If...Then Algorithmic Power and Politics

Conclusion

Given the 1.28 billion active daily Facebook users, algorithmic practices, techniques, and procedures play a powerful role in deciding what gets shown, to whom, and in what ways. Algorithms are never neutral but reflect the values and cultural assumptions of the people who write them. This does not mean that we can simply understand algorithms by asking programmers. While ethnographic work exploring how algorithms are created may tell us something about what goes into the ways an algorithm functions, programmers are no more equipped than other people to talk explicitly about their biases, values, or cultural assumptions. Instead, this chap ter reveals how an understanding of the power and politics of algorithms can be approached by considering their sociomateriality. Guided by the question of how the news feed works, for what possible purpose, I argued for a critical, architectural reading of algorithms. Through a close reading of publicly available documents de scribing the workings of the Facebook news feed as well as systematic observations and analysis of my own news feed, several aspects were revealed about the ways in which sociality and subjectivity are imaged by the platform and how it is systemati cally embedded and engineered into algorithmic arrangements and mechanics. Despite the emphasis on the material politics of algorithms in this chapter, material ity is never decoupled or distinct from the domain of the social. With much recent focus on the algorithm as an object of analysis within the social sciences and humanities, there has been a certain tendency to imbue too much causal power to the algorithm as if it were a single, easily discernible thing. What has become apparent is that there is an urgent need to historicize algorithms, to put algorithms into broader contexts of production and consumption. At the same time, cultural theorists have to grapple with the inevitable ephemerality of social media algorithms. In this chapter, I have tried to engage with this historicity by considering the cultural beliefs and values held by creators of software and by being explicit about the dates and times of data capture. This is to say that we need to acknowledge that the news feed and its algorithm(s) emerge from somewhere; evolve in particular ways; are talked about in certain terms; mystified; reflect certain beliefs and assumptions; work to accomplish certain technical, economic, and cultural goals; stabilize temporarily, then change again; and so on. It also means that tapping into the algorithmic assemblage will only ever be a tapping into a momentary tracing and stabilisation of a series of events.

Indeed, the analysis of the news feed algorithms presented in this chapter reveals the “variable ontology” characteristic of software (Mackenzie,  2006). While it might be a stretch to claim that Mark Zuckerberg’s love of gray T-shirts is an algorithmic metaphor, the piece of clothing does reveal something about the values and beliefs he holds. In spite of restricted access to Facebook engineers, we still find plenty of traces of how platforms operate, how their CEOs and managers think and talk, how code is made and work organized, the ways in which culture maters, how culture gets branded, and the values, norms, and cultural assumptions held by the people involved. These traces include but are not exclusive to the principle of moving furniture, the notion that the physical layout of Facebook’s headquarters is designed to change the thinking of Facebook employees, the technical specifications described in patent documents, the engineering rhetoric perpetuated at developer conferences, media reports, and user interface. The study shows how algorithms are ontogenetic simply because the problems that require a solution continuously change. While the problem of showing the most “interesting” content remains, what constitutes interestingness or relevance depends on the given context. Facebook is never finished. Whether we are talking about the physical refurbishing of sofas or the news feed ranking algorithm, the object of study is inherently dynamic, changing, and mutable. In an age of machine learning in which the models used to predict user behavior constantly change as a result of increasing user data, it may prove particularly challenging to capture the underlying logic of how the system works. If, as Foucault suggested, disciplinary power is efectuated by means of training subjects to reach their full potentiality as useful individuals, what we see emerging now is a certain disciplining of the algorithm by means of training the machine to learn correctly from a corpus of known data. What we are still only at the beginning of understanding, let alone addressing in the first place, is what “correct training” means and which techniques are put into place to ensure that algorithms reach their full potentiality as useful machines.

90

I F . . . T H E N

Many of the characteristics associated with disciplinary power described by Foucault, such as the function of enclosure, the creation of self-control, and the training of human multiplicity, are apt characterizations of the kind of enclosed architecture of Facebook and its subtle demands for participation and interaction. However, if we follow Foucault in his understanding of surveillance as a form of “permanent visibility,” then this notion fails to capture the algorithmic logic of creating modalities of visibility that are not permanent but temporary, that are not equally imposed on everyone and oscillate between appearing and disappearing. While it is true that Foucault described organizations of power within a somewhat fixed technological and architectural form, the idea that architectural plans structurally impose visibility does not seem to come in conflict with the unstable and changing arrangements characteristic of new media. Spaces, both digital and physical, are always delimitated in one way or another. That does not mean they cannot also expand or change as new elements come along. As Foucault sees it, discipline is centripetal, security centrifugal. This is to say that discipline functions to isolate a space or determine a segment, whereas the apparatuses of security “have the constant tendency to expand” by having new elements integrated all the time (Foucault, 2007: 67). Security does not supersede discipline, just as discipline does not replace sovereignty but rather complements and adds to it (Foucault, 2007: 22).

Life at the Top

91

With reference to Foucault’s concept of panopticism, my aim in this chapter has been to argue for the usefulness of applying an analytics of visibility to (im)material architectures. Following Foucault’s assertion that “the Panopticon must be understood as a generalizable model of functioning; a way of defining power in terms of the everyday life of men” (1977: 205), a diagrammatic understanding of the news feed algorithm provides a constructive entry point for investigating how diferent regimes of visibilities materialize. Taking the material dimensions of social media seriously reveals how politics and power operate in the technical infrastructure of these platforms. In Foucault’s terms, the news feed can be understood as a form of government in which the right disposition of things are arranged to lead to a suitable end (2007: 96). Algorithms, it was argued, are key to the disposition of things or what Ranciere would call the distribution the sensible. What the operational logic of the news feed ranking algorithm reveals is a reversal of Foucault’s notion of surveillance. Participatory subjectivity is not constituted through the imposed threat of an all-seeing vision machine but, rather, by the constant possibility of disappearing and becoming obsolete.

92

I F . . . T H E N

Affective Landscapes

Everyday Encounters with Algorithms

I have been a Twiter user for more than ten years now, but I still do not know how to use it. Tweeting does not come naturally to me, and I would not really be able to explain what the platform is for. However, like most users, I think of Twiter as a real-time micro-blogging platform from which you can tweet small snippets of thoughts, pieces of information, or links to a group of followers. For many, the realtime reverse chronological feed is the most defining aspect of Twiter’s user experience. It, therefore, felt somewhat transgressive when Twiter announced in early 2016 they would introduce an “algorithmic timeline” to replace its iconic real-time feed. It did not come as a surprise that the news that Twiter was preparing to change this defining feature completely would cause a massive public outcry. Like any big platform change nowadays, enraged users gathered around a hashtag to protest. For days and weeks, #RIPTwiter demonstrated the dismay people felt at the arrival of the algorithm. “In case you never hear from me again, you can thank Twiter’s genius new algorithm system . . . #RIPTwiter.”1 “The main thing I love about Twiter is that it does not use an algorithm. Sounds like that’s changing. #RIPTwiter.”2 “Every time I hear the word ‘algorithm’ my heart breaks. #RIPTwiter.”3 I want to start this chapter with the case of #RIPTwiter, not because the user protests surrounding platforms are somehow exceptional, but because of the emerging ordinariness of algorithms that these tweets and many more like them seem to convey. As Christian Sandvig (2015) notes, in the years 2006–2013, “There has been a five-fold increase in the number of times the word ‘algorithm’ appeared in the major newspapers of the world.” This number has likely just exploded since 2013, with the news media now routinely reporting on algorithms. While many people are still not aware of the extent to which algorithms are curating their online experiences (Eslami et al., 2015), we might nevertheless—or precisely because of it—sense the emergence of a new “structure of feeling” (Williams, 1977) surrounding algorithms as evident in various emerging discursive impulses surrounding the algorithm.4

93

The starting premise for this chapter is that algorithms are becoming habitual and part of a felt presence as people inhabit mediated spaces. This is not to say that algorithms are necessarily consciously present to people but that they have the capacity to call forth “something that feels like something” (Stewart,  2007: 74). Sometimes, this something may take the form of an event such as the #RIPTwiter protest, a moment when “collective sensibilities seem to pulse in plain sight” (2007: 74). Most of the time, however, algorithmic impulses become barely noticeable in the mundane moments of everyday life. Recall, for example, the targeted ads for the Lisbon hotel recommendation or the party dress that seemed to creep up on me while I was surfing the Web (as described in chapter 1), or the oddity of having your credit card company put a security ban on you without being able to explain why (as mentioned in chapter 3). In moments like these, algorithms make themselves known even though they often remain more or less elusive. They become strangely tangible in their capacity to create certain afective impulses, statements, protests, sensations and feelings of anger, confusion or joy.

In this chapter, we move away from the disciplinary power of algorithms explored in the previous chapter toward the “micropolitics” of power imbued in the afective and phenomenological dimensions of algorithms. By “micropolitics,” I mean something akin to Foucault’s (1978) notion of microphysics as a way to pay atention to the generative potential of specific everyday practices and techniques. More specifically, micropolitics “refers to the barely perceived transitions in power that occur in and through situated encounters” and the idea that “diferent qualities of encounter do diferent things” (Bissell, 2016: 397). This is not a kind of power and politics that is necessarily repressive, discriminatory, or hierarchical but, rather, productive in the sense that it produces certain capacities to do and sense things. What I am interested in is the question of how algorithms and people meet, and what these encounters make possible or restrict? The notion of afect becomes important in terms of describing what is at stake in thinking about the barely noticeable and more sensible dimensions of algorithms as a key to understanding the power and politics of algorithmic life. Without going much further into the notion of “afect” (many books have been dedicated to this concept alone), sufice it, at this point, to say that “afect” describes the domains of “the more-than or less-than rational” in life, including “mood, passion, emotion, intensity, and feeling” (Anderson, 2006: 734).5 I specifically focus on the ways in which algorithms have the capacity to afect and be afected (Deleuze & Guatari, 1987) by looking at how users perceive and make sense of algorithms as part of their everyday lives.

The point, however, is not to assess the extent to which people actually feel algorithms as such but to highlight experience and afective encounters as valid forms of knowledge of algorithms (see chapter 3 for a discussion on the phenomenological approach to algorithms). The argument is made that algorithms do not just do things to people, people also do things to algorithms. The word to is important as it extends more classic accounts of media use and arguments of audience power. It is not just that people do things with the help of algorithms. By using algorithms, they also do things to them—modulate and reconfigure them in both discursive and material ways. As we discussed in chapter 2, the machine-learning algorithms proliferating on social media platforms are never set in stone. These algorithms are continuously molded, shaped, and developed in response to user input. If the previous chapter made a case for analyzing algorithms as governmental techniques directed at the “right disposition of things” through rankings and weights as architectural forms of power, this chapter considers the “productive power” of cultural imaginaries and the significant role users have in reconfiguring the algorithmic spaces that they themselves inhabit.

94

I F . . . T H E N

Encountering Algorithms

Consider the following scene: Michael presses the post buton and waits. Normally, it should not take longer than five minutes before the “likes” and “comments” start ticking in. Nothing happens. As an independent musician, Michael has to find his own ways to spread the word about his music and make sure he reaches an audience. Facebook seems like the perfect platform for self-promotion. Michael thinks he has goten beter at playing “Facebook’s game,” as he calls it. For example: “Statuses do beter based on what night you post it, the words you choose to use, and how much buzz it initially builds.” He knows from previous experience that, “if the status doesn’t build buzz (likes, comments, shares) within the first ten minutes or so, it immediately starts moving down the news feed and eventually gets lost.” Michael has just released a new album and needs to get the word out. He picks what seems to him the perfect day of the week, carefully crafts the words of the update, deliberately uses phrases such as “wow!” and “this is amazing!” to make the post more visible. Or so he thinks. But nothing happens, and the post eventually disappears, no downloads, just some scatered “likes.”6

One more: Rachel is an avid Facebook user, but lately she thinks “the Facebook algorithm is destroying friendships.”7 She’s been a Facebook member almost since the very beginning and has over 700 friends. She considers herself a fairly sociable and outgoing person, who likes to post regular status updates and frequently comments on other people’s posts. Then, a few weeks back, she saw a post by an old friend from high school. Rachel had totally forgoten they were even Facebook friends. She had not seen this old friend in ages, and, all of a sudden, she “pops up in her news feed.” Rachel is curious—how much has she been missing out on? What’s been going on in her friend’s life that Facebook decided not to broadcast on her own feed? “I’m constantly taken aback by all the info and people Facebook hides from my feed on a daily basis,” Rachel says. “So, in that sense, it does feel as if there is only a select group of friends I interact with on the social network, while I’ve practically forgoten about the hundreds of others I have out there.”8

Affective Landscapes

95

More than simply describing the strange feelings of media users, these scenes describe some of the many mundane moments in which people variously encounter the algorithmic realities and principles underlying contemporary media platforms. For 21-year-old Michael, the Facebook algorithm has become an important part of his everyday life. As a musician, he depends on the Facebook platform to promote his music, connect with fans and followers, and engage in a community of likeminded artists in L.A., where he lives. Michael describes his encounters with the algorithm as a game of popularity that hinges on the number of likes and comments his updates are able to atract. In many ways, Michael’s story is reminiscent of the “threat of invisibility” described in the previous chapter. To counteract this threat, Michael has developed strategies and tactics that he thinks will guarantee him a beter chance at retaining atention and geting noticed. Based on previous experience and videos about the underlying logic of the news feed, Michael has developed theories about how the algorithm works. What is frustrating, he says, is when it turns out not to be true. For Rachel, the Facebook algorithm is not subject to the same kind of frustration as it is for Michael. Rachel, a 24-year-old journalist based in New York City, does not “depend” on the algorithm for her own livelihood in the way that Michael does. As a journalist, however, she has noticed that there seems to be an increasing number of news reports on algorithms lately. An article in the Washington Post dealt with the Facebook algorithm; and, ever since, Rachel has been trying to find signs of it in her own news feed. When an old friend from high school appeared in her feed seemingly from nowhere and even “liked” one of Rachel’s posts, she finally sensed the algorithm at work. With respect to forgeting people, Rachel says, it feels like the algorithm is trying to make decisions on her behalf, ultimately influencing the way in which she conducts her friendships. The question going forward is how we might want to think about such everyday encounters with algorithms and what, if anything, they might tell about the ways in which algorithms shape our experiences?

Asking about people’s encounters with algorithms seems to suggest a conscious meeting of some sort or that people are already aware of and have a grasp of what algorithms are. As existing research has shown this is far from always the case (Eslami et al. 2015; Rader & Gray, 2015). In many ways, the question of how people atend to algorithms is akin to the question of how people start to make sense of that which is not readily available to the senses, which has always been an important area of research in the history of science.9 Many scientific and technological innovations are not visible as such. Just think of nanotechnology, RFID, or sensor technologies of various kinds. Like algorithms, they are embedded in computers, pieces of clothing, and other more visible interfaces. Dealing with uncertainties and unknowns is part of what it means to be a human; it is part of what it means to live toward a future, what it means to be in love, or what it means to understand things such as physics and climate change. When people are faced with the hidden and uncertain aspects of life, they tend to construct what has variously been described as “mental models” (Schutz, 1970; Kempton, 1986), “interpretive frames” (Bartunek & Moch, 1987), or simply “folk theories” (Hirschfeld, 2001; Johnson-Laird & Oatley, 1992; Johnson, 1993). To interact with technologies, people have to make sense of them, and they do so in a variety of ways—for example, through visualizations, analogies to more familiar domains, or by the use of metaphors. Media discourses, cultural artefacts, stories, anecdotes, and shared cultural beliefs can “provide people with explanatory power and guide behavior surrounding use of a particular technology” (Poole et al., 2008). The increased ubiquity and popularity of depicting unfamiliar phenomena such as artificial intelligence and robotics in films and popular imagery have arguably made these phenomena more available to public reasoning and debate (see Suchman, 2007).10 This is no diferent when it comes to the algorithm, which is rarely encountered in its mathematical or technical form. When people encounter or “see” the algorithm, it is typically in the form of simplified images of a formula (e.g., EdgeRank formula) or a newspaper story about a recent controversy as evident, for example, in the recent censorship accusations against Facebook and their “algorithm” removing the iconic “Terror of War” image or through other means of cultural mediation such as films and artworks of various kinds.

96

I F . . . T H E N

Unsurprisingly, perhaps, algorithms are dificult to represent or culturally mediate even for computer scientists.11 However, as Sandvig (2015) notes, in an atempt to make the algorithm more intelligible to the public we are seeing an epistemic development in which “algorithms now have their own public relations.” That is, beyond the technical or educational imagery of algorithms characteristic of earlier times, algorithms materialize in diferent representational setings. Algorithms have become objects of marketing, represented in commercialized images of mathematical formulas and circulated by various third-party marketing consultants, technology blogs, and other trade press publications. The process of making algorithms a product of public relations is not so diferent from what Charles Bazerman (2002) calls the production of “representational resting points.” In examining the cultural history of Edison’s electric light, Bazerman argues that, before the invention “could create its own new world of experience and meaning,” it was crucial to develop accepted representations of electric light as stable points of reference (2002: 320).12 What exactly these representational resting points are with regards to algorithms and machine learning is still up for grabs, but simplified mathematical formulas or cyborg figures are certainly part of it. Though the formula may portray algorithms as objects of science and neutral conduits, this is arguably not what people experience in their day-to-day life. As the stories of Michael and Rachel suggest, what people experience is not the mathematical recipe but, rather, the moods, afects, and sensations of which algorithms are generative. The notion of mental models, then, only brings us so far—not because the theories that people construct about unfamiliar technologies and hidden processes often prove to be inaccurate (Adams & Sasse, 1999; Wash,  2010), but because it does not necessarily mater whether they are correct or not.

Affective Landscapes

97

From a phenomenological perspective, what is of interest is how the seemingly hidden and unfamiliar technological processes permit or limit people’s movements and whereabouts. For people in their everyday lives, what maters is not necessarily the truth (whatever that is) but “practical knowing” (Thrift,  2007: 121); that is, people “know as they go, as they journey through the world along paths of travel” (Ingold, 2011: 154). In other words, people learn what they need to know in order to engage meaningfully with and find their way around an algorithmically mediated world. For Michael, for example, Facebook’s news feed algorithm feels like a gatekeeper that stands in the way of his music career. Knowing the algorithm just enough to ensure that his followers receive his updates maters to him—not just personally but also professionally. The kind of knowledge that I am concerned with here, then, treats the beliefs people form about algorithms not as right or wrong but as a point of access for understanding how and when algorithms mater.

Researching Personal Algorithm Stories

Contrary to the notion that “the individual user is incapable of really experiencing the efect that algorithms have in determining one’s life as algorithms rarely, if ever, speak to the individual” (Cheney-Lippold, 2011: 176), I suggest in this chapter that people do experience algorithms; and, while algorithms might not speak directly to individuals, they might speak through them. Here, I report on a series of personal algorithm stories, that is, stories about algorithms derived from people’s online experiences. Over a two-year period from October 2014 to August 2016, I collected personal algorithm stories from social media users through email and face-to-face interviews with 35 individuals about their perceptions and awareness of algorithms. The data set also contains over 2000 tweets writen by social media users that contain popular hashtags such as #RIPTwiter and #InstagramAlgorithm as well as blog posts and archived comment sections of newspaper articles discussing algorithms. Taking my lead from Zizi Papacharissi (2015), who sees the collective discourses organized around hashtags on Twiter as indicative of an emergent structure of feeling, I shall expand this notion to include not just hashtags but also disparate tweets about algorithms as indicative of an afective landscape surrounding the felt presence of algorithms. Going back to Michael and Rachel, their algorithm stories were originally collected using their own tweets as a starting point. Expressing feelings of bewilderment and defeat, their tweets simply read: “This Facebook algorithm is just super frustrating” (Michael) and “Facebook algorithm destroys friendships” (Rachel).13 Seeing these kinds of uterances as part of my Twiter feed, I could not help wondering where they came from. What had prompted Michael and Rachel to tweet those things?

Methodologically, I engage in what might be called a scenographic inquiry into the afective landscapes generated in and through algorithms by atending to the scenes, situations, episodes, and interruptions that give rise to the felt presence of algorithms in everyday life. By “scenographic,” I mean a methodological sensibility akin to writers and scholars such as Lauren Berlant and Kathleen Stewart, both of whom atend to cases or scenes of the ordinary as a way of making afective processes more palpable. For Stewart (2007), scenes are method and genre, a way of writing about ordinary afects as a series of happenings that, taken together, form a story about what it feels like to live in contemporary America. As McCormack writes, the scene or situation “becomes a way of gathering the sense of worlds that mater while also posing the question of how the force of these worlds might become part of their stories” (2015: 93). A situation, as Berlant defines it, “is a state of things in which something that will perhaps mater is unfolding amid the usual activity of life. It is a state of animated and animating suspension,” one “that forces itself on consciousness, that produces a sense of the emergence of something in the present” (2011: 5). Algorithms, I suggest, may be productive of such emerging presences and “accessed” via the ways in which they make people feel and the stories that they tell about those encounters. This is closely related to the notion of afect but not exactly the same. As Papacharissi (2015) points out, afect is what permits feelings to be felt, it is the movement that may lead to a particular feeling. Afect, she suggests, can be thought of as the “rhythm of our pace as we walk” (2015: 21). For example, a fast-paced rhythm may lead to and amplify feelings of stress; a slow and light-paced rhythm may make us calm. In the context of this study, the question is how we might think of the force of movement that algorithms create, and how it prompts a “reason to react,” as Stewart puts it (2007: 16).

98

IF . . . THEN

Taking tweets about algorithms writen by seemingly ordinary people as my starting point, I was interested in geting a beter sense of the kinds of situations and circumstances that had solicited people to react and respond to the ways of the algorithm.14 During the two years of data collection, I regularly searched Twiter for keywords and combinations of keywords, including: “Facebook algorithm,” “algorithm AND Facebook,” “Netflix algorithm,” “algorithm AND Netflix” and so forth.15 I queried Twiter every few weeks and manually scrolled down the stream of tweets and took screenshots of the ones that seemed to be more personal rather than marketing-oriented. Using a research profile I had set up on Twiter, I occasionally contacted people who had recently tweeted about algorithms to ask whether they would be willing to answer a few questions related to that tweet. These tweets included statements such as: “I’m ofended by the awful taste Twiter’s algorithm thinks I have,” “Facebook, perhaps there isn’t an algorithm that can capture human experience,” “It feels like Spotify’s algorithm can read my mind—it’s both awesome and creepy,” “The Netflix algorithm must be drunk,” “The algorithm on Facebook confuses me. It seems to think I should lose weight, am broke, pregnant and single.” Altogether, I engaged in asynchronous email conversations and synchronous computer mediated chats with 25 individuals, based on questions concerning those tweets.16 During the spring of 2016, I conducted an additional ten semistructured interviews with long-time social media users in their early twenties about their media use practices and how they viewed the role of algorithms in those interactions.17

Affective Landscapes

99