Today, Big Data analyses and data-driven profiling are applied in many different areas. In general, "(Big) data captured through digitalized devices are processed by algorithms aimed at predicting what a person will do, think and like on the basis of their current (or past) behaviors."21 Such algorithms are for example used for predicting security threats and risks, forecasting crimes in predictive policing, in insurance and employment screening, or in establishing credit scores for predicting the likelihood of loan default. In managerial contexts, data and numbers are seen as an essential basis for organizational steering and transparency. Big Data analyses are used in the areas of marketing and consumer research. Increasingly, psychometrical categories are used to create clusters of consumer types (profiles) and to predict consumer behavior.22 The idea of "predicting people" has also become central in Human Capital Management.23

In all of these areas, Big Data analyses promise to deliver a basis for "evidence-based decisions." Data are frequently presented as neutral forms of information that represent objects in the world and provide insight into social, economic and environmental phenomena. The assumption is that (big) data comprehensively map reality, while algorithmic procedures (computer-based data mining) allow identifying and "discovering" the underlying structures and patterns of this reality. McAfee and Brynjolfsson formulated the conviction of Big Data proponents succinctly: "Data-driven decisions are better decisions—it's as simple as that. Using big data enables managers to decide on the basis of evidence rather than intuition."24 Critical researchers in various fields instead have questioned this assumption on the grounds that (big) data and algorithms might be inherently biased, creating, reproducing or even intensifying unfair discriminations and established inequalities.25 In addition, Kitchin reminds us that (big) data are neither simply given nor representing the world. Instead, "data are captured from the world, but in turn do work in the world."26

For Derrida, a decision that deserves the name presupposes "undecidability" and requires going through the "ordeal of the undecidable." Algorithmic decisions are in his sense not decisions at all since "a decision that didn't go through the ordeal of the undecidable […] would only be a programmable application or unfolding of a calculable process."27 So called decision trees that characterise algorithmic decisions effectively automate responses and force a yes/no or if/then choice. In this sense, the agonism and radical uncertainty of a true decision is annulled or rather denied. There is no hesitation or fear, no struggling heart, no ambiguity. Similarly, an "apparent decision taken on the basis of what is 'seen' evidently, via the calculation of experts, or in the screened results of algorithmic visualization, is not a decision at all."28 In fact, algorithmic procedures that produce a decision (or rather a yes/no switch) as the outcome of a calculated series of steps (e.g. decision trees) reframe ethical and political questions as technical problems of systems engineering. Operating through logical and mathematical procedures, they become "supercarriers of formal rationality"29 that promise to overcome the "bounded rationality" of humans. While proponents of algorithmic decision-making assume that the inherent undecidability, which traverses human systems, can be—or ideally is—removed, Derrida reminds us that the undecidability is "not a moment to be traversed and overcome." It continues to inhabit the decision "as an essential ghost."30 The programmed decision and the objects, divisions and demarcations it creates are both haunting the world and haunted by its own shadow. In the context of data-driven decisions, the ghost(ly) and the notion of haunting draw attention to all that exceeds measurement and meaning, to the "noise" in the system, to all that cannot be captured and computed, to that which occasionally and unexpectedly interrupts systems and sometimes also allows for novelty, creativity and the creation of the new.31 As Cheyney-Lippold has demonstrated, even "algorithmic identities"32 inferred upon individuals can never finally remove the undecidability and fix the subjects they refer to. On the contrary, they are pervaded by undecidability. For example, the ratio male/female or other characterizations is constantly changing according to the feed-back data; definitions of maleness/femaleness etc. can shift according to the logic of the algorithm.

Algorithms can be understood as "series of generalized procedures for turning disorganized data inputs into manageable outputs through a series of logical rules."33 As such, algorithms are organizers that are both socially produced and socially productive. They organize and structure the reality producing process as they select, filter, and frame information and create facticity. Hence, no algorithm is neutral. "Pattern discrimination,"34 the ability to filter information from data is never innocent. It is inherently tied to value judgments and it often reproduces or amplifies the very phenomenon it seeks to describe or represent. Values are thus not external to the algorithm, but they are folded into it. Even the (seemingly) trivial acts of naming, of storing data, of "deciding where to make cuts in the system,"35 contain value judgements. Moreover, algorithms that are "inherently framed and shaped by all kinds of decisions, politics and ideology" are created for purposes that "are often far from neutral: to create value and capital; to nudge behaviour and structure preferences in a certain way; and to identify, sort, classify people."36 Algorithms codify assumptions and are socially productive. A data-mining algorithm for example that is designed to find or "discover" valuable data and patterns, sorts, selects, filters data based on specific assumptions about what makes data "valuable" and for whom. In Hacking's terms, it is not just an "engine of discovery"—as the term "Knowledge Discovery in Data Base" (KDD) suggests—but also at the same time an "engine of making up people."37 Algorithms are thus performative. They generate objects of knowledge that loop back and shape the world. There are many different—highly complex and sophisticated—technical methods of data mining (e.g. decision trees, cluster analysis, neural nets, text mining, anomaly detection and many others). What is distinctive about these methods is the specific interested perspective that guides the search for "valuable" data and patterns.38 From these data it allows to construct a "person of interest," for different actors, institutions, organizations for different reasons. A hiring algorithm for example mines data from the interested perspective of an employer and creates an image of the attractive/unattractive candidate; a security algorithm makes up an image of the potential security risk; in consumer profiling, types of more or less profitable consumers are created and sorted; in credit scoring an image of the more or less creditworthy or trustworthy person is produced. While it is often assumed that the outcomes represent a preexisting and underlying reality in a neutral way, critical researchers have argued that the outcomes are rather "carefully crafted fictions"39 that help to make us overlook or forget that underlying data are often messy, incomplete, full of gaps and errors, as well as the fact that algorithms may work in biased or even systematically distorted ways. However, such fictions are not simply illusions, but rather social objects and artefacts that have concrete material effects in the world. This will be theorised and illustrated in the following section on profiling.

All profiles are abstractions. In the process of profiling images of the person are created for the purpose of diagnosis or prediction. A profile is never a representation of an individual but a selective image that is constructed for specific purposes. In the process of profiling the complexity of the person is reduced to a finite number of traits, indicators, etc. Profiles are in this sense "model(s) or figure(s) that organize […] multiple sources of information to scan for matching or exceptional cases."48 Bogard locates these "models or figures" between fiction and reality at the intersection of virtual and actual worlds. Such figures may be fictions (in the sense of not representing an underlying reality) but these are "a fiction difficult to separate from fact, whose effects are indiscernibly real."49 Fictions create mental images and expectations. They are operationally effective, as they influence decisions, shape and intervene in the world. Consequently, profiles should not be understood as descriptive, but as performative.

Harding has taken a similar perspective and argued that profiles that are assembled and derived from data should not be understood as representations of the self, but as "mere specters of our actual selves—something metaphorically akin to the theatrical illusion called Pepper's Ghost."50 Understanding profiles as "mere specters of our actual selves" reminds us that they should not be confused with the embodied human being, yet at the same time, they are more than simply illusions. I argue that such categories may not be able to capture the world, but nevertheless they may have material effects in the world. In her seminal book on "ghostly matters," Gordon51 has introduced the "ghost" as a social figure that helps understanding and theorizing phenomena that are somehow located between the dead and the living, between the abstract and the concrete; phenomena which are absent, but nevertheless make their presence felt, phenomena which are illusionary or unreal, but are at the same time real in the sense of producing material effects. The figure of the "ghost" and the "ghostly" has also been taken up by organization scholars52 and data-scientists53 for theorizing a "spectral working" of forces that destabilize and contaminate any self-sufficient present, for disrupting linear concepts of time and for attending to how policies or data become "haunted" when their implementation disrupts local cultures, traditions and histories. Continuing this work and taking some inspiration from Derrida's analysis of the ghost and the "ghost effect" in The Specters of Marx, I suggest that profiles can be understood as ghosts that are the outcome of a "spectrogenic process."54 Derrida describes this as a process in which thoughts, ideas, data, etc. are extracted from the living body and integrated into a more abstract body. "Ghost effects" are produced with every step of abstraction. Transferred to our context this means, that ghost effects are engendered on multiple levels. Firstly, when human experience is abstracted and turned into "information," secondly, when data are extracted, resulting in what Haggerty and Ericson called a "decorporealized body, a 'data double' of pure virtuality."55 The data-double as the aggregation of personal information represents already a high level of abstraction; an even higher level is reached when the data-double is "mined" and selective information is assembled and reassembled into a profile that allows institutions and organizations to make discriminations and separations among people. In the process of application, institutionally and technically formed and shaped data return to the world of real-life events, where they haunt those with whom profiles are associated. As Gordon put it: "[i]n haunting, organized forces and systemic structures that appear removed from us make their impact felt in everyday life in a way that confounds the social separations themselves."56

While all profiles are abstractions, profiling can be distinguished first, by the way how these abstractions are produced and second by the way how these abstractions (profiles) are applied in decision-making processes. I elaborate this in the following.

There are different forms of profiling. Traditional profiling is a process of classification, which is based on ex-ante categorizations of individuals. It assumes that the individual can be characterized by a set of traits that can be identified by professional/scientific methods. A specific type of knowledge is created that typically results from experience or the tradition of expert knowledge. Typically, such knowledge is deductive and derived from specific predefined hypotheses. In the traditional sense, the profile results from targeted questioning based on specific hypotheses. Profiling involves the definition of categories that are assumed to characterize the individual and the establishment of methods for measuring the individual along these categories. Measurement allows to identify differences and to determine the distance to the norm. Profiling allows to position individuals in a normative matrix of behavior and to make them comparable to others.

Traditional profiling resembles the classical disciplinary technology of examination. This "tiny operational schema that has become so widespread (from psychiatry to pedagogy, from the diagnosis of diseases to the hiring of labour)," which Foucault saw at the heart of the procedures of discipline, "establishes over individuals a visibility through which one differentiates them and judges them."57 It is the scientific or quasi-scientific fixing of differences and their representation in numbers, charts, and reports, which provides the basis for administrative decision-making.

As an administrative technology, described by Foucault as a "scientific or administrative inquisition which determines who one is,"58 classical profiling works as a productive technology of power. It creates objects of knowledge, allows classifying and ranking these objects, and turns individuals into "cases" that can be handled and managed in particular ways. Emerging profiles are inherently normative constructions that have intrinsic moral value since they are "kinds that people want to be or not."59 Like stereotyping, profiling works as an a priori ordering that tells us who "is" good/bad, trustworthy/untrustworthy, healthy/unhealthy, suspicious/un-suspicious, profitable/not-profitable, etc. Profiles also provide those who are profiled with an image of themselves. They allow subjects to reflect on themselves and to monitor their behavior in the light of normative expectations. In this sense, profiles create an imaginary ideal ("idéal speculative") that works as a subjectifying force.60

In "new profiling,"61 profiles are often the result of automated and algorithmically enhanced systems of pattern recognition and -discrimination.62 This presupposes the transformation of the flux, flow, and mess of real-life events into data that are readable by machines. Knowledge in this case is inductive. It is shaped by the algorithms, which direct attention, focus on specific points in the data set and cancel out other data. Profiles in this sense are less about measurement but about "detection." In contrast to traditional profiling, categorization emerges ex-post, from clustering detected patterns. As Hildebrandt explains, "Pattern recognition, based on 'blind' correlations (i.e., correlations that do not derive form predefined hypotheses and do not necessarily imply causes or reasons) allows those that use the ensuing profiles to anticipate the state or behavior of the objects or subjects that are profiled."63

Traditional profiling follows the logic of inspection. It seeks to identify the presence of certain predefined characteristics in the individual and categorize it accordingly. In contrast, new profiling follows the forward-looking logic of prospection. It is based on the assumption that detected patterns and correlations allow creating "predictive transparency,"64 which provides the basis for preemptive decision-making and anticipatory governing. Predictions are derived from patterns in past behavior or they are derived from similar patterns of "groups" or "neighbors." Categorizations thus not only depend on individual actions, behaviors and histories, but on those of others who are similar to him or her.65 The generated knowledge is based on probability rather than on the construction of causalities. In short, while traditional profiling seeks to determine, who or what the person is, new profiling "render(s) us transparent in a rather counterintuitive manner. We become transparent in the sense that the profiling software looks straight through us to 'what we are like,' instead of 'what or who we are.'"66

New profiling, based on pattern recognition, constructs or assembles a person of interest by connecting data-points and visualizes the person for example on a screen, a scoring sheet or a risk map. Algorithms function as means for directing attention and for focusing on specific data while neglecting or cancelling out all other data. In the case of new profiling, they make up people in the form of "data doubles" or "data derivatives."67 Such objects of knowledge can be generated from a variety of different and heterogeneous sources (e.g. discrete digital acts like searching, messaging, blogging, purchasing, liking, tweeting, and posting, etc.). Profiles are composed by combining these actions and connecting bits and pieces of data. They are not representations of the 'real' person or of 'individuals,' but selective constructions which consist themselves out of sub-individual elements, the fragments of registered behavior, which are extracted from the flow of data for specific purposes. They resemble the coded "dividual"68 which, according to Deleuze, replaces the undivided in-dividual in the "society of control." The dividual consists of separate data-points, which can be scrutinized, calculated, circulated, assembled, and reassembled for specific purposes. As Hildebrandt explains, "these elements or properties are not given; they are attributed by whoever writes the algorithms of data analytics; taking into account that whatever dividuals are sought after they must be inferred from machine-readable data by machine-readable algorithms."69

The aim is not to produce exhaustive knowledge about a particular individual and his or her intrinsic characteristics, but to create a basis for acting on similar individuals. These profiles are not a reflection of a given identity, but a projection of possible future behavior. Amoore stresses that the digital alter ego that emerges from algorithmic profiling "is precisely a projected person: an image of a potential future person as yet to come,"70 e.g., the potential terrorist, the potential criminal, the potential buyer, the potential toxic employee, the potential credit risk, etc.

While traditional profiling fixes the individual in a relatively stable normative matrix, profiles that emerge from clustering patterns "create momentary groupings that might disappear into the white noise of the database in the next moment."71 They therefore do not constitute a stable and fixed norm that defines and disciplines bodies according to a predefined model, but rather create a fluctuating network of categories that modulate behavior. In terms of Deleuze, such "modulation" functions "like a self-transmuting molding continually changing from one moment to the next, or like a sieve whose mesh varies form one point to another."72 In such dynamic networks, subjects can be assessed and sorted for specific purposes on a continuous basis. Categorizations are themselves evolving and shifting depending on feed-back data and the logic embedded in the algorithm.

What is called "autonomic profiling"73 goes one step further. Here, algorithmic functions are not only applied for the purpose of detecting patterns and generating profiles, but also for adapting the algorithms to changing environments and for automatic sorting and treatment of profiles, for example in the form of a differentiated or "(micro)targeted" response that is machine-generated.

An organizational example that comes close to autonomic profiling can be seen in concepts of "people analytics."74 The individual here is a "walking data generator."75 Sensor technologies automatically extract data from moving bodies. The algorithm built into the technology mines the data and autonomously transforms extracted data into a virtual profile, which is continuously updated and modified. People analytics visualizes individuals and teams and makes them comparable, measurable, and manageable. In this version, data-traces left by the moving bodies are calibrated as "derivatives." Amoore has described the specific form of abstraction that characterizes this form of creating profiles: "The data derivative comes into being from an amalgam of disaggregated data—reaggregated via mobile algorithm-based association rules and visualized in 'real time' as risk map, score or colour coded flag."76 In Table 1 the main differences are summarized.

In sum: In particular data-driven profiling based on (algorithmic) pattern recognition illustrates how emerging profiles are increasingly decoupled from the social relations from which they are drawn. Both, the process of arriving at categories and the multiple decisions about what becomes visible or invisible, what to select and what to exclude within the system become black-boxed and disappear in the created image. Such profiles emerge as "artefactual bodies" that can be and are disconnected from the individual. They can be combined with other profiles and they can be used for different purposes in different contexts. They can be continuously modified and they are extremely mobile, particularly in their electronic form. Thus, profiles become "prediction products"77 that circulate in cyberspace, can be sold and traded and turned into capital.

As pointed out, profiles are not descriptive but performative. They are not only a product of worldly and material processes, but in turn they do work in the world where they have material effects. They generate a frame for decision makers, produce the "evidence" to which decision-makers can refer to, they instruct decision-makers on what to see and what to ignore and they "act back on those with whom data are associated, informing us who we are, what we should desire or hope for, including who we should become."78 Application is the process in which profiles perform their work of informing or shaping decision-making in various places. Application is often seen as a technical, linear process in which context-specific 'solutions' or decisions are derived from abstract rules or concepts, or in our case, from profiles. By contrast, Derrida has suggested "a new concept of application which is in agreement with dissemination" and implies "a total spectral structure."79 Such a concept of application leads beyond the mere technical or instrumentalist understandings and considers application as a creative and performative process, which generates something unpredictable in "contexts which nobody can master in advance."80

In this process, profiles travel through time and space and have expected and unexpected effects. As ghosts, they are not bound by time and space. They can be created now, but return later; they can be produced here, but return elsewhere. Similarly, Amoore reminds us:

As data and profiles move from context to context, they can be interpreted and framed in different ways. Profiles can mean different things in different contexts. A regular buyer of antidepressant drugs can be an attractive customer for the pharmaceutical industry, but can be classified as an employment risk in other cases; a credit score can become a proxy for the trustworthiness of persons etc. Even though, any interpretation or sensemaking is context-bound, context cannot be fixed. Every profile is a sign or a mark that signifies some social status—creditworthiness, employability, riskiness, profitability, etc. As such, it can be removed from its "original" context. It can break from a given context and can be redeployed in new contexts; data collected for one specific purpose can be subsequently used for another authorized or unauthorized purpose, a phenomenon that has been described as "function creep."82 As is frequently demonstrated, profiles can easily flow between so-called "data sharing partnerships" that are created for optimizing the user experience. In moving from context to context profiles do not move in a vacuum, but they can be, and actively are transferred and transformed in the context of other profiles, actors, traditions, and institutions. The de/re-contextualization of profiles may change not only the content (the data-body) but also the ascribed meaning of the profile. Thus, when profiles travel form context to context, they may also be reframed according to specific interests and ideologies. Profiles are scrutinized, assembled, and reassembled for specific purposes in a variety of distributed "centres of calculation,"83 which—like Cambridge Analytica—seek to develop a knowledge base for commercial or political strategies for influencing and shaping behavior of customers or voters or, more generally, for nudging people to make decisions, which decision architects consider good decisions.

In the context of "surveillance capitalism,"84 profiles are the form of capitalizing data. Digital traces are mined and transformed into profiles that can be bought and sold, packaged and repackaged for specific purposes. Since every piece of data is of potential value for someone it might end up in some profile that is useful and can be sold to somebody. The monetization is the business of various data brokers, direct-marketers and other organizations and companies that sell and resell profiles to interested parties.85 In these conditions the emerging profiles are further removed from their primary source or context. As the legal scholar Frank Pasquale says, there are endless permutations:

What Pasquale has called "runaway profiles" illustrates the spectral logic of profiling, where data are decoupled and torn loose from their context in ways that make it impossible to understand how data from various sources are transformed into a profile, on what kinds of decisions, assumptions and prerogatives the constructed object has been formed, how it has been transformed and modified according to technical, economic or political considerations. While the decision process in itself becomes obscure and black-boxed, profiling and profiles—particularly when produced by "data-intelligence"—produce and reproduce the "illusions of transparency,"87 the promises of predictability and governability that support a (neoliberal) regime of truth and justify decisions and demarcations on the basis of seemingly rational and neutral calculations.

When abstractions circulate and are materialized (i.e., when the profiles are 'applied') they have powerful effects. They serve to allow or to deny access to services and benefits, they generate credibility or suspicion, and they provide the basis for a differential treatment of people. In particular, runaway profiles "can lead to cascading disadvantages as digital alchemy creates new analogous realities."88 Once a person is attributed the label of credit risk, toxic worker, marginal consumer, potential dropout or high-risk traveler, that attribute may appear with decisional force on various occasions throughout the social field. Profiles—regardless of whether we are talking about traditional (disciplinary) or about data-driven profiling, based on pattern recognition, mining of data, etc.—are discriminatory technologies that work as "social sorting."89 Profiling sorts people into social categories, so that the classification allows distributing opportunities and risks. While algorithmic and data-driven profiling is promoted as means to overcome human irrationality, to optimize judgmental accuracy and to produce more fine grained and more objective analyses, it is perhaps better understood as a movement in which mechanisms that generate demarcations become increasingly opaque and incomprehensible for those who are objects of profiling.

In the context of security governance, Leese has identified a triple "loss" that is associated with data-driven profiling (in comparison to traditional profiling).90 First, there is a "loss of traceability." With data-driven profiling, it is increasingly difficult to determine where data from which profiles are assembled come from, since they can literally come from everywhere. It is increasingly difficult to know and understand how categories are defined and how classifications are produced. Second, there is a "loss of visibility." Profiling increasingly operates out of sight. Data-mining procedures produce "artificial categories," the individual is likely not even to notice when or how he or she becomes categorized as risky, profitable, suspicious, etc. Consequently, most of the time people will not even notice if or how they have been discriminated against. Third, there is a "loss of accountability." In many cases, the human decision-maker is "rendered likely to comply with the truth claims of the algorithm"91 even if "autonomy" is attributed to him or her. The ability of individuals to reflect or challenge decisions vanishes to the degree that the process of making them becomes impenetrable. An inherent opacity characterizes predictive algorithms, particularly in the case of machine learning.92 Even experts are often overwhelmed by the complexity to comprehend the internal procedures of algorithmic systems. In many cases, it is difficult, if not impossible, to give an account of what the profile means and how it has been constructed.

I would add that there is—paradoxically—also a loss of predictability. While data-driven profiling seeks to make actions, thoughts, feeling, and behavior of citizens (customers, voters, debtors, employees, etc.) predictable, it becomes highly unpredictable how, when and for what reason created profiles will affect those who are profiled. Occasionally, we realize that we are in the visor of specific organizations and are targets of specific profiling machines. This happens, for example, when we are sent an individualized or personalized message, when we are offered a tailored product, are singled out as a risky passenger at the airport, or, when we are denied access to public spaces, insurance, employment, credit etc. because we (or our profile) does not fit to the criteria defined in advance by anonymous agencies. Moreover, there is a loss of response-ability. The ability to respond to the concrete other is a precondition for exercising or enacting moral responsibility. In his analysis of the Holocaust Bauman93 has analyzed how bureaucratic procedures and abstract classifications work as "moral sleeping pills." According to Bauman, dehumanization starts when people are objectified and reduced to a set of quantitative measures. In data-driven profiling where people and social relations are transformed into bits of information and electronic impulses, dehumanization works on a more abstract level. When the person appears on screen as a result of algorithmic visualization, the Other as a "face" (in Levinas' sense of "an authority without force"94 that demands a moral response) disappears completely and moral responsibility for the Other is suspended and rendered ineffective.

To sum up: Profiling increasingly creates an "invisible visibility" that is associated with what Derrida has called the "visor effect: we do not see who looks at us."95 We are observed and increasingly denied the possibility to understand, question, or challenge how, why, when, by whom we are observed and evaluated by what criteria. This also has consequences for the modes of subjectification and the specific ways in which truth and power are linked to each other.

Rouvroy has argued that algorithmic forms of profiling affect people in so many different contexts, yet without addressing them as persons directly. In her view, data-driven profiling circumvents subjectivity and subjectification, and in some sense even works "without subjectivity."101 Algorithmic governmentality, she says, "bypasses consciousness and reflexivity, and operates on the mode of alerts and reflexes."102 Yet it can also be argued that the emerging regime of truth creates conditions that are related to a specific mode of subjectification which Žižek has called "proto-paranoiac":

But the conditions of subjectification are even more paradox. On the one hand there is the phenomenon that Žižek describes. On the other hand, these data practices are widely regarded as normal, necessary or beneficial. There are a lot of incentives for making ourselves visible to various others, or to (actively) profile ourselves. In order to proof our creditworthiness or attractiveness to employers or potential partners, for example we provide our data and actively create our profiles. This is a new form of truth obligation that modifies the confession which according to Foucault "became one of the West’s most highly valued techniques for producing the truth,"104 which constitutes subjects in the double sense of the word. In the digital context one confesses in the form of providing data. Often this is not forced, but happens because of ignorance, laziness or indifference. The promise here is not salvation in the other world, but the promise is to get access to relevant services (e.g. credit, employment, friends, etc.) and pleasures in this world. In the digital context, the obligatory act of speech turns in many cases into a desiring self-exposure and exhibition of intimate details.105

Data-driven profiling provides the basis for a new mode of governing or, in Foucault's sense, a new "way of linking the art of governing and the game of truth."106 It is a way of linking them, which at first sight seems paradox:

Foucault traces this way of linking the art of governing and truth to the (physiocratic) idea, of "govern(ing) according to the rules of evidence." According to this idea, it would ideally be "things in themselves, rather than men, that govern." This, of course implies a utopian moment in which the exercise of power is nothing more than an "indicator of the truth."108

Today, in the context of the regime of practices which Zuboff109 has called "surveillance capitalism," a similar utopian moment is present in the work of prophets of the new data-driven society, who believe that total knowledge is within reach. They argue that "Big data give us a chance to view society in all its complexity, through the millions of networks of person-to-person exchanges […] we could potentially arrive at a true understanding of how society works and take steps to fix our problems."110 "Evidence" (i.e., truth) is derived from a multiplicity of data (Big data), drawn from a huge variety of sources and brought into a specific form by algorithms. It is thus, in the terms of Foucault, a specific form of evidence that is linked to a specific "modality of veridiction."111 Evidence here is not produced by or attributed to personal experience or courageous truth-telling (the parrhesiastic modality), it is also not produced with reference to some general form of knowledge (the modality of wisdom) nor by referring to professional expertise and (causal) knowledge (the technical modality). Rather, evidence is associated with the "objectivity" of data and the effectiveness of computational procedures. In this context, there is a multiplicity of generators of evidence, which provide the basis for a form of governance that is based on the assumption that "computational truth must necessarily replace politics as the basis of instrumentarian governance."112 In the context of ubiquitous computing, the idea that things in themselves govern takes on a new form that goes beyond Foucault's analyses. The more the things that surround us become "intelligent"—from cars to refrigerators, from cellphones to Alexa, etc. the more, Zuboff remarks, the "things that we have" are turned into "things that have us."113 Increasingly, driven by the utopia of "N = all" the whole spectrum of our experience, our homes and our bodies, our movements and our emotions, etc. are transformed into objects that can be calculated, processed and managed in the name of profit. Equipped with what Zuboff calls "actuation capability," technologies like smart sensors are interlinked and can register and analyze any kind of human behavior and then actually figure out how to change it. Automatically generated impulses and "nudges" condition and incentivize behavior and bypass human reflexivity. Governing the conduct of individuals, influencing, shaping, forming conduct may ultimately become automatized and freed from the incertitude and strain of interpretation, negotiation, and responsibility. This is ultimately hyper-facticity at work.