If...Then Algorithmic Power and Politics

The systematic collection of data about people has afected not only the ways in which we conceive of a society, but also the ways in which we describe our neighbour. It has profoundly transformed what we choose to do, who we try to be, and what we think of ourselves. (1990: 3)

If we accept Hacking’s notion of “making up people,” it becomes necessary to interrogate the ways in which counting, quantification, and classification limits the condition of possibility for subjects on Facebook. The manner in which the categories and classifications are constituted is not arbitrary or neutral; nor are the implications incidental. Here, I want to suggest that Facebook friends are “made-up people” in the sense described by Hacking. Friends are not natural kinds but, rather, constructions that serve specific purposes in a specific historical and cultural context. As has already been pointed out, the category “friend” as used by the Facebook platform does not even require members of that category to be human beings. While all Facebook users may belong to the set of “friends,” subsets of “friends” or “users” are dynamically being made up to serve diferent purposes. As exemplified in a Facebook patent application detailing the idea of adaptive ranking of news feed: “The social networking system divides its users into diferent sets, for example, based on demographic characteristics of the users and generates one model for each set of users” (Gubin et al.,  2014). In other words, sets are powerful classification devices implying a politics in terms of demarcating who or what belongs and what does not, “what the set ‘counts’ in, what counts as members in the set” (Baki, 2015: 37). Subsets are “made up” based on demographic information, atributes of connections, frequency of interaction, and other factors afecting the features used for modeling what users can see or not see (we will return to the politics of visibility in chapter 4). If friends are not of a natural kind, what kinds of friends or, rather, subsets are there? Again, we might turn to Facebook patent applications for some tentative answers. According to a document describing a technique for optimizing user engagement, some friends seem more valuable than others (Chen et al., 2014). Particularly useful friends are called “top friends,” defined as persons having the highest measures of relatedness to a specific user (Chen et al., 2014). The determinants of relatedness are generated using a so-called coeficient module. These, in turn, depend on a variety of factors (as is the case with all of Facebook’s computations)—for example, “based on how many times or how frequently interactions occurred within the last 30 days, 60 days, 90 days, etc.” (Chen et al., 2014). Top friends are used for a number of purposes and in a variety of contexts such as:

10

IF . . . THEN

To identify participants to play online games; to identify relevant connections to the user for inclusion in her social network; to display a listing of photos of persons having highest relevance to the user; to otherwise display or list an identification of persons having highest relevance to the user; to identify persons with whom the user can engage in an instant message or chat session; etc. (Chen et al. 2014)

Above all, top friends are used to prioritize information associated with them above others. Top friends are made-up people insofar as “those kinds of people would not have existed, as a kind of people, until they had been so classified, organized and taxed” (Hacking, 2007: 288). The subset of algorithmic top friends can be seen as a new category of people, emerging in the age of programmed sociality and algorithmic life. There are many more. As the notion of top friends shows, computable friendships hinge on measuring and evaluating users in order to be able to determine their friendship status. While friendships have always been qualitatively determined, as the notion of “best friend” suggests, the extent to which Facebook now quantifiably produces and classifies friendships works to dehumanize sociality itself by encouraging an empty form of competitiveness. Like most social media platforms, Facebook measures social impact, reputation, and influence through the creation of composite numbers that function as a score (Gerlitz & Lury, 2014: 175). The score is typically used to feed rankings or enhance predictions. The computing of friendships is no diferent. In another patent application Facebook engineers suggest that the value of friendship is not confined to users but also serves an essential role in sustaining the social networking system itself. As Schultz et al. (2014) suggest, beter-connected users tend to increase their use, thereby increasing advertising opportunities. A socalled friendship value is not only computed to determine the probability of two users “friending” but also to make decisions as to whether to show a specific advertising unit to a user. The higher the score, the “beter” Facebook deems the friendship to be, increasing the likelihood of using the connection to help promote products. The value of a friendship is produced as a composite number based on a “friendship score, sending score, receiving score or some combination of the scores as determined by value computation engine” (Schultz et al., 2014). According to Schultz et al., “the sending and receiving scores reflect the potential increase in the user’s continued active utilization of the social networking system due to a given connection” (2014: 2). From a computational perspective, friendships are nothing more than an equation geared toward maximizing engagement with the platform.

Far from loving the friend for the friend’s own sake, which would be exemplary of the Aristotelian notion of virtue ethics and friendship, Facebook “wants” friendships to happen in order to increase engagement with the social network, ultimately serving revenue purposes. The quantification and metrification of friendship are not merely part of how connections are computed by Facebook’s algorithmic infrastructure but increasingly make up the visuals of social networking systems through the pervasive display of numbers on the graphical user interface. With more than 3 billion likes and comments posted on Facebook every day, users are both expressing their sentiments and engagement and reminded and made aware of these actions and afects through the visual traces thereof. As the software artist Ben Grosser notes, “A significant component of Facebook’s interface is its revealed enumerations of these ‘likes,’ comments and more” (Grosser 2014: 1). Grosser questions whether people would add as many friends if they were not constantly confronted with how many they have or whether people would “like” as many ads if they were not always told how many others have liked them before them. Grosser’s artwork The Demetricator is a software plugin that removes all metrics from the Facebook interface and critically examines these questions. According to Grosser, Facebook draws on people’s “deeply ingrained ‘desire for more’ compelling people to reimagine friendship as a quantitative space, and pushing us to watch the metric as our guide” (Grosser, 2014). The pervasive enumeration of everything on the user interface function as a rhetorical device, teaching users that more is beter. More is also necessary if we consider the operational logics of the platform. The drive toward more is evident when considering the value of friendship, given that more friends increases the likelihood of engaging with the site. Friends are suggested based on mutual friends, but also on factors such as low activity or few friends. The idea is that, by suggesting friends with low activity level, Facebook “can enable those users to likely have more friends as a result of being suggested [. . .] and thereby likely increasing the candidate user’s engagement level with the social networking system” (Wang et al.,  2012: 6). Friendships, then, are variously made and maintained by humans and non-humans alike. The specifics of how friendships are susceptible to computation is immeasurable in and of itself. The purpose, however, of explicating the term “programmed sociality” as core to understanding algorithmic life is to draw atention to software and computational infrastructure as conditions of possibility for sociality in digital media.

Programmed Sociality

11

Guilt by Association

Whereas “friend” in the sociological sense signifies a voluntary relationship that serves a wide range of emotional and social aims, “friends” as seen from the perspective of the platform are highly valuable data carriers that can be utilized for a variety of reasons. One of the core principles underlying networked media is that information is derived as much from the edges (connections) as it is from the nodes (users, businesses, objects). This means that users do not just provide data about themselves when they fill out profiles, like things, or comment on posts; in doing so, they simultaneously reveal things about the people and things they are interacting with. If data are missing from a user’s personal profile or that user is not as engaged as the platforms would prefer, the best way to extract more information about the user is through his or her various connections. From a platform perspective, friends are in the data delivery business. This becomes particularly evident when looking at the patent documents by Facebook describing techniques related to advertising. For example, given the insuficient personal information provided by a particular member, ads are tailored and targeted based on friends. Facebook calls this “guilt by association” (Kendall & Zhou, 2010: 2). While the authors of the patent document acknowledge that they are “giving credence to an old adage,” the word “guilt” is worth pondering. Guilt evokes notions of responsibility, autonomy, and accountability. Who or what is responsible for the content shown on Facebook, and who might be held accountable? While it might be easier to understand how users’ own actions determine what content they see online, it seems more dificult to come to terms with the notion that users also play a crucial role in determining what their friends see on their feeds. Guilt by association, as Facebook uses the term, implies that users are made “complicit” in their friends’ ad targeting, which seems highly problematic. While it is now commonplace to say users are the product, not the media platforms they are using, the extent to which users are used to promote content and products—often, without their explicit knowledge—is unprecedented in the age of algorithmic media. If the classical notion of friendship is political in the sense that it assumes gendered hierarchy through the notion of brotherhood (Derrida, 2005), the politics of algorithms suggests hierarchies of a diferent sort— of what is “best,” “top,” “hot,” “relevant,” and “most interesting.”

12

I F . . . T H E N

When examining the contours and current state of algorithmic life, it is important to understand the mechanisms through which algorithmic media shaping sociality is deeply intertwined with power and politics. This book is not just about highlighting the role of algorithms as a core governing principle underlying most online media platforms today, but also about showing how algorithms always already invoke and implicate users, culture, practice, ownership, ethics, imaginaries, and afect. It means that talking about algorithms implies asking questions about how and when users are implicated in developing and maintaining algorithmic logics, as well as asking questions about governance, who owns the data, and to what end it is put to use? While friends have always been valuable for advertisers, algorithms seem to lessen the autonomy and intentionality of people by turning everything they do into a potential data point for the targeting of ads and news feed content. Such is the network logic, which users cannot escape. For neoliberalism, “friendship is inimical to capital, and as such, like everything else, it is under atack” (Cuterham, 2013: 41). Moreover, as Melissa Gregg holds, “‘friendship’ is labour in the sense that it involves constant atention and cultivation, the rewards of which include improved standing and greater opportunity” (2007: 5). As Langlois and Elmer point out, “social media seek to mine life itself ” (2013: 4). That is, social media platforms “do much more than just sell users’ atention to advertisers: they actually help identify the very strategies through which atention can be fully harnessed” (Langlois &  Elmer, 2013: 4). Algorithms are key to this end. If we want to understand the ways in which power and politics are enacted in and through contemporary media, we need to look more closely at the ways in which information, culture, and social life are being processed and rendered intelligible. In this book, I set out to do so.

Programmed Sociality

13

Examining algorithmic media and the ways in which life is increasingly afected by algorithmic processing, means acknowledging how algorithms are not static things but, rather, evolving, dynamic, and relational processes hinging on a complex set of actors, both humans and nonhumans. Programmed sociality implies that social relations such as friendships are not merely transposed onto a platform like Facebook but are more fundamentally transduced. The concept of transduction names the process whereby a particular domain is constantly undergoing change, or individuation, as a consequence of being in touch or touched by something else (Mackenzie, 2002; Simondon, 1992). Rather than maintaining an interest in what friendship is, transduction and the related term “technicity” help to account for how domains such as friendship come into being because of sociomaterial entanglements. Using Facebook to access a social network transduces or modulates how a person connects with friends. When using Facebook, the technicity of friendship unfolds as conjunctions between users and algorithms (e.g., the PYMK feature), coded objects (e.g., shared video), and infrastructure (e.g., protocols and networks). As Kitchin and Dodge point out: “This power to afect change is not deterministic but is contingent and relational, the product of the conjunction between code and people” (2005: 178). Transduction and technicity become useful analytical devices in exemplifying the concept of programmed sociality as they point toward the ways in which software has the capacity to produce and instantiate modalities of friendship, specific to the environment in which it operates. The productive power of technology, as signified by the concept of technicity, does not operate in isolation or as a unidirectional force. Algorithms and software, in this view, do not determine what friendships are in any absolute or fixed sense. Rather, technicity usefully emphasizes the ways in which algorithms are entities that fundamentally hinge on people’s practices and interaction, in order to be realized and developed in the first place. Taking such a perspective allows us to see friendship and other instances of programmed sociality as emerging sociomaterial accomplishments.

Back to the rainy November day introduced at the beginning of the chapter: The question of what categories were used to determine the specific ads or the content of my and my friends’ news feeds persists. Was it my clicking behavior, my age and gender, pages that I have liked, the cookies set by the online design store where I bought the gift for my mom, my friends’ clicking behavior, the friends of my friend, everything or nothing of the above? Whatever the exact reason might be, online spaces are always computed according to underlying assumptions, norms, and values. Although we simply do not know and have no way of knowing how exactly our data and the algorithmic processing are shaping our experiences online, a critical perspective on sociotechnical systems, along with personal encounters and experiences with algorithmic forms of connectivity and sociality, might help to illuminate the ways in which “categorization is a powerful semantic and political intervention” (Gillespie, 2014:171). Judging solely by the content and ads served up on my news feed, I am perceived as having children, being overweight, and single— none of which is true, at least for the time being. While the case of Facebook and programmed friendship provides a useful entry point to questions of how information is governed and organized online, an understanding of algorithmic power and politics cannot simply be reduced to a single social media platform. As this book will show, how we come to know others, the world, and ourselves as mediated through algorithms is the result of complex sociomaterial practices that exceed the specific coded instructions. Facebook and the notion of programmed sociality are but one way in which algorithmic arrangements bring new possibilities, realities, and interventions into being. And there are many more.

14

IF . . . THEN

By now, I hope to have instilled enough curiosity in the reader to keep you exploring the power and politics of algorithms in the contemporary media landscape with me throughout the next chapters. The goal of the book is for readers to not simply be the subjects of algorithmic judgment but, rather, to be in a position to critically judge the workings of the algorithmic apparatus for themselves.

Outline of the Book

The overall aim of this book is to sketch the contours of an algorithmic media landscape as it is currently unfolding. While algorithms and software are starting to catch the interest of social scientists and humanities scholars, having become somewhat of a buzzword in media and communication studies during the past years, we are only at the beginning of understanding how algorithms and computation more broadly are afecting social life and the production and dissemination of knowledge as we know it. This book seeks to contribute to these discussions by ofering conceptual, theoretical, and empirical analyses of the ways in which algorithms produce the conditions for the sensible and intelligible.

In the chapters 2 and 3, which comprise the conceptual framework of the book, I focus on the ontological, epistemological, and methodological dimensions of algorithms. Chapter 2 provides an outline for understanding what algorithms are and how they are conceptualized in diferent manners. While the chapter functions as a conceptual introduction to the interdisciplinary field of critical algorithms studies, merging perspectives from computer science, social sciences and the humanities, it mainly does so for analytical reasons. I argue that these are not simply diferent perspectives on a static object called an algorithm, but rather, following insights from STS (Law, 2002; Mol 2002), provide diferent versions of what an algorithm is. Even if we assume that we are talking about the same algorithm (say, the “Facebook algorithm” or “K-nearest neighbor”), the algorithm is always “many diferent things. It is not one, but many” (Law, 2002: 15).

Programmed Sociality

15

Chapter 3 tackles the question of how algorithms can be known, proposing an epistemology of algorithms that moves beyond the popular conception of the algorithm as a black box. I draw on the concept of the black box as a heuristic device to discuss the nature of algorithms in contemporary media platforms, and how, we might atend to and study algorithms, despite, or even because of, their seemingly secret nature. The chapter develops a conceptual repertoire that moves beyond the textbook definition of algorithms as step-by-step instructions for solving a computational problem. It does so to beter account for the fundamental interlinking or entanglement of the social and material implicated by algorithms. To support such a view, chapter 3 builds on insights from relational materialism and process-relational philosophy (e.g., Barad, 2007; Mol, 2002; Whitehead, 1978). Here, I suggest a conceptual vocabulary that allows for investigating algorithms as eventful, understood as constituents that co-become, and their power and politics as tied to the ways in which these configurations have the capacity to produce certain orderings and disorderings of the world. In chapter 2 I introduced the notion of “ontological politics” (Mol, 2002) to convey how realities are never given but shaped and emerge through interactions. In chapter 3 I expand on these ideas by suggesting, perhaps somewhat paradoxically, that algorithms are not always important. Rather, their agency emerges as important only in particular setings or constellations. The argument is made that by shifting atention away from asking what and where agency is, to when agency is and to whom it belongs in specific situations, we see how the notion of algorithms as black boxes is a political claim. The chapter concludes by ofering some possible methodological routes for the study of algorithms premised on their eventfulness, which are operationalized in the chapters that follow.

Having outlined the ontological and epistemological contours of algorithms, the chapters 4–6 turn to specific constellations of algorithmic power and politics as it materializes is the contemporary media landscape. Each chapter presents a case study that is meant to get at two related concerns: the ways in which algorithmic systems govern the possible field of action of others, and how these possibilities are made more or less available or unavailable to certain actors in specific setings. Taken together, these case studies are concerned with the world-making capacities of algorithms, questioning how algorithmic systems shape encounters and orientations of diferent kinds, and how these systems are endowed with difused personhood and relational agency.

Chapter 4 pays atention to the mechanisms of power that work as a concerted distribution of people, information, actions, and ways of seeing and being seen. It uses Facebook’s news feed as a case study to develop the argument that algorithms now play a fundamental role in governing the conditions of the intelligible and sensible. The chapter looks at how the news feed operates algorithmically to govern the “distribution of the sensible [. . .] defining what is visible or not in a common space” (Ranciere,  2004: 12–13). While users feed Facebook with data, the techniques and  procedures to make sense of it, to navigate, assemble, and make meaningful connections among individual pieces of data are increasingly being delegated to various forms of algorithms. The question raised in this chapter is, how does this kind of algorithmic intervention into people’s information-sharing practices takes place? What are the principles and logics of Facebook’s algorithmic form of editing the news feed? What implications do these algorithmic processes have for users of the platform? Through an analysis of the algorithmic logics structuring the flow of information and communication on Facebook’s news feed, I argue that the regime of visibility constructed, imposes a perceived “threat of invisibility” on the part of the participatory subject. As a result, I reverse Foucault’s notion of surveillance as a form of permanent visibility, arguing that participatory subjectivity is not constituted through the imposed threat of an all-seeing vision machine, but by the constant possibility of disappearing and becoming obsolete. The intention is not so much to ofer a definite account of the role played by Facebook in capturing the world in code, but to open avenues for reflection on the new conditions through which in/visibility is constructed by algorithms online.

16

I F . . . T H E N

Chapter 5 considers the barely perceived transitions in power that occur when algorithms and people meet, by considering how social media users perceive and experience the algorithms they encounter. While it is important to interrogate the operational logics of algorithms on an infrastructural level, materiality is only half the story. To do the topic of algorithmic power and politics full justice, there is a need to understand how people make sense of and experience the algorithms with which they persistently interact. Technical systems and infrastructure alone do not afect use. Users’ perceived knowledge of how the systems work might be just as significant. The questions raised in chapter 5 are: How do social media users imagine algorithms, and to what extent does their perception and knowledge afect their use of social media platforms? The chapter reports findings from an exploratory study of 35 social media users who were asked about their perceptions of and experiences with algorithms online. The chapter examines the specific situations in which users notice algorithms and start reflecting and talking about them. Focusing on a few of these user-reported situations, the chapter shows how users respond to and orient themselves diferently toward algorithms as. Moving beyond a call for intensified code literacy, I argue that these personal algorithm stories provide important insight into the ways in which algorithms are currently imagined and understood, and how users negotiate and resist algorithms in their everyday life.

Chapter 6 looks at how algorithms acquire the capacity to disturb and to compose new sensibilities as part of situated practices, particularly in terms of how they become invested with certain political and moral capacities. While the previous chapters considered how algorithms are publicly imagined and how they work to produce impressions of engagement, chapter 6 looks at how algorithms materialize in the institutional seting of the news media. More specifically, in this chapter we  continue to consider how algorithms are not just maters of fact but also an important mater of concern (Latour, 2004). By taking the field of journalism as a case in point for understanding the politics and power of algorithms as they are manifested in the current media landscape, I aim to illuminate how algorithms are always diferently materialized—how they are made out and carved into existence at the intersection of technology, institutional practice, and discourse. I make an argument to the efect that the ontology of algorithms is up for grabs. Based on 20 interviews with digital editors and managers at leading Scandinavian news organizations, as well as on field observations at the Swedish news app Omni, chapter 6 explores how institutional actors are responding to the proliferation of data and algorithms. If algorithms have an audience or are otherwise consumed, as previous chapters have highlighted, they are also produced and variously put to work, which is the focus of this chapter. The question is how and when algorithms come to mater and whom they mater for. On the one hand, news organizations feel the pressure to reorient their practices toward the new algorithmic logic governing the media landscape at large, and often with a close eye to what’s perceived as one of their biggest competitors—Facebook. On the other hand, the interview data suggest that algorithms work to disturb and question established boundaries and norms of what journalism is and ought to be. In order to understand the power and politics of algorithms in the contemporary media landscape, my claim is that it is vital to understand how algorithms are neither given nor stable objects, but rather made and unmade in material-discursive practices. Sometimes algorithms are rendered important, at other times, they are deemed insignificant. Knowing more about processes of matering, I suggest, enables us to understand the multiple realities of algorithms, and how these relate and coexist.

Programmed Sociality

17

Taken together, the chapters ofer diferent versions of how and when algorithms are made present and important to individual and institutional actors. An important suggestion here is that power and politics are never reducible to question of materiality or perception, but rather about making algorithms mater in specific ways and for specific purposes. In the concluding chapter, I revisit the core arguments of the book, and suggest that there’s a need to blend an understanding of the material substrates of algorithmic media with an understanding of the multiple ways on perceiving, feeling, acting, and knowing which coalesce around algorithms as an object of social concern. Here, I revisit some of the initial questions asked at the beginning of the book (including this introduction), and look at how algorithmic power and politics can be understood if power and politics are not necessarily about imposing force from above (or below for that mater, if understood as code). Ultimately, this chapter serves to summarize the key contributions of the book in terms of: (1) providing an understanding of algorithms that is not commited to one ontological position, but instead sees algorithms in terms of a multiple and variable ontology; (2) helping to identify forms of algorithmic power and politics; and (3) ofering a theoretical framework for the kinds of work that algorithms do and the landscapes they help to generate.

18

I F . . . T H E N

The Multiplicity of Algorithms

Consider the following descriptions of what an algorithm is: “a step-by-step instruction of how to solve a task,” “a recipe,” “a form of programmed logic,” “made of people,” “code that tries to accommodate personal interests,” “an application that determines what you see,” “a type of technology,” “an automated filtering mechanism,” “what platforms use to guarantee the best user experience possible,” “an appetizer to make you crave for more of the same,” “magical.” You may already recognize some of these descriptions from standard computer science textbooks, common language, media reports, or from the way in which platforms market their algorithms. Indeed, these descriptions are the accounts of computer scientists, marketers, lay people, critics, journalists, or media users. These are the textbook definitions, tropes, PR slogans, tweets, folk theories, and popular descriptions of algorithms. Yet you may wonder, are these people describing the same thing? What does magic have to do with a recipe? Can something both determine what you see and accommodate your own interests?

This chapter suggests that algorithms are multiple in the sense that there is more than one kind of thing that we call “algorithm.” Not only is this true in the technical sense of there being many diferent types of algorithms but also in the social sense. When people speak of algorithms they often have diferent things in mind, diferent concerns, worries, conceptions, and imaginations. In a more theoretical and philosophical sense, one can talk of such multiplicity as “manyfoldedness, but not pluralism” (Mol, 2002: 84). Whereas plurality assumes that there is one stable thing and diferent perspectives one can take on that thing, to speak of manyfoldedness implies the idea that algorithms exist on multiple levels as “things done” in practice. In other words, algorithms can be magical and concrete, good and bad, technical and social. Yet, the notion of multiplicity also reminds us that puting it in such binary terms risks perpetuating an overly simplistic understanding of what algorithms are and what they can be. What is at stake, then, is the meaning of a variable existence for the ways in which we understand algorithmic power and politics. While I want to stress a view on algorithms that sees it not as one object (or subject, for that mater) but many, I shall first need to say a few things about the ways in which algorithms are commonly understood.

2

19