If...Then Algorithmic Power and Politics

The answers were first coded for information about the kinds of situations and circumstances that provided the informants a “reason to react.” As feminist philosopher Sara Ahmed suggests of strange encounters, “To be afected by something is to evaluate that thing. Evaluations are expressed in how bodies turn towards things. To give value to things is to shape what is near us” (Ahmed, 2010: 31). This became clear in the process of coding the interviews and email transcripts as well. Thus, in a second iteration of coding, all transcripts were analyzed in terms of how they were either explicitly or implicitly evaluating and making sense of the algorithms, the extent to which their awareness of the algorithm afected their use of the platforms in question, what kinds of tactics and strategies they developed in response to the algorithm (if any), and the kinds of issues and concerns they voiced about the algorithms. While these transcripts would not stand up in a court of law, they should be seen as records of happenings that are “difuse yet palpable gatherings of force becoming sensed in scenes of the ordinary” (McCormack, 2015: 97). What is of importance, then, is not whether the stories that people tell are representative of how algorithms “really are” (if that is even possible). Instead, personal algorithm stories may atest to the ideas and perceptions people have about what algorithms are and how they function and, perhaps more importantly, their imaginings about what algorithms should be and the expectations that people form about them.

Living with Algorithms

People encounter algorithms in all kinds of situations: a recently purchased Christmas gift that appears on the shared family computer as an ad (ruining the surprise), weird recommendations, the almost-forgoten friend who suddenly reappears on the Facebook news feed after years of silence. Although the people I talked to for this chapter claimed to have varying degrees of knowledge about algorithms, a few themes emerged as to when and how algorithms came to mater—here, presented under a few subheadings paraphrased from participants’ responses: “People aren’t a math problem” describes the diferent kinds of confounding connections that algorithms seem to be generative of. Whether these connections come across as surprising, uncanny, funny, or bewildering, what unites them is that they do not feel inconsequential to the people afected by them. “Enter the popularity contest” is about the ways in which algorithms become arbiters of diference by “Distributing the sensible” (Ranciere,  2004), demarcating the regimes of visibility in terms of who or what becomes visible. Related to the first theme concerning how algorithms shape the ways in which identity is constituted and experienced, the notion of the popularity contest articulates the ways in which people become aware of and evaluate algorithms based on their capacity to filter content in diferent ways. Finally, “Clicking consciously” addresses how algorithms or, more precisely, beliefs about how algorithms work trigger particular responses and ways of being on social media platforms. People develop various strategies and tactics to make themselves “algorithmically recognizable” (Gillespie, 2016) or unrecognizable for that mater. What the data reveal is how being afected by algorithms is not simply a mater of passive observation or subtle awareness but of actionable consequences and movement, of reaction and creativity.

100

I F . . . T H E N

PEOPLE AREN’T A MATH PROBLEM

Meet Jessa. She and her boyfriend, both in their mid-20s, recently moved to New York City for work. The couple has been subleting an apartment for one and a half months, sleeping on an air matress that gradually seems to be deflating ever faster. They regularly talk about how much they look forward to sleeping on a real mattress. While Jessa trawls Craigslist for new apartments, her boyfriend looks at mattresses on Amazon. Then, one morning, while Jessa scrolls through her Facebook news feed, there it is—an ad for air matresses. Jessa is so perplexed that she tweets about it: “How on earth did the Facebook algorithm know she was sleeping on an air matress?”18 Although the connection may have been coincidental, the efects are not incidental. While Jessa understands that clicks and browser behavior are routinely tracked and used to tailor ads online, there is something almost inexplicably creepy about the ways in which algorithms seem to know you.19

Similarly, Kayla, a 23-year-old student from New York, is confounded by the ways in which the Facebook algorithm seems to have her pegged as “pregnant, single, broke and overweight.”20 Sure, she posted a status update a few days ago about being broke and single, she says, so the suggestions for dating sites and loans made sense. But how weird was it when the algorithm for suggested apps showed her dating sites next to apps for expecting mothers? Like Jessa, Kayla had theories about the logic behind the algorithmic profiling of her. As she speculates, “upon further thought, I think the pregnancy ones were because I had been looking at nursery decorations on another website for a baby shower gift.”21 While the inferences that algorithms make on the basis of user behavior might logically make sense to people, there is something socially and culturally “of” about puting dating sites next to advice for expecting mothers. Despite declaring that they do not really know what an algorithm is, many participants, when asked how they would define an algorithm, seem to make a distinction between a technical aspect and its socially perceived outcomes. Take Shannon, a career counselor in her 40s, who happened to blog about Taylor Swift. Now, the Facebook algorithm seems to think she is a “12 year-old with an Amex card.”22 Shannon is not sure “what the Facebook algorithm is—at least, the technical part of it.” She understands that “it picks up clues, perhaps using keywords” from her own activities and information she herself has provided “to try to target relevant content.” Although it makes technical sense, it may not feel right, Shannon explains. Asking her to provide an example of her experience of being algorithmically profiled, Shannon says:

Affective Landscapes

101

I noticed after I published my blog post on Facebook earlier in the week that, suddenly, the ads that Facebook “chose” for me were often about Taylor Swift concert tickets and products she endorses (like American Express), instead of the usual middle-aged woman assumptions like ads for wrinkle cream and fat loss.23

While Shannon thinks that the Taylor Swift incident is amusing, she often finds targeted ads to be “slightly ofensive as they make assumptions about me, which I don’t like to think are true.” Such is the work of “profiling machines” (Elmer, 2004) that seek to produce a sense of identity through detailed consumer profiles, which are geared toward anticipating future needs. Based on statistical inferences and inductive reasoning, profiling algorithms do “not necessarily have any rational grounds and can lead to irrational stereotyping and discrimination” (de Vries, 2010: 80). Still, the question of whether “correct” identifications are being constructed may be beside the point. As de Vries (2010) argues, misidentification is not simply a mismatch or something that should be considered inappropriate. Misidentifications may also give some leeway for thinking about how identity construction is experienced.

Experiencing algorithmic landscapes is as much about what the algorithm does in terms of drawing and making certain connections as it is about people’s personal engagements. To draw on Ingold (1993), a particular landscape owes its character to the experiences it afords to the ones that spend time there—to their observations and expectations. For Robin, the ways in which Amazon seems constantly to be confused by her gender is not just amusing and confusing but seems to make her transgender subject position simultaneously more real and unreal. As she explains:

I am transgender and just starting transition. This seems to confuse Amazon like crazy. It tries to make suggestions and gets things wrong quite often. Sometimes, it’s obvious but amusing: “I see you just bought some makeup but wouldn’t you like some power tools?” (I have never ordered power tools).

Robin has been thinking about “how much a company is willing to go out on a limb on things like this,” whether a company like Amazon would actually be willing to try and categorize people according to all possible demographic buckets, not just the ones that do not seem to endanger their profits:

If I were to examine my purchase history and come up with a narrative, I’d almost certainly read it as one transgender individual, rather than a heterosexual couple who happens to wear the same size clothing and shoes. But the negative reaction that comes from a false positive might outweigh any potential sales.24

102

IF . . . THEN

What seems to bother Robin is not so much the fact that the apparatus is unable to categorize her but the apparent heteronormative assumptions that seem to be reflected in the ways in which these systems work. As a person in transition, her queer subject position is reflected in the ways in which the profiling machines do not demarcate a clear, obvious space for her. A similar issue came up in my interview with Hayden, who is also transgender and uses social media primarily to read transition blogs on Tumblr and Instagram. He is not particularly impressed by the way  in which algorithms seem to only put “two and two together” when “the way  things work in society can be quite the opposite.” If this is how algorithms work by simplifying human thoughts and identity, Hayden thinks social media platforms are moving in the wrong direction. As he tellingly suggests, “people aren’t a math problem.”25

Algorithmic mismatches may be perceived as problematic when they are no longer perceived as “innocent” but turn into a more permanent confrontation that feels inescapable and more intrusive. For Lena, the Facebook algorithm is at odds with how she identifies in life. She is no longer the person she used to be. Yet, she feels constantly trapped by her past. She no longer hangs out with the same crowd. She does not share their interests. She has moved across the country from a small rural town in Texas to New York City, where she now goes to grad school. She identifies strongly as liberal and, of course, votes for the Democrats, she says. Yet, all the updates Lena sees on her news feed seem to stem from Republican voters and people who are only interested in celebrity gossip. This obvious disconnect between who she identifies as being and how Facebook seems to have her figured out annoys her. She is annoyed that her “own social network is so out of sync” with her interests and beliefs. Although she speculates that her inability to escape her past could be a result of adding most of her FB “friends” when she was atending high school in rural Texas and not adding as many since she moved to New York City, Lena thinks the algorithm “must be incorrectly assessing my social networking desires.”26 While “real” life allows the past to be the past, algorithmic systems make it dificult to “move on.” As Beer (2013) notes, the archive records and works to shape memory by defining what is relatable and retrievable. What happens when the world algorithms create is not in sync (to use Lena’s own words) with how people experience themselves in the present? To what extent do existing social networking profiles remain forever muddied by past lives and experiences?

Memory is a powerful thing. It lives in the flash of a second and the duration of a lifetime, it emerges in lifelike dreams and the drowsiness of mornings. Memories can be recalled at will or willfully withdrawn. A memory is an event that connects us to the past and makes it possible to project a path for the future. Memory is an encounter, a distraction, an opportunity, a daydream, or a denial. In the digital age, memories can be made, encouraged, and programmed to make people participate more, share more, engage more. For Albert, this is where machines fall short—“just maybe, there isn’t an algorithm capable of capturing human experience,” he tweets.27 When asking Albert to provide some context to his tweet, he explains:

Affective Landscapes

103

I was tweeting about Facebook’s recent “Year in Review” fiasco, where the algorithm they created was generating reviews that included some terrible memories—deaths of children, divorces, etc. My point in the tweet was that, while algorithms might (or might not) be a good tool for efectively serving up the right ad to the right user, they might not be the best way to create emotional content or connect on a human level.28

The incident that Albert cites, and which prompted him to speak out about the limits of algorithms, is the story of Eric Meyer and the “cruel algorithm” serving up a picture of his recently deceased daughter as part of Facebook’s “year in review” feature. For Eric Meyer and others experiencing these unwanted flashes of painful memories, algorithmically curated and repackaged content does not just materialize in helpful birthday reminders. It also lingers as insensitive reminders of the tragedies of life. Emerging at odd moments, these digitally mediated memories make you “raise your head in surprise or alarm at the uncanny sensation of a half-known influence” (Stewart, 2007: 60). More than anything, however, the incident made explicit for Albert “the most obvious thing about algorithms—they’re just machines.” What was obviously missing, says Albert, was the “human judgment that says, ‘You know, this guy probably doesn’t want to be reminded that his daughter died this year, so even though the post got tons of atention, I’ll leave it out’.” Indeed, algorithmic systems “judge individuals against a set of contrasting norms, without human discretion intervening or altering decisions” (Beer, 2013: 77).

Similarly, Ryan, a 21-year-old avid YouTube user, thinks that mechanical ways of trying to solve human problems and desires is where the algorithm hits its barrier. The problem with algorithms, Ryan suggests, is “that they are just based on trends and past behavior. It can’t predict for you in the future, it just assumes you are going to repeat everything you’ve done in the past.” Algorithms do not take into account that “people are unpredictable and like to try new things all the time,” he says.29 This is why Ryan loves people who put considerable efort into their YouTube channels—particularly, the semifamous YouTube celebrities he follows, who explicitly ask fans to contribute to their next videos. For Ryan, these YouTube celebrities are almost like “human algorithms,” making sure they give their audience what they want. Ryan thinks they are “more accurate than an automatic algorithm, because computer algorithms don’t really understand human preferences or traits like actual humans do.” This is precisely what Hayden seems to touch on when he claims that people are not a math problem: People are more complicated than an equation, more complex and unpredictable than what can be broken down into a few steps of  instructions for a computer. In fact, Hayden thinks algorithms are fundamentally not that smart. They were not able, for example, to parse the fact that “views” means beautiful scenery, while it also happens to be the name of the rapper Drake’s newest album. So, when Hayden uses the hashtag “views” for all his recent images of sunsets and landscape pictures on Instagram, he is not interested in following a bunch of Drake fans, which an algorithm apparently suggested. What became apparent in many of the participants’ responses, then, is not just how algorithms are capable of misjudging humans but how they might not even be able to judge humans at all.

104

I F . . . T H E N

E N T E R T H E P O P U L A R I T Y C O N T E S T

While algorithms might be bad at judging human complexity and emotional valences, they also seem to augment and amplify traditional social norms and hierarchies. For participants, this is most evident in terms of the perceived capacity of algorithms to filter and sort content and people according to their popularity and importance. Algorithms, of course, are not unique in this respect. The media have always operated according to the principle of privileging certain voices and personalities over others. As Altheide and Snow (1979) atested in their influential account of media logic, the mass media fundamentally work by assigning some expressions more weight and importance than others. The logic of popularity is highly evident in the operational logics of social media platforms, which often incorporate some form of personalized ranking mechanism to distinguish popular content from less important items (van Dijck & Poell, 2013). Technically, however, popularity and personalization are not easily squared (Amatriain,  2013). Whereas popularity is generally about recommending items according to a global notion of relevance, personalization is about satisfying members with varying tastes. As Amatriain writes in a paper explicating the models behind Netflix recommendations, the goal “becomes to find a personalized ranking function that is beter than item popularity” in order to cater to a variety of tastes (2013: 3). This goal does not just apply to Netflix but to prety much every platform that seeks to filter and rank its content to retain user atention in some way or another.

Indeed, the balancing act between quantity and quality, between high degrees of links/likes and individual practical engagement, constitutes a core challenge for platforms. When the balancing act fails—as it often does—users notice. Why, for example, did Netflix recommend a horror movie when Katy only ever watches romantic comedies, or why does OKCupid’s matching algorithm only seem to suggest outdoorsy types when all Robin wants to do is go to concerts and eat good food? Just because many people like horror movies or because the fit and healthy seem to be a thing on dating apps, this does not mean either is right for everyone. People particularly notice the popularity contest in terms of the ways in which social media platforms such as Facebook have a tendency to make the same people or the same type of content visible time and again. Lucas, a 25-year-old quality assurance engineer, says he spends a lot of his spare time on social media. Lately, he has noticed that there is absolutely no new content popping up on his Facebook feed.

Affective Landscapes

105

The same 5-6 stories were showing up at the top of my feed across a 4-hour period. I have a fairly large social graph on Facebook (1,000+ “friends”), so to see absolutely no new content show up agitated me, so I tweeted about my feelings on the algorithm.30

Lucas thinks that “Facebook’s filtering consistently degrades what he prefer in his online social experience.” Like Lena, who seems to be caught in a Republican loop on Facebook, Lucas feels that the platform is holding back important information. The problem for the participants lies not necessarily with filtering in and of itself but with the feeling of not knowing what they could have known. This is not simply the fear of missing out but of having “your own content” and relationships controlled or even censored by a platform, as some participants suggest.

Going back to Michael and Rachel’s stories, their frustration with Facebook stems in part from the feeling of having someone else decide on their behalf the conditions under which their contributions are either made visible or silenced. Indeed as Amber, another participant tells me, “I am feeling annoyed that the algorithm decides for me [. . .] it is so unpredictable and strange.”31 Nora, a 21-year-old Canadian public policy student, thinks there is “definitely something strange going on with the Facebook algorithm.”32 In a series of tweets, Nora speculates about the underlying mechanisms of Facebook, noting that some of her posts seem to get far more “likes” and comments than others without any apparent reason. The number of “likes,” “shares,” and “comments” fuel the popularity game supported by social media platforms in which algorithms feed of on the social disposition towards interaction. Nora worries that the popularity bias of social media algorithms potentially diminishes what people get to see. Specifically, she worries that the algorithmic bias toward “likes” and “shares” makes viral videos such as the “ice bucket challenge” much more prominent, hiding more important but less “liked” current events such as the racial conflicts in Ferguson. As Nora says, “I don’t like having an algorithm or editor or curator or whatever controlling too much of what I say—if they’re trying to go for more ‘trending topics,’ will my posts on ‘non-trending topics’ get shutered away?”33

The notion that algorithms contribute to making the popular even more popular is not just apparent in the participants’ understanding of algorithms. It is very much evident in the public discourses surrounding algorithms as well. Let us for a moment return to the #RIPTwiter incident mentioned at the beginning of this chapter and consider some of tweets that contained this hashtag: “Algorithms are a form of censorship. I look to Twiter for unfiltered news in real time.”34 “One of the great rewards of being an adult is deciding ON YOUR OWN who (and what) you should be interested in.”35 “Think about it, with an algorithm you might never see a tweet from specific people ever again. RIP Twiter.”36 “Replacing the chronological time line with an algorithm based feed. Turning Twiter into a popularity contest. No thanks.”37 In my interview with Ryan, he brings up similar issues and worries. He thinks it would be “a big mistake” if Twiter started using algorithms on a large scale, as “it will obviously favor media and links and videos and photos and stuf like that, and it will not favor text posts, which is a majority of what the average user is posting on that platform.” Hayden voiced a similar concern about the possibility that Instagram would change into an algorithmically filtered feed. For Hayden, the “Instagram algorithm” would merely make the “Instagram famous” even more famous since they would “automatically pop up on the top, and you won’t see your friends who might have fewer followers.”38 Like Hayden, the users who gathered around the hashtag #InstagramAlgorithm only two months after Twiter announced its algorithmically curated feed voiced their worries about the platform turning into a popularity contest—for example: “Meh about the Instagram algorithm change. Popularity contests were never my thing.”39 “Why is everything a popularity contest on social media? Why @snapchat is the healthiest platform for your self esteem. #InstagramAlgorithm.”40 “Why does @instagram think that this is a good idea? It’s ‘insta’gram not ‘popular’gram #RIPInstagram.”41 “The new @instagram algorithm is the end for small timers like me. I’m not winning any popularity contests against big names.”42 These tweets are worth quoting at length not because they are necessarily representative of the general sentiments of the hashtag publics that form around these specific events but because of the ways in which they show how algorithms are variously imagined and the expectations people have of them. What these sentiments show is how algorithms are imagined as control mechanisms and gatekeepers that work to enforce existing power structures.

106

I F . . . T H E N

However, algorithms are not always pictured as “bad” or controlling. The same person may talk about algorithms as problematic in the context of one platform but not seem to have a problem with a similar mechanism on another platform. Some algorithms are regarded as useful, while others are seen as obtrusive. For example, while Nora is worried about the Facebook algorithm controlling too much of what she sees, she says that, “with regards to Google, I think that is an algorithm I  would actually place my faith in.”43 When I probe her about her reasons for having  faith in Google but not in Facebook, she says she is struggling to find an exact  reason. Perhaps, it is because Google is search, she speculates, or because Facebook somehow

feels much more intrusive because the things they are intruding on is really personal things that really could mess someone up [. . .] Whereas Google, it is just a search engine. When you go to Google you already have an idea of what to find.44

Affective Landscapes

107

The ambiguity surrounding diferent kinds of algorithms became particularly evident in the semistructured interviews conducted over Skype. Algorithms are not inherently regarded as problematic but, rather, become more or less problematic in terms of how restrictive they feel to users. Several of the interviewees seemed to focus on the degree of agency and control they felt they had in terms of navigating diferent platforms. Ryan, for example, makes a distinction between helpful algorithms and controlling ones. As with Amber and Nora, Ryan also says “it feels like with Facebook, the algorithm is controlling me.” He categorizes Facebook’s algorithms as big while Twiter’s algorithm, despite the recent news, still feels “small” and “simple.” As he explains,

[T]his simple algorithm that Twiter has is that it shows you briefly that, while you’ve been away, these are the major things that has happened in the feed [. . .] I think the Twiter algorithm is very nice, because it is very simple, and it doesn’t control that very much, it just helps you a litle bit.45

When asking Ryan to expand on what he means by the notion of “helpful algorithms,” he brings up Netflix, his favorite platform. To him, the Netflix algorithm is like Facebook’s algorithm and Twiter’s algorithm: Like Facebook because it “shows what is popular right now,” and like Twiter in that it focuses on “your own customized feed without any other interference” but still tries to help you find new content “based on what I have watched.” Similarly, Melissa, a 22-year-old Canadian college student, thinks that Netflix gives you the “greatest control, because it will pull specifically from what you yourself are doing. They will not pull from what your friends are doing, or what you are looking at.”46 While all of the platforms try to find content that is relevant to users, there is a sense in which it makes a diference whether the user feels in control by seting the parameters themselves and the algorithm is merely helping along, as opposed to a situation in which the algorithm seems to determine relevancy more or less on its own or, at least, without any obvious explicit input from users. In many cases, personalization is seen as an asset if the platforms get it right but an intrusion if the popular seems to get in the way of the personal. Melissa links the demand for personalized content to her generation’s desire to feel special. This is how she puts it:

You are always looking for something that makes you special. And that is one reason why Starbucks’ business model was so successful. Because giving the cup with your name on it makes you feel like an individual in the sea of people. And I think the online platforms try to do a similar thing by ofering you things they think you want the most. And it makes you feel like it is almost like the media platforms are looking out for you. They try to give you the best experience you can have on that platform. And some succeed and some not so much. But I find the personalization much more gratifying for the user than having the same standard experience as everybody else.47

108

I F . . . T H E N

Just as the platforms themselves struggle to find the right balance between ranking based on a global notion of relevance and/or a more personal model, users also seem to oscillate between a desire for popular content on the one hand and personalized content on the other. More importantly, what these diferent responses show is that algorithms are not just factual but also normative in the sense that, both explicitly and implicitly, people have an understanding of how things ought to be, that is, how algorithms should perform.

C L IC K I N G C O N S C IO US LY E V E R Y DAY

Living in an algorithmically mediated environment does not simply mean that algorithms are overruling users. Wanting to collect likes and comments is not just a “narcissistic cry for help” (Agger, 2012: 45). For many users, it has become a necessary strategy to deploy if they want to impact the algorithm’s “willingness” to show their posts more prominently. Many of the avid social media users I talked to have in one way or another tried to play or experiment with the operational logics of the platforms they use—to try to be seen or not seen or otherwise to circumvent perceived control mechanisms. Several of the participants, for example, mention the experience of being caught in a temporary feedback loop. Ryan says of the YouTube recommendation algorithm’s tendency to reinforce a viewing patern, no mater whether it is correct or not:

Sometimes, I click on a video and think it is something and it is actually about something else, but now that I watched it, YouTube will show me a  bunch of videos connected to that in the recommended section, and I don’t really care about any of it.

In order to “correct” the algorithm, Ryan says he either stops clicking on anything for a while or just starts clicking on as many other more-related videos as he can. Similarly, Rachel, the woman who tweeted about Facebook destroying friendships, says that:

I find myself “clicking consciously” every day. This is going to sound crazy, but I remember the one time I accidentally clicked on a former high school classmate’s profile and cursed under my breath: “damn it, now I’m gonna see her stuf 24/7.”

“Clicking consciously” emerged as a key theme in people’s responses to how algorithms afect their use of platforms. Moreover, the notion of being trapped in a temporary bubble because of the way one clicks and likes seemed to constitute a rather common experience of being in algorithmic spaces. As the comments of Rachel and Ryan suggest, either you take charge of the clicks or the clicks will catch you up in a perpetual feedback loop.

Affective Landscapes

109