navigation of which follows parameters and patterns inscrutable to most.

To the extent that the UNHCR Program and institutional accounts of this initiative manage to sustain an experience of shared, credible, common sense at work, this seems to have been achieved (to date) mostly through the dazzle of technology: an expectation that “the machine can't make a mistake.” However, this is an expectation unlikely to withstand scrutiny, given popular and scholarly critiques of technological determinism long in circulation.Footnote ²⁴⁶ The “machine” clearly can make a mistake, as UNHCR staff would probably acknowledge if pressed on the issue. What needs to be better understood in this context, then, is how those participating in the UNHCR Program have come to disengage so from the complexities and instabilities of their distributive predicament, in contrast to the more agonistic approach that they were encouraged to take in the era of decision-making described in Part III. The description of sensory practice set out in this Part IV has offered some clues as to how this disposition has been cultivated amid mundane practices in the field.