If...Then Algorithmic Power and Politics

Chapter 1

1. Facebook’s tagline as of June 2017.

1. Netflix’s tagline as of June 2017.

Chapter 2

1. htp://cs-exhibitions.uni-klu.ac.at/index.php?id=193

1. According to Hromkovič, Leibniz was the first philosopher who conceptualized mathematics as an “instrument to automatize the intellectual work of humans. One expresses a part of reality as a mathematical model, and model, and then one calculates using arithmetic” (2015: 274).

1. Not all algorithms depend on “if. . . then” statements. Other control structures include “while” or “for” loops.

1. Elegance, as first proposed by Donald Knuth in Literate Programming (1984), can be measured by four criteria: the leanness of the code, the clarity with which the problem is defined, the sparseness of the use of resources such as time and processor cycles, and implementation in the most suitable language on the most suitable system for its execution.

1. Some of the most basic data structures include the array, the record, the set, and the sequence, of which array is probably the most widely used. More complicated structures include: lists, rings, trees, and graphs (Wirth, 1985: 13).

1. For more on the history, cultures, and technical aspects of databases, see Bucher (2016), Codd (1970), Driscoll (2012), Dourish (2014), Manovich (1999), and Wade & Chamberlin (2012).

1. In a program writen in the programming language C, the diferent steps can usefully be illustrated by the program’s file-naming convention. The source code file ends in “.c,” the object code ends in “.obj,” and the executable files end in “.exe.”

1. While it is beyond the scope of this chapter to lay out how the extremely complex fields of cognitive science or children’s cognitive development understand human learning, it may sufice to say that “many machine-learning researchers take inspiration from these fields” (Domingos, 2015: 204).

1. A fourth category of machine learning that is usually listed is reinforcement learning, which is about learning in situ as the system interacts with a dynamic environment, for example, in the case of self-driving cars.

1. For a good overview of the technical details on how data mining and machine learning work, see Barocas & Selbst (2016). They provide an excellent description of the diferent processes and steps in data mining, including defining the target variable, labeling and collecting the training data, selecting features, and making decisions on the basis of the resulting model.

Notes

161

1. Probabilistic models or classifiers called Naive Bayes usually power spam filters. For more on the history and working of Naive Bayes classifiers, see Domingos (2015: 151–153) or Rieder (2017). According to Domingos, it is supposedly the most used learner at Google.

1. See Barocas & Selbst (2016: 9) on the creditworthiness.

1. Rules of thumb and systematic searching through the performance of diferent parameter values on subsets of the data (cross-validation) which are also often used.

1. Numbers based on consulting Facebook’s newsroom on December 31, 2014: htp://newsroom.fb.com/products.

1. Neural networks in machine learning do not work exactly like the brain but are simply inspired by the ways in which the brain is thought to learn. The first formal model of a neuron was proposed by Warren McCulloch and Walter Pits in 1943. It was not until 1957, when Frank Rosenblat pioneered neural nets with his conception of the perceptron, that neurons were thought to be able to learn to recognize simple paterns in images. For more on the early history of neural nets, see Minsky & Papert (1969) and M. Olazaran (1996). For a literature overview of the field of neural networks and deep learning, see, for example, J. Schmidhuber (2015).

1. For more on how Google image recognition works using deep learning and neural networks, see htp://googleresearch.blogspot.dk/2015/06/inceptionism-going-deeper-into-neural.html.

1. Bayes’ theorem sits at the heart of statistics and machine learning, though what a Bayesian means by probabilities may difer from the ways in which statisticians use the concept. As Domingos (2015) explains, statisticians usually follow a much stricter “frequentist” interpretation of probability.

1. For the most comprehensive bibliography on critical algorithm studies, put together by Tarleton Gillespie and Nick Seaver, see: htps://socialmediacollective.org/reading-lists/ critical-algorithm-studies (last accessed, May 2, 2016).

1. The need for coding skills is one of the most discussed topics in software studies. While it certainly helps to know the principles of writing code, some syntax of a programming language, and how the computer works, a sociologist of algorithms needs to know how to code no more than a television scholar needs to know mechanical engineering.

1. On the notion of “algorithmic culture,” see also Galloway (2006b); Kushner (2013).

1. For an overview of Foucault’s concepts of power, see Lemke, 2012; Faubion, 1994; Rabinow, 1994. Foucault’s public lectures (held at the Collége de France as part of his appointment as chair of the “history of systems of thought” between 1970 and his death in 1984) provide a comprehensive survey of his various conceptualizations of power, including the notions of disciplinary power (Foucault, 2015), biopower (Foucault, 2007, 2008), and pastoral power (Foucault, 2007). Foucault’s text on The Subject and Power (1982) provides a classic statement of how he understood the notion of power and how it can be studied.

1. Foucault’s ideas of omnipresent power, however, have a much longer historical philosophical tradition going back, at least, to the philosopher Baruch Spinoza (1632–1677) who is known for his metaphysics of substance monism, the view that everything is a “mode” of one ontological substance (God or Nature). Monism is typically contrasted with Cartesian dualism or the view that the world is made of two fundamental categories of things or principles. For Spinoza, modes (or subsets of the substance) are always in the process of entering into relations with other modes (i.e., humans, animals, things, etc.). As Jane Bennet puts it, Spinozan nature is “a place wherein bodies strive to enhance their power of activity by forging alliances with other bodies in their vicinity” (2004: 353). Foucault’s way of understanding power as omnipresent and relational bears resemblance to Spinoza’s monism and his concept of immanent causality, see Juniper & Jose, 2008; Deleuze, 1988.

1. Gatekeeping is the idea that information flows from senders to receivers through various “gates.” Originally conceived by the social psychologist Kurt Lewin in 1947, the idea of gatekeeping became crucial to journalism studies and information science as a way to explain the process of editing information. In the domain of communication, the gatekeeping concept is usually atributed to Lewin’s research assistant David Manning White, who in 1950 studied how “Mr. Gates,” a wire editor in a small US newspaper, based his selection of news on some highly subjective criteria (Thurman, 2015). Ever since, the notion of gatekeeping has been used to designate the editorial decision-making process in journalism and media.

162

NOTES TO PAGES 26–34

1. Within scholarship on computational systems, previous research emphasizing a “values in design” perspective includes discussions about the politics of search engines (Introna & Nissenbaum, 2000; Zimmer, 2008), cookies (Elmer, 2004), knowledge infrastructures (Knobel & Bowker, 2011; Bowker & Star, 2000), and game design (Flanagan & Nissenbaum, 2014). “Values in design” approaches typically hold that technology raises political concerns not just in the way it functions but also because the ways in which it works often seem to be at odds with people’s expectations (Introna & Nissenbaum, 2000: 178).

1. For a good discussion of the concept of government and governmentality, see Michel Senellart’s overview in Foucault, 2007: 499–507.

1. For his conceptualization of government in his College de France lectures, Foucault draws heavily on the anti-Machiavellian writer Guillaume de La Perrière and his work Le Miroir politique, contenant diverses manières de gouverner (1555). Despite characterizing La Perrière’s writing as boring compared to that of Machiavelli, Foucault sees great merit in La Perrière’s way of conceptualizing government as being concerned with the relationship between men and things (with an emphasis on relationship).

1. As Bröckling et al. point out, such technical means may, for instance, include social engineering strategies embedded in various machines, medial networks, and recording and visualization systems (2011: 12).

1. Foucault’s thinking on subjectivation and power relations changed somewhat throughout his career. Whereas his earlier work conceptualized subjects as “docile bodies” shaped by disciplinary processes of power, his later works focused on what he termed “technologies of self,” whereby subjects are seen as much more self-suficient in terms of shaping their conditions of living.

1. David Golumbia (2009) makes a similar argument that computation can be understood as an “ideology that informs our thinking not just about computers, but about economic and social trends.”

1. htp://www.bbc.co.uk/programmes/b0523m9r

1. As Law & Singleton usefully break the notion of the multiple down: “it is most unlikely that whatever we are looking at is one thing at all. So, for instance, a ‘natural’ reality such as footand-mouth disease is not just seen diferently by vets, virologists and epidemiologists (though indeed it is). It is actually a diferent thing in veterinary, virological and epidemiological practice. It is made or done to be diferent in these diferent practices. It is a multiple reality” (2014: 384).

Chapter 3

1. See Wendy H.K. Chun (2011) for more on how the Enlightenment vision connects to software.

1. Secrecy has long been a topic of concern for disciplines such as sociology (most notably, in the works of Simmel, 1906), anthropology (Bellman, 1984), and philosophy (Derrida & Ferraris, 2001). On the relations between secrecy and transparency, see Birchall (2011).

1. The unknown here is simply understood as lack of knowledge or information. For more on how the unknown can be conceptualized further, see, for example, the extant literature on the notion of ignorance (Gross, 2007; McGoey, 2012; Roberts, 2012; Smithson, 2012). Scholars distinguish among diferent types of ignorance (or lack of knowledge). For example, between known unknowns and unknown unknowns, where the former “denotes knowledge of what is known about the limits of knowledge; there are certain things that we know that we do not know” while the later “refers to a total lack of knowledge” (Roberts, 2012: 217).

NOTES TO PAGES 35–43

163

1. For one of the first discussions on black boxes within science and technology studies, see Callon and Latour (1981), where they conceptualize power as the ability to sit on top of black boxes. For a critical account of “opening up the black box,” see Winner (1993). A quick Google Scholar search revealed that over 18,000 articles have been published containing the phrase “opening the black box,” including the black boxes of finance, nanotechnology, soil microbial diversity, aid efectiveness, and media efects.

1. Galison exemplifies his notion of antiepistemology by way of one of the most mythical examples of trade secrets, the Coca-Cola formula. As Galison writes, quoting security personal Quist: “the recipe for Coca-Cola Classic has been kept a secret for over one hundred years. It is said that only two Coca-Cola company executives know that recipe [which] is in a safe deposit box in Atlanta, which may be opened only by vote of the company’s board of directors . . . We probably would not know if a national security secret was as well-kept as the secret of Coca-Cola” (2004: 239).

1. Legal debates over algorithmic regulation have largely centered on free speech jurisprudence under the First Amendment. Pasquale and Bracha (2008), for example, argue not only that search engines need to be regulated but, more specifically, their ability to structure search results needs to be regulated, and that the First Amendment does not encompass search engine results. Others disagree, seeing search engines as fully protected by the First Amendment (Volokh & Falk, 2011).

1. Bataille’s notion of “non-savoir” has been translated as both unknowing (Bataille & Michelson, 1986) and nonknowledge (Bataille, 2004).

1. At Twiter, they describe their mode of platform development in terms of a “culture of experimentation”: htps://blog.twiter.com/2015/the-what-and-why-of-product-experimentationat-twiter-0.

1. There are many resources providing an overview of what a relational materialism entails, its key thinkers and core tenets—for example, Bennet, Cheah, Orlie, Grosz, Coole, & Frost (2010) for an overview on new materialism; Thrift (2007) for an introduction to many of the core issues at stake in a relational ontology; and Latour (2005) for an introduction to actornetwork theory. Scholarly debates about these issues also proliferate in special issues of academic journals and within specific academic disciplines (i.e., information systems, education, human geography) that seem to have adapted to a particular strand of relational materialism (i.e., agential realism, new materialism), often with regard to specific concepts (i.e., sociomateriality, material-discursive, assemblage) and thinkers (i.e., Barad, Whitehead, Latour).

1. As Lemke (2015) notes, the term “more-than-human” was coined by Braun and Whatmore (2010: xx). Braun and Whatmore use it as the preferred term over “posthuman.” These approaches also emphasize terms such as practice, performance, movement, process, entanglement, relation, materiality and the nonhuman.

1. Prehension is a key concept in Whitehead’s metaphysics and refers to the elements, including energies, emotion, purpose, causation, and valuation, that combine (or what Whitehead refers to as concrescence) to produce actual entities (Michael, 2004).

1. For Whitehead, actual entities or occasions become concrete through a process he calls concrescence, that is, the “production of novel togetherness” (1978: 21). This insistence of the becoming or thickening of an actual entity from a multiplicity of possibilities (or potentiality) has had an enormous influence on Deleuze’s philosophy of the virtual.

1. Lucy Suchman (2007: 283–286) addresses the methodological necessity of making arbitrary analytical cuts in the network. The boundaries drawn and cuts made are never just naturally occurring but constructed for analytical and political purposes.

1. Derived from Austin’s (1975) account of the performative function of language as having the power to enact that which it merely seems to state, the notion of performativity has been widely used in critical theory (Butler, 2011), STS (Callon, 1998; Pickering, 1995), and the sociology of finance (MacKenzie, 2008) as a way to depict the world not as an already existing state of afairs but, rather, as a doing—“an incessant and repeated action of some sort” (Butler, 1990: 112). The notion of performativity has also played a central role in critical discussions of software, code and algorithms (see Galloway, 2006a; Hayles, 2005; Introna, 2016; Mackenzie, 2005; Mackenzie & Vurdubakis, 2011).

164

NOTES TO PAGES 43–50

1. Some scholars would argue that the sociotechnical and sociomateriality are not the same but denote diferent things. The concept of sociomateriality has been particularly important to research in information systems and organizational studies. For a discussion on how what these terms mean and how they might be distinguished, see Leonardi, Nardi & Kallinikos (2013).

1. As with the terms sociotechnical and sociomaterial, the concepts of network, assemblage, and hybrids have their own intellectual history that do not necessarily overlap entirely. While they all denote a composite term that implies some form of co-constitution between humans and the more than human, the term hybrid relates to actor-network theory and the work of Bruno Latour in particular, whereas assemblage is more of a Deleuzian concept (although applied and used by thinkers afiliated with actor-network theory such as Michael Callon). For an understanding of hybrids, see Latour’s example of the citizen-gun or gun-citizen that he describes on several occasions but, most explicitly, in Latour (1994). Whereas hybrids refer to the relations between heterogeneous entities, the notion of assemblage also points to the process of assembling, not merely to the existence of a composite entity. In Deleuze and Guatari’s account, an assemblage is not just a thing but also an ongoing organizing of multiplicities (see Deleuze & Guatari, 1987). Assemblage should not just be understood as a gathering of subjectivities and technical objects but, rather, in the sense of its original French meaning of agencement—a process of assembling rather than a static arrangement (see Packer & Wiley, 2013; Callon, 2007). For a good overview of the similarities and diferences between actor-network theory and the concept of assemblages, see Müller (2015).

1. See DeLanda (2006) for a discussion on these terms and how they difer. The main ontological diference can be found between those who focus on “relations of interiority” and those who maintain a focus on “relations of exteriority.” The former notion claims that nothing exists outside of relations, whereas the later contends that the parts of an assemblage can have intrinsic qualities outside of its associations (Müller, 2015: 31). For Karen Barad, an influential philosopher commited to a relational ontology (or agential realism as she calls it) “relata do not preexist relations” (2003:815). An agential realist account emphasizes entanglements not as intertwining of separate entities, but rather as relata-within-phenomena that emerge through specific intra-actions (ibid.). As Barad puts it, “why do we think that the existence of relations requires relata?” (2003: 812). If Barad’s ontological commitments constitute one side of the spectrum, most theorists commited to a relational ontology are either less explicit about their metaphysical claims or formulate a theory that atempts to account for the relative autonomy of entities outside of their relations. In contrast to Barad, Deleuze explicitly states that “relations are external to their terms” (Deleuze and Parnet, 2007: 55), meaning that “a relation may change without the terms changing” (See Delanda, 2006: 11).

1. The US Senate Commerce Commitee sent a leter to Facebook CEO Mark Zuckerburg, looking for answers on its trending topics section. In the leter, Senator John Thune, chairman of the commitee, accused Facebook of presenting the feature as the result of an objective algorithm while, in reality, human involvement made it much more “subjective.”

1. Here, I am borrowing from Bloomfield et al. (2010: 420), who revisit the notion of afordances and urge researchers to consider the question of when an afordance is.

1. Lucy Suchmann makes a similar argument in her book Human-Machine Reconfiguration in which she turns away from questions of “whether humans and machines are the same or different to how and when the categories of human or machine become relevant, how relations of sameness or diference between them are enacted on particular occasions, and with what discursive and material consequences” (2007: 2).

NOTES TO PAGES 50–55

165

1. As the continuing saga of Facebook’s trending topic shows, the figuration of the humanmachine relationship is only ever temporarily stabilized. Only two days after Facebook announced that it would replace the human writers with robots, the algorithm started to highlight fake news as part of the trending topic. Without human oversight, the robots were not able to detect that a story featuring the Fox News anchor Megyn Kelly wasn’t true. This, of course, caused much outcry again, proving to critics that humans were needed after all. I am including this in a note because of its obvious neverending character. At the time of writing, Facebook might have deployed more humans again to counter the robots gone awry, while in two months’ time it might just as well be the other way around.

1. While many scholars who subscribe to a performative notion of ontology claim that reality emerges not just in interactions but intra-actions, referencing Barad (2007), I am not commited to making what is essentially a very strong ontological claim. Rather than commiting to a view that claims nothing exists outside or external to a relation, which is what Barad’s notion of intra-action implies, my view is more in line with Deleuze’s notion of assemblage, which assumes a certain autonomy to the terms they relate. See Hein (2016) for a discussion of how compatible Barad and Deleuze are.

1. Thanks to Michael Veale for pointing out these distinctions.

1. Technography has previously been described as an “ethnography of technology” (Kien, 2008), a research strategy aimed at uncovering the constructed nature of technoculture (Vannini & Vannini, 2008), a way to explore artefacts in use ( Jansen & Vellema, 2011), or a method “to tease out the congealed social relations embodied in technology” (Woolgar, 1998: 444). Moreover, the notion of technography has been used to describe the “biography of things” (Kahn, 2004) or simply as a synonym for a “technical drawing,” for example, within the field of architecture (Ridgway, 2016). My understanding comes closest to that of Woolgar (1998) and of Vannini and Vannini (2008), who describe technography as a general atitude and research strategy aimed at examining the structural aspects of complex technocultural layers. However, the way in which I use the term difers from the notion of technography in Woolgar and in Vannini and Vannini in that the descriptions of technology do not necessarily involve a description of situated practices of production and implementation in particular setings. The question is, rather, what can we know about the workings of algorithms without necessarily asking humans or studying particular social setings as an ethnographer would do?

Chapter 4

1. htp://newsroom.fb.com/company-info (accessed June, 2017).

1. As Leah Lievrouw (2014) has argued, the tendency for STS and communication scholarship is to privilege the social side. In emphasizing the opposite, this chapter is highly indebted to scholarship on materiality in media studies (Kitler, 1999; Meyrowitz, 1994; McLuhan, 1994), cultural theory (Galloway, 2004; Packer and Crofts Wiley, 2013; Parikka, 2012), sociology, and organization studies (Leonardi et al., 2013; Mackenzie, 2008; Marres, 2012), social media research (Gerlitz and Helmond, 2013; Langlois, 2014; van Dijck, 2013), and software studies (Berry, 2011; Chun, 2011; Fuller, 2008).

1. htp://vimeo.com/111171647

1. See Andrew Bosworth’s depiction of the Facebook boot camp at Quora (last accessed February 9, 2015): htp://www.quora.com/How-does-Facebook-Engineerings-Bootcampprogram-work

1. See FB tech talk “behind the code” in which Facebook engineers discuss the boot camp and the company culture (last accessed February 9, 2015): htp://www.livestream.com/fbtechtalks/ video?clipId=pla\_c1161c6b-dd52-48e7-a8ea-abeb3b198375&utm\_source=lslibrary&utm\_ medium=ui-thumb

166

NOTES TO PAGES 56–70

1. Throughout the company’s history, Facebook has been accused of violating the privacy of its users. There is a rich existing body of literature concerning Facebook and privacy. See, for example, boyd (2008); Debatin, et al. (2009); Hargitai (2010); Tufekci (2008).

1. In Gilles Deleuze’s (2006) reading of Foucault, the concept of diagram signifies the function or operationality of power. Deleuze suggests that Foucault is a new cartographer, someone intent on mapping out the relations between forces, to show how power produces new realities.

1. The dimension of subjectivity is key in the writings of Foucault (especially in the later ones). Foucault used the concept of governmentality to analyze everything from the production of orderly and compliant “docile bodies” through pastoral guidance techniques (see Foucault, 1978) to the emergence of liberalism in which notions of freedom are produced so as to replace external regulation by inner production (Bröckling et al., 2011: 5).

1. Many scholars have argued that Bentham’s Panopticon and Foucault’s adaptation of it in terms of disciplinary power provides a limited model for understanding contemporary forms of surveillance and power (e.g., Boyne, 2000; Green, 1999; Haggerty & Ericson, 2000). One of the more pervasive arguments used against the notion of disciplinary power is what we may call the Deleuzian turn in surveillance studies (Lyon, 2006). In a short piece called the Postscript on the Societies of Control, Deleuze argues that discipline has been superseded by something he calls “control.” Despite the brevity of Deleuze’s essay (only a few pages), the fact that he explicitly connects control to code, automation and networked technology, probably accounts for the main reason scholars find the notion of control to be more useful. However, as Mark G.E. Kelly (2015) points out, much of the arguments voiced against Foucault’s notion of discipline have failed to recognize that “discipline is control.” It is not the case, as Deleuze suggested, “we are in the midst of a general breakdown of all confinements” (1992: 178). Prisons and schools still persist. Nor is it the case that discipline means confinement in the way that Foucault introduced the notion. As Kelly (2015) suggests, networked technologies such as CCTV and GPS are not evidence that discipline has been surpassed by control, but rather complete the idea of disciplinary power as discipline is essentially about the infinite management of bodies. Others have in turn argued that Foucault moved away from notions of discipline in his later work, focusing instead on notions of biopower, security, and governmentality. Yet, in Security, Territory, Population Foucault makes sure to not posit a historical break between discipline and mechanisms of security when he writes: “So, there is not a series of successive elements, the appearance of the new causing the earlier ones to disappear. There is not the legal age, the disciplinary age, and then the age of security. Mechanisms of security do not replace disciplinary mechanisms, which would have replaced juridico-legal mechanisms” (2007: 22). While I can certainly understand the appeal of moving away from a notion of discipline toward a more fluid understanding of control as proposed by Deleuze or the notion of an expansive power indicative of Foucault’s security apparatuses, I don’t think we have to choose. It is not an either/or; not discipline or control; not discipline or security. I concur with Foucault when he says that discipline and security are not opposites but part of the same atempts at managing and organizing social spaces. For these reasons I still think the notion of disciplinary power holds great merit for an understanding of the ordering mechanisms of algorithms.

1. htps://newsroom.fb.com/news/2013/08/announcing-news-feed-fyi-a-series-of-blogs-onnews-feed-ranking.

1. For more information on the news feed disclosure, see the video broadcasts of the technical sessions of the 2010 Facebook developers conference, especially the session called “Focus on the feed” in which Facebook engineers Ari Steinberg and Ruchi Sanghvi talk about the details of EdgeRank: htp://www.livestream.com/f8techniques/video?clipId=pla\_5219ce25- 53c6-402d-8ef-f3f8f7a5b510

1. While the term “Facebook algorithm” may be somewhat misleading in terms of giving the impression that there is one single Facebook algorithm as opposed to many diferent algorithms accomplishing diferent tasks and working in tandem, my reference to the algorithm in this singular way should be read as an act of convenience on my part. Facebook algorithm is shorter and more commonly used in everyday discourse than EdgeRank or the longer and more precise “news feed ranking algorithm.”

NOTES TO PAGES 71–77

167

1. As exemplified in a Slate article about the Facebook news feed: “It doesn’t just predict whether you’ll actually hit the like buton on a post based on your past behavior. It also predicts whether you’ll click, comment, share, or hide it, or even mark it as spam. It will predict each of these outcomes, and others, with a certain degree of confidence, then combine them all to produce a single relevancy score that’s specific to both you and that post” (Oremus, 2016).

1. Kairos refers to the ancient Greek notion of the opportune or the right time to do or say something. For more details on the concept of kairos, see, for example, Boer (2013), or Smith (1969).

1. For a detailed description on how trends are defined and algorithmically processed in Facebook, see Li et al. (2013).

1. Daniel Ek, who is the founder of the music streaming site Spotify, which entered into a close partnership with Facebook in 2011, tellingly suggests: “We’re not in the music space—we’re in the moment space” (Seabrook, 2014).

1. These are just some of the changes made to the news feed as documented by the Facebook News Feed FYI blog (htps://newsroom.fb.com/news/category/news-feed-fyi): May 2017, reducing the amount of posts and ads that link to “low-quality web page experiences” by using artificial intelligence to “identify those [web pages] that contain litle substantive content and have a large number of disruptive, shocking or malicious ads.” January 2017, emphasizing “authentic” content by looking at “signals personal to you, such as how close you are to the person or Page posting.” August 2016, reducing the ranking of updates that classify as having clickbait-like headlines. June 2016, posts from friends and family to get top priority followed by posts that “inform” and posts that “entertain.” March 2016, prioritizing live videos. June 2015, tweaking the algorithm to take more account of user actions on videos such as turning on sound or making the video full screen as indications of heightened interest. June 2015, taking into account time spent on stories as indicated by users scrolling behavior. April 2015, making posts from friends rank even higher and relaxing previous rules about not showing multiple posts in a row from the same source. September 2014, prioritizing more timely stories, for example as indicated by two people talking about the same issues (i.e., a TV series or a football game). Timeliness, moreover, is indicated by when people choose to like or comment on a post. If more people like a post right after it was published but not a few hours later, the algorithm takes it as an indication that it was more relevant at the time it was posted. What’s interesting to note about the blog posts documenting changes is that the posts from 2017 more readily mention artificial intelligence as part of the procedures and techniques for making the news feed more relevant.

1. See chapter 3 for a discussion of the possibilitities and challenges of applying a reverse engineering ethos to algorithms.

1. Facebook alternately calls these types of stories aggregate or cluster stories. See Luu (2013) for a description of cluster stories.

1. The most recent feed used to have a counter next to it indicating the number of new stories that had been posted and published by a user’s Facebook connections since the last time they had checked the feed. The counter only went as far as 300 new posts, so any number above the limit would just be indicated as “+300.” Usually, my most recent feed would reach this limit only after one or two days of not checking.

1. In September 2011, I did this kind of comparison between my “top news” feed before and after checking the most recent feed a couple of times; and, every time, there was a considerable change in the stories displayed between the first and second time I checked.

Chapter 5

1. @Desareon, February 8, 2016.

1. @doughj, February 7, 2016.

168

NOTES TO PAGES 78–93

1. @YazibelleXO, February 6, 2016.

1. For Raymond Williams, the notion of “structures of feeling” denotes a historically distinct kind of feeling and thinking around a particular quality of social experience and relationship (1977: 130). What defines this social experience is its emergent character as something felt but still not fully articulable as such due to its historical unfolding presence in the here-andnow. For Williams, “structures of feelings” became a way of recognizing moments of new experiential impulses and paterns as a way to analyze cultural change and the unruly reality of social life.

1. As Clough and Halley (2007) point out, the social sciences and humanities have seen an “afective turn” in times of ongoing conflict and crisis, a turn that has sought to make analytical sense of what cannot necessarily be explained or made sense of using the language of reason. The study of afect, understood as the study of how bodies are moved, how the world in some way or another afects us and how we, in turn, afect the world, has atracted a number of scholars and thinkers within philosophy, feminist theory, psychology, political science, cultural and social geography, media studies, and critical theory—to name but a few. As suggested by Gregg and Seigworth in their introduction to The Afect Theory Reader (2010), two vectors have been particularly pertinent in the turn to afect, both culminating with publications in 1995. On one hand, there was the essay by Sedgwick and Frank called “Shame in the Cybernetic Fold” (1995) and the corresponding book Shame and Its Sisters (1995), both of which introduced readers to the psychology of afect developed by Silvan Tomkins. On the other hand, there was an essay by Brian Massumi on the “Autonomy of Afect” (1995), which expanded on Deleuze’s Spinozan ethology of bodily capacities. So, if the first vector of afect sets afect apart from previous psychoanalytic conceptions of drives as the primary motivational system in human beings, the second strand concerns the many contemporary reinterpretations of Spinoza’s ethico-political writings (Spinoza, 2000)— most notably, by Deleuze (1988, 1990) and his followers. On the most basic level, afect can be understood as the capacity to afect and be afected.

1. Email interview with “Michael,” October 3, 2014.

1. Paraphrased from a tweet “Rachel” posted on October 14, 2014.

1. Multiple email interviews and online chats with “Rachel,” October 15–26, 2014.

1. On a more general level, the question of how people encounter and make sense of new technology is, of course, one the core questions not only within media studies but also in related fields such as STS and IS research. In media studies, these questions have been extensively discussed within audience and media use research—for instance, through the lens of domestication theory, which focuses on the emotional and symbolic dimensions of communication technologies and how the meanings of technology have to be understood as emergent properties of contextualized audience practices (see Berker, Hartmann & Punie, 2005; Hirsch & Silverstone, 2003; Morley & Silverstone, 1990; Morley, 2003). In science and technology studies, the publication of The Social Construction of Technological Systems in 1987 popularized the notion of the social construction of technology, or SCOT, as a way to theorize and analyze technological innovation not as the result of designers but as fundamentally co-constructed in negotiations with users and other stakeholders (see Bijker & Law, 1994; Bijker et al., 2012; Pinch & Bijker, 1984). One of the most influential ideas in terms of understanding the meanings users atribute to technology is the notion that artefacts have “interpretative flexibility,” meaning that there is flexibility in the design, use, and interpretation of technology (see Orlikowski, 1992). The idea that technologies do not have fixed meanings or determine use has also been a cornerstone of much research within information systems and organization science—particularly, in fields such as computer-supported cooperative work (CSCW), human-computer interaction (HCI) and social informatics (see, for example, Lamb & Kling, 2003; Kling, 1980; Orlikowski & Gash, 1994; Orlikowski, 2000). One of the earliest accounts of the relationship that people forge with computers is Sherry Turkle’s book The Second Self (1984) in which she describes the computer as a “marginal object” that defies easy categorization. A computer, she found, means more to people than merely a physical thing. It is also a metaphysical thing that influences how people think of themselves and others (Turkle, 1984: 16).

NOTES TO PAGES 93–96

169