If...Then Algorithmic Power and Politics

Clicking consciously, however, is not just a defense strategy of sorts. It emerges quite prominently in a much more proactive sense as well. Take Kate, a former school teacher who now runs a Facebook page for parents in her neighborhood. For Kate, it is vital that her posts are seen and circulated as widely as possible. In order to secure “maximum reach” as she puts it, Kate takes great care to post “consciously,” which, in her case, means using multiple pictures instead of one, always trying to choose the right words, and the right time of the day. As a page owner, Kate says, “I have completely changed how I share information to make it work best for the algorithm.”48 In a similar vein, Nora tailors her updates in specific ways to fit the Facebook algorithm. Like Michael, Nora has been observing the working of the news feed for a while, taking notes of what seems to work and what does not. As she puts it, “If I post things and they receive no likes within the first 5 minutes, or very sparse likes (1–2 in the first minute), then they’ll drop of and not get many comments or likes at all.” Over the years, Nora has developed diferent strategies for making her “posts more frequently recognized” by the algorithm.49 These strategies include posting at a certain time (“usually around late evening on a weekday that’s not Friday”), structuring the post in specific ways, making sure that other people are not shown in her profile pictures (otherwise, they are “likely to get fewer likes”) and making sure to avoid or include certain keywords in her updates.

As Gillespie (2016) has usefully pointed out, adapting online behavior to social media platforms and their operational logics can be seen as a form of optimization, whereby content producers make their posts “algorithmically recognizable.” When users wait until a certain day of the week and for a particular time of day, use multiple pictures instead of one, carefully choose their words, and deliberately use positive sounding phrasing, they are not just strategically updating their social media profiles or hoping to be seen by others. Reminiscent of Gillespie’s (2014) argument about using hashtags for optimizing the algorithm, the personal algorithm stories shared as part of this study suggest that many of the participants are redesigning their expressions in order to be beter recognized and distributed by the algorithm. Like Kate, Ryan thinks the underlying logic of social media maters a lot. As he says, “how social media works really, really impacts what I post and how I post it.” For example, posting “life events” on Facebook or using certain hashtags on Twiter and Instagram was seen by some of the participants as a particularly efective means of increasing the probability the algorithm will make them visible.

Of course, people have always developed strategic eforts to be seen and recognized by information intermediaries. As Gillespie writes, “sending press releases to news organizations, staging events with the visual impact that television craves, or making spokespeople and soundbites available in ways convenient to journalists”

110

I F . . . T H E N

have long been media strategies for visibility (2016: 2). Users have in a sense always tried to reverse engineer or playfully engage with the seemingly impenetrable logics of media institutions and powerbrokers. However, in this day and age, the logic that people are trying to crack is not about human psychology but machine processing.

Part of this task involves figuring out how these platforms function. For example, posting “life events” on Facebook or using certain hashtags on Twiter and Instagram were seen by some of the participants as particularly efective means for increasing the probability of being made visible by the algorithm. According to Zoe, a 23-yearold self-employed Canadian designer, it is not just about knowing which hashtags to use in order to get noticed but also about knowing how diferent platforms treat and value them in the first place.50 For an up-and-coming designer like Zoe, who depends on reaching an audience (much like Michael), being unable to navigate certain platform conventions could be detrimental to business. The blogging platforms Tumblr and Instagram, for instance, have diferent hashtag conventions that are not just about knowing which are the most popular. Take a seemingly small technical detail such as whether or not a platform allows typing spaces in hashtags. Zoe says it took her a while to figure out that cross-posting images between the blogging platform Tumblr and Instagram failed to atract new followers or interest simply because the hashtags would not synchronize in the right way. While the hashtags worked on Tumblr, they turned into gibberish on Instagram. Zoe speculates that this may have cost her valuable opportunities to be noticed due to the algorithm’s unwillingness to feature her profile on other users’ “explore” section of Instagram. Using the right hashtags in the right way will increase your chances of becoming one of those suggested profiles, Zoe says.

Unlike a traditional public relations person or strategic communicator, the majority of people in this study did not tailor their messages for business reasons but often did so for what they thought their friends and family would like to see. Particularly striking was the notion of not wanting to burden people with unnecessary postings, that is, of not wanting to add to their friends’ information overload; and the social norms around accepted sharing behavior were frequently mentioned in this regard. For Ryan, catering to what he thinks his network is interested in seeing goes hand-in-hand with knowing how platforms work. Because he knows that a photo album on Facebook “is much more likely to appear on someone’s feed than a text status update,” he makes sure that the photos he posts are worthy of other people’s atention. He will only share a photo album with “thoughtful, funny captions” because he knows people are probably going to see it, given the algorithm’s perceived preference for photo albums. In a similar vein, Ryan only rarely “likes” things on social media platforms because as he says “people notice.” If you “like” too many things, the “likes” become meaningless. So, Ryan tries to “save those for actual important times.” This way, when people see he liked something, it actually means he really enjoyed it. “I save my likes for interesting articles or funny videos that are posted by major commercial pages that already have a million followers or likes,”

Affective Landscapes

111

Ryan explains, because as he speculates, accumulated likes on big, important pages are more likely to show up on other people’s news feeds.

Taking care of personal online communication in this optimized and strategic manner takes a lot of time and energy. Indeed, algorithms act as atention gatekeepers of sorts that users need to “crack” or play along with in order to increase their chance to get their message across. As the public discourse around the Twiter and Instagram algorithmic timeline suggests, algorithms are seen as powerbrokers that you either play with or play against. Part of the reason Ryan thinks it is a big mistake that Twiter and Instagram are “using algorithms” on a large scale is that it “will change what people post and change how people use the platform, and they won’t use it the same way as they used to.” For Ryan, the diference between a service such as Snapchat and Facebook is that the former “takes less thinking time.” As he elaborates:

But, with Facebook, there are so many things going on there, so I have to plan out strategically what I am saying, how I am posting it, when I post it, all that stuf. Just to make sure that everyone I want to see it actually sees that.51

The notion of strategically planning what to say is just another side of “clicking consciously,” a way of being oriented toward the operational logics of platforms. For Ryan, the problem with Twiter rolling out an algorithmic logic on a much greater scale than before is not so much about the algorithm in and of itself but the consequences it entails for practical engagement with the platform. So, if Twiter were to “start with an algorithm, you really have to start thinking about what you’re posting,” Ryan suggests.

While users often need play along with the algorithm if they want to get noticed, there are also the inverse eforts of not making oneself “algorithmically recognizable.” Indeed, as the personal algorithm stories atest, avoiding or undermining the algorithm requires people to pay atention to the workings of the platform. Louis, a respondent from the Philippines, says he is deeply fascinated by the Facebook algorithm; yet, he thinks of it as a trap. For Louis, users have various options to counteract the logic of algorithms. Indeed, as Hill argues, “resistance cannot merely be about opting out, but about participating in unpredictable ways” (2012: 121). As Louis sees it, “privacy online does not really exist. So, why not just confuse those who are actually looking at your intimate information? That way, it misleads them.”52 Others, too, reported engaging in activities of data obfuscation, whether explicitly or implicitly. Lena has been trying to “manipulate content” with which she interacts in order to “control the suggestions” Facebook gives her, while Jessa atempted to confuse the algorithm by liking contradictory things.

Whether or not people try to make themselves algorithmically recognizable or not, the responses suggest that playful learning on the part of users should be thought of as a mode of knowing that is geared toward an active exploration of unknown objects.53 In many ways, the various ways in which users orient themselves toward the platform’s operational logic is not that diferent from the ways in which I tried to “reverse engineer” Facebook’s news feed algorithm in the previous chapter. While we might not know the exact workings of algorithms, their consequences are often both observable and felt. According to Eleanor Gibson, “the active exploration of objects, leading to observable consequences and more specialized exploratory activities, has important results for learning about what an object afords, what can be done with it, its functional possibilities and uses” (1988: 24). Not only does the playful exploration of algorithms that some (albeit quite reflexive and knowledgeable) users engage in result in users learning more about what systems aford and what can be done with them, these activities also have the power to shape the very algorithms that perpetuate these responses in the first place.

112

IF . . . THEN

THE ALGORITHMIC IMAGINARY

Algorithms seize the social imaginary through the various afective encounters of which they are part. As the many diferent personal algorithm stories analyzed in this chapter atest, algorithms are generative of diferent experiences, moods, and sensations. The diferent scenes and situations can be understood as forming part of what might be called an algorithmic imaginary—ways of thinking about what algorithms are, what they should be, how they function, and what these imaginations, in turn, make possible. While there is no way of knowing for sure how algorithms work, the personal algorithm stories illuminate how knowing algorithms might involve other kinds of registers than code. Indeed, “stories help to make sense of events, set forth truth claims, define problems, and, establish solutions and strategies” (Ainsworth & Hardy, 2012: 1696). Stories account for events and experiences in ways that help actors make sense of what is going on in specific, situated encounters. Personal algorithm stories illuminate what Charles Taylor (2004) has termed “the social imaginary,” understood as the way in which people imagine their social reality. The imaginary, then, is to be understood in a generative and productive sense as something that enables the identification and engagement with one’s lived presence and socio-material surroundings.

However, my use of the imaginary is not about the social imaginary per se. The algorithmic imaginary is not necessarily a form of “common understanding that makes possible common practices” (Taylor, 2004: 23), nor is it necessarily about the ways in which algorithms, for instance, can be generative of an “imagined community” (Anderson, 1983) or a “calculated public” (Gillespie, 2014). While these notions could certainly be part of an “algorithmic imaginary,” I want to use the term here to suggest something slightly diferent but no less related. I have no reason or desire to claim, for example, that people’s personal algorithm stories are part of a more commonly held cultural belief about algorithms in general. While they might be shared or overlap in important ways and be part of larger debates and discourses about algorithms in society, they are also imaginaries that emerge out of individual and habitual practices of being in algorithmically mediated spaces. In other parts of the book, I have discussed the notion of programmed sociality as the ways in which software and algorithms support and shape sociality and lived experience that are specific to the architecture and material substrate of the platforms in question. In this sense, algorithms may, indeed, have the power to generate forms of being together that are reminiscent of Benedict Anderson’s notion of imagined communities. As Anderson famously declared, a nation “is an imagined political community—and imagined as both inherently limited and sovereign” (Anderson, 1983: 15). Importantly, as Strauss points out, Anderson emphasized the role of media, particularly newspapers and books, in “creating a reader’s sense of being part of a larger community of other assumed readers” (Strauss, 2006: 329). The algorithmic imaginary that I want to address in this chapter, however, is not a public assembled by the algorithm but the algorithm assembled, in part, by the public. In other words, the algorithmic imaginary emerges in the public’s beliefs, experiences, and expectations of what an algorithm is and should be.54

Affective Landscapes

113

Using the notion of the imaginary to describe what the encounters between people and algorithms generate is not to suggest that people’s experiences of algorithms are somehow illusionary. Quite the opposite, they are “real.”55 Algorithms are  not just abstract computational processes; they also have the power to enact material realities by shaping social life to various degrees (Beer,  2013; Kitchin & Dodge, 2011). When Rachel finds herself “clicking consciously everyday” to influence what will subsequently show up in her news feed, the algorithm is not merely an abstract “unreal” thing that she thinks about but something that influences the ways in which she uses Facebook. Similarly, Lucas says his awareness of the Facebook algorithm has afected not just how he posts but also how he responds to others. As Lucas explains:

I know that, if a friend of mine posts something they are passionate about, I will go out of my way to “like” and “comment” because I know that will programmatically “support” them and hopefully put them into more people’s feeds because EdgeRank will give them more points for my participation.56

Lucas’ willingness to go out of his way to like his friends’ posts and enhance their visibility echoes some of the findings in recent work on social media surveillance. As Trotier and Lyon (2012) have shown, Facebook users engage in “collaborative identity construction,” augmenting each other’s visibility through practices of tagging, commenting and liking.

Users’ perceptions of what the algorithm is and how it works shape their orientation toward it. Several of the participants reported having changed their information-sharing behavior “to make it work best for the algorithm,” as Kate put it. The practical engagement with social media platforms as lived-environments implies developing tacit knowledge about the underlying logic of the system. While most technologies are designed in such a way that people do not have to know exactly how it works (Hardin, 2003), people tend to construct “mental models” and theories about its workings as a way of navigating and interacting with the world (Orlikowski & Gash, 1994). Despite explicitly pointing out that they did not know what an algorithm is, most participants had more or less elaborate theories about what algorithms are and ought to be. In the case of Facebook, for example, Kayla says she has “no idea what the algorithm is” but suspects it works in response to all the data tracked by the platform. Similarly, Michael has “no clue what the actual algorithm is” but still had a clear idea about how best to construct a status update in order to increase the likelihood of geting his posts widely distributed.

114

I F . . . T H E N

Far from naming an illusory relation, the algorithmic imaginary is a powerful identification that needs to be understood as productive. The sites and situations through which people encounter and experience algorithms arguably shape ways of thinking, talking, and feeling about them. While seeing an ad for wrinkle cream may not be surprising when you are 45 or an ad for a dating site when you have declared yourself as “single” on Facebook, these connections may not feel incidental. Algorithms create a “cybernetic relationship to identification” by constructing “categories of identity” (Cheney-Lippold, 2011: 168, 172). These statistically derived paterns of cybernetic categorization, however, may be in conflict with how users feel about and see themselves. Some participants, such as Shannon and Kayla, feel uncomfortable with the ways in which they are apparently being categorized, while others, such as Lena and Larry, feel distanced and even angry at the algorithm for “thinking” they would be the kinds of persons who would actually be interested in the content they are served. While it might be dificult to escape the digitally constructed categories of identity, Papacharissi suggests that afect may extend beyond “just emotions and feelings to describe driving forces that are suggestive of tendencies to act in a variety of ways” (2015: 12). Whether or not the algorithm makes a correct inference does not necessarily mater. For example, a child who is mistakenly addressed as an adult may like the fact he or she is being taken “seriously” and start to behave in a more adult-like manner (De Vries, 2010: 78). Similarly, it seemed that Lena’s anger at being “wrongfully” identified and associated with her former classmates provoked her to update even more on the Democrats to counter the algorithm’s insistence on showing “Republican” updates.

The algorithmic imaginary does not merely describe the mental models that people construct about algorithms but also the productive and afective power that these imaginings have. Driven by machine learning, the algorithms that drive social media platforms evolve and change as a result of being exposed to an ever-increasing set of data. As Emilee Rader and Rebecca Gray (2015) point out, the feedbackloop characteristics of these systems make user beliefs an important component in  shaping overall system behavior. When users “click consciously,” disrupt their “liking” practices, comment more frequently on some of their friends’ posts to support their visibility, only post on weekday nights, or emphasize positively charged words, these movements or reactions are not just afected by the algorithm (or, rather, by people’s perceptions of the algorithm); the practices also have the ability to afect the very algorithms that helped generate these responses in the first place. The social power of algorithms—particularly, in the context of machine learning— stems from the recursive relations between people and algorithms.

Affective Landscapes

115

Conclusion

Everyday encounters with algorithms constitute an important site for analyzing the power and politics of algorithms in a way that does not necessarily assume a topdown or macroscopic view of power. Instead, in this chapter I have focused on the micropolitics of powers that emerge in the lived afects of people as they encounter algorithmic forces in their online surroundings. Micropolitics becomes a useful concept for thinking about what these encounters generate. As events, these situated encounters “might have powerful consequences through the way that [they transform] relations of enablement and constraint” (Bissell,  2016: 397). It is not just that algorithms may constrain or enable people’s capacities to do things—for example, to be seen and heard online, to permit feelings to be felt or to prompt reactions and evaluations about one’s lived environments. The barely perceived transitions that occur in situated encounters are by no means unidirectional; people certainly constrain or enable the algorithm’s capacity to do things as well.

The algorithmic body—particularly, the one driven by machine learning—is characterized by its capacity to change as a result of an encounter. Just as people learn to navigate their whereabouts, algorithms learn and adapt to their surroundings as well. While we might not be able to ask the algorithm about its capacity to afect and be afected, its outcomes can certainly be read as traces of how it evaluates its encounters with the social world. It is important to understand the afective  encounters that people have with algorithms not as isolated events that are somehow outside the algorithm but as part of the same mode of existence. When people experience the Facebook algorithm as controlling, it is not that the actual news feed algorithm is powerful in the restrictive sense. In its perceived restrictive functionality, the algorithm also becomes generative of various evaluations, sensations and moods.

If power is to be understood relationally, then we must think about the encounters between algorithms and people as both adversarial and supportive. Power, Foucault reminds us, “traverses and produces things, it induces pleasure, forms of knowledge, produces discourse. It needs to be considered as a productive network, which runs through the whole social body, much more than as a negative instance whose function is repression” (1980: 119). More than anything, then, power is a transformative capacity that always already implies forms of resistance (Foucault, 1978: 95). Resistance is not so much about the refusal to be governed by algorithms or to use a particular platform but about the practical engagement with an ascending force. The personal algorithm stories examined as part of this chapter are precisely indicative of such engagement. As Williams argued, people’s afective encounters with algorithms should not be read as “formally held or systematic beliefs” about algorithms but, rather, as generative of “meanings and values as they are actively lived and felt” (Williams, 1977: 23). Still at the cusp of semantic availability, these meanings, perceptions, and expectations cannot be adequately described as a social imaginary if we take that to mean a culture’s ethos or shared sense of understanding. As individual and idiosyncratic, these stories point to the more subtle conditions, ways of being and imaginaries that surround algorithms.

116

I F . . . T H E N

In this chapter, I have suggested viewing these “ordinary afects,” to use Stewart’s term, as forming part of an algorithmic imaginary, understood as ways of thinking about what algorithms are and, perhaps more importantly, what they should be. While encounters between algorithms and people are generative of stories and certain beliefs about algorithms, these imaginations have a transformative capacity, too. In an important sense, then, it is not the algorithm, understood as coded instructions that enables or constrains actions but, rather, the perceptions and imaginations that emerge in situated encounters between people and algorithms—indeed, the algorithmic imaginary. If we want to understand the social power of algorithms, then, critiquing their workings is not enough. While algorithms certainly do things to people, people also do things to algorithms.

Affective Landscapes

117

Programming the News

When Algorithms Come to Matter

“People are beginning to believe what they want.” Such was the explanation for Macquarie Dictionary to name “fake news” the word of the year for 2016. Similarly, the Oxford English Dictionary named “post-truth” word of the year 2016, suggesting: “Objective facts are less influential in shaping public opinion than appeals to emotion and personal belief.” Questions about the status of truth, news, and facts have acquired renewed prominence in the wake of the complex political developments of 2016, particularly with regard to the possible consequences of a misinformed public in the aftermath of the US presidential election. But the arrival of “fake news” and “post-truth” into the popular vocabulary do not merely describe the notion that beliefs and emotions trump facts and truth.1 These concepts are symptomatic of a time when algorithms have emerged as a particularly prominent mater of concern in public discourse. When debates about fake news and whether we live in a post-truth era emerged, they were already part of a much longer and ongoing discussion about the powerful role that social media platforms have in circulating news and information.

In 2016 alone, Facebook was involved in at least three major public controversies surrounding its alleged editorial responsibility. I briefly discussed the first controversy, which centered on the human role of editing what was believed to be a fully automated process for determining trending news, in chapter  3. When the news about Facebook employing human editors to curate its “Trending Topic” section first broke in May 2016, the public response was one of outrage. After all, the algorithm was believed to be neutral, free from human intervention and subjective bias. Through these controversies, among other things, the world learned that algorithms do not function apart from humans. Part of the controversy centered around a belief that humans were interfering with the supposed neutrality of the algorithm. But the mater was made worse by the way it turned out that the humans involved held leftleaning political inclinations. The story became one where politically biased humans deliberately edited out and thereby prevented conservative news from making it into the trending topic section.

6

118

After much public pressure, Facebook’s responded in August 2016 by diminishing the role of human editors, making “the product more automated,” in that it no longer required people to write descriptions for trending topics (Facebook, 2016). In an interview, Facebook’s manager in charge of news feed, Adam Mosseri, claimed the trending topics section “is a  ‘beter’ product without human curators” (Chaykowski, 2016). However, mere days after removing the human curators, socalled fake news started to appear in the trending section. This included a factually incorrect headline involving Fox News personality Megyn Kelly (Ohlheiser, 2016). Laying of 26 employees (19 curators and 7 copyeditors) and puting the robots in charge apparently did not make the trending topic section more accurate or beter. Why Facebook would make any such promise in the first place, given that it repeatedly denies being a media company, is, of course, another question to raise.

At nearly the same time that Facebook fired the journalism graduates editing the trending topic feature and the first accusation of fake news emerged, Facebook became embroiled in another major controversy surrounding its algorithms. In early September 2016, the social media platform came under much public scrutiny for having deleted the famous Pulitzer Prize–winning photograph “The Terror of War.” Nick Ut’s photograph of a naked 9-year-old girl running away in agony after a napalm atack during the Vietnam War was deleted from Facebook for violating the company’s terms of service. Although the image had been deleted from the platform many times before, this time it did not go unnoticed. On September 8, the editor-in-chief of Norway’s largest newspaper Aftenposten, Espen Egil Hansen, used the front page to publish an open leter to Mark Zuckerberg, where he expressed a worry about Facebook’s inability to “distinguish between child pornography and a famous war photograph.” “Dear Mark Zuckerberg” the leter began, “[editorial responsibility] should not be undermined by algorithms encoded in your ofice in California.” Like most reporting on the topic of algorithms, Hansen’s leter was quick to construct a dichotomy where the automatic and mechanical contrasted with human decision making. In addition to highlighting a strong concern over Facebook’s growing power as an information intermediary, the incident also says something about the ways in which the algorithm is strategically called out and framed as a potential threat for a democratic society. In what is becoming an all too familiar tale, the algorithm is the villain, and the human—in this case, the press—is portrayed as the hero.

The above controversies call our atention to a particular messy and complex part of the current algorithmic media age. It is not simply that the Facebook algorithm does or does not do certain things—say, censor a famous picture or favor click-bait journalism and the spreading of fake news. What the trending topics controversy, censorship allegations, and fake news story suggest is the extent to which various actors are always diferently involved in seting the stakes for when the algorithm is called out as particularly powerful in specific situations. In chapter 3 the argument was made that algorithms only mater sometimes, and that their power and politics have to do with the circumstances of making them mater in specific ways and for specific purposes. In this chapter I continue this discussion by examining how algorithms are “problematized” in the current media landscape. In writing about the history of sexuality, Foucault described how sexual behavior was “problematized, becoming an object of concern, an element for reflection, and a material for stylization” (2012: 23–24). Similarly, this chapter takes the news industry as a case in point to address the circumstances under which algorithms are rendered more or less problematic, and how their power and politics can be understood by examining the conditions under which algorithms become objects of concern.

Programming the News

119