University of Illinois Press

Chapter Title: CODING AND ENCODING STREAMED MEDIA: The Cultural Infrastructure of
the Netflix Recommender System
Chapter Author(s): FATIMA GAW

Book Title: Media Backends

Book Subtitle: Digital Infrastructures and Sociotechnical Relations
Book Editor(s): LISA PARKS, JULIA VELKOVA, SANDER DE RIDDER
Published by: University of Illinois Press. (2023)
Stable URL: [Link]

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@[Link].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
[Link]

University of Illinois Press is collaborating with JSTOR to digitize, preserve and extend
access to Media Backends

This content downloaded from [Link] on Tue, 17 Mar 2026 18:48:30 UTC
All use subject to [Link]
 4
CODING AND ENCODING
STREAMED MEDIA
The Cultural Infrastructure
of the Netflix Recommender System

FATIMA GAW

Netflix reached almost 220 million subscribers in 2021, making it the largest
streaming media service in the world. Undergirding its streaming media infra-
structure is a set of intricate algorithms that distinguishes Netflix from others—
personalized, up-to-date, and “so good at suggestions that [it’s] able to show you
exactly the right film or TV show for your mood.”1 The Netflix Recommender
System (NRS), the algorithms’ official name, is at the core of the streaming plat-
form. It organizes Netflix’s vast library of content, analyzes billions of hours of
watch time, and generates tailored recommendations for audiences across 190
countries. By relying on taste attributes instead of demographic information
such as location, race, and gender, NRS purportedly “transcends the predictions
of stereotypical demographics.”2 This makes the NRS a model among engineers
and programmers on the promise of algorithms, machine learning, and AI in
innovating cultural experiences.
The NRS exemplifies algorithms as ubiquitous, deeply entangled, and in-
dispensable media technology in contemporary culture. The era of big data
requires computational technologies to mediate between platforms and end
users to organize knowledge, processes, and experiences. For instance, Google’s
PageRank algorithm crawls the web to catalog billions of pages and websites
and match them with user searches.3 Twitter employs an algorithm to identify
“trending” topics both within your location and personalized to your network.4
Spotify algorithmically curates playlists from millions of tracks to fit particular
dayparts, moods, and activities of its users.5 Algorithms form part of media
backends as sociotechnical systems that facilitate the production, distribution,
and consumption of cultural experiences beneath layers of obscured materiali-
ties, processes, and practices.

This content downloaded from [Link] on Tue, 17 Mar 2026 18:48:30 UTC
All use subject to [Link]
 coding and encoding stre amed media . 81

Despite their necessary pervasiveness, algorithms are elusive. Apart from

their technical complexity, they are black boxes by virtue of being “trade secrets”
involving proprietary information and techniques.6 When required to be dis-
closed to regulators, algorithms are obfuscated in very simplistic or inexplicably
complicated terms to strategically deflect questioning. Conceptually, algorithms
are also difficult to locate. They are foremost considered as technological arti-
facts whose values and standards are impressed in their material code, input,
and output.7 They are also produced through collective cultural practices in
diverse social contexts, which Nick Seaver argues not only situate algorithms
in culture but also enact them as culture.8 Further, algorithms are enveloped
within platforms and are examined as components, among others, that facili-
tate programmable sociality and experiences.9 On their own, algorithms enact
their rules and decisions that shape a range of cultural processes, exhibiting and
performing computational agency.10
In all these conceptions, algorithms are made meaningful not by themselves
but in relation to various actors and objects—programmers and users, data and
code, and protocols and structures. The contingency of algorithms on social
interactions as well as their capacity to set the conditions that govern these en-
gagements point me to investigate them as infrastructures. Infrastructures are
produced out of this “double articulation”11 where objects are born out of other
objects to create “multiple, overlapping, and perhaps contradictory infrastruc-
tural arrangements.”12 Problematizing algorithms as infrastructure allows us to
study them at multiple scales and scopes and accommodate inconsistencies and
conflicts in their operations.
In this chapter, I investigate “when” algorithms are enacted as infrastruc-
tures through the cultural processes that mutually animate their technical, cul-
tural, and social aspects.13 Using the NRS as one of the most popular algorith-
mic systems in the world, I trace these processes by adapting Taina Bucher’s
methodological tactic of interrogating algorithms as configurations, which are
constructs constituted by heterogeneous, incomplete, and contradictory parts
that “form the appearance of more or less coherent relations.”14 I assemble this
configuration in situated practices where the technical and cultural become
inextricable—algorithms as designed by developers, imagined by media, and
experienced by users. While the NRS’s materiality remains obscure, I draw
from the translations of “technical specifications” by its engineers from Netflix
technical and corporate documents.15 Then, I weave the narratives that estab-
lish, dispute, and expound popular conceptions and controversies of the NRS
in media, in the form of Benjamin Burrough’s “streaming lore” formulated by
industry, media, and cultural discourses and practices.16 Lastly, I encapsulate
the social encounters of users with the NRS as publicly discussed on Twitter,
particularly during experiences of algorithmic breakdown and anomalies as
well as user resistance.17

This content downloaded from [Link] on Tue, 17 Mar 2026 18:48:30 UTC
All use subject to [Link]
82 . sensing , auto mating , mediating

I conceptualize this configuration of the NRS as a “cultural infrastructure”

whose structures and protocols are designed to reconcile and reconstruct the
technical, cultural, and social discrepancies emerging from its relocation from
the articulations of code into contexts of encoding. Cultural infrastructures
perform the dual work of the governance of social processes and generation
of cultural standards through the mutually constitutive relations between the
infrastructures’ material and social conditions. Algorithms as cultural infra-
structures not only direct paths for action but also create spheres of influence
where decisions are made. Where there are windows of negotiation and resis-
tance, however, algorithms tend to supersede them through socializing users
into deference, restricting user agency, and privileging algorithmic agency. This
theoretical intervention integrates intersections and incompatibilities of research
on algorithms from different disciplinal strands by acknowledging their duality
as both products and producers of cultural processes.18

Algorithm as Infrastructure, Infrastructures as Algorithmic

In an influential paper, Robert Kowalski defines an algorithm as a computing
program that identifies information and employs strategies to solve complex
problems.19 Ted Striphas argues that algorithms not only work to make sense
of enormous amounts of data but also “to expose some truth or tendency
about the world.”20 Ed Finn calls algorithms “culture machines” as they simu-
late reality by using “all sorts of technical and intellectual inferences, inter-
ventions, and filters.”21 These algorithmic mechanisms operate recursively,
creating feedback loops where “each iteration introduces a deviation.”22 The
variations fold back infinitely to the system, creating “models [that] analyze
the world and the world responds to the models.”23 As such, an algorithm both
anticipates outcomes and generates its own procedures in filtering, ordering,
and structuring reality.
Algorithms are embedded in larger systems, from software to platforms and
infrastructures, often sinking into a singularity with other components. Platform
studies often subsume the discussion of algorithms with that of affordances, big
data, and platform governance.24 Yet, the same works acknowledge algorithms
as consequential to the operations and outcomes of platforms. Critical algo-
rithm studies examine algorithms as conceptually distinct objects of inquiry,
problematizing their power to regulate cultural production, circulation, and
consumption.25 I argue that both areas of study fall short in capturing the con-
tingency of algorithms on other structures, arrangements, and organizations and
their capacity to set the conditions of cultural processes within these systems.
In the case of the NRS, its functioning is dependent on the Netflix platform,
Netflix Inc., and the cultural contexts in which it is embedded. The relationship

This content downloaded from [Link] on Tue, 17 Mar 2026 18:48:30 UTC
All use subject to [Link]
 coding and encoding stre amed media . 83

of the NRS with these structures then configures how the algorithms are en-
acting cultural experiences, such as the creation of user taste profiles based on
platform affordances or the expectations of demographic agnosticism from
corporate claims of algorithmic neutrality. This double articulation requires
reconceptualizing the study of algorithms from being neither part nor whole to
whole parts of other wholes. Infrastructure studies might bring to the fore this
character of algorithms.
Infrastructures are “built networks that facilitate the flow of goods, people,
or ideas and allow for their exchange over space.”26 They are heterogeneous
material and intellectual structures, mechanisms, and protocols that organize,
standardize, and inform the order of things. Leading infrastructure researchers
emphasize that an infrastructure’s power lies in its capacity to create the condi-
tions to which all decisions and actions are subjected and orient them toward
particular directions.27 Recently, media scholars examine this “possibility-fixing”
power in media infrastructures,28 particularly through the ways they shape
media environments, distribution processes, and cultural experiences.29
Of interest in infrastructure as an analytical lens is its relational character,
“becoming real infrastructure in relation to organized practices.”30 This reframes
the question from “what is an infrastructure” to “when it becomes an infra-
structure,”31 locating its constitutive components in conjunction with techni-
cal, social, and cultural processes. I conceptualize algorithms as infrastructures
along this line of analysis and argue that they represent a new class of media
infrastructures built on artificial intelligence. I draw from Susan Leigh Star and
Karen Ruhleder’s dimensions of infrastructures to build this case: embedded-
ness, transparency, reach and scope, membership-based learning, linkage with
conventions of practice, embodiment of standards, superimposition over an
installed base, and visibility upon breakdown.32
First, algorithms have become a “kind of invisible structural force,” con-
cealed behind technical complexity and commercial interests, enmeshed in
heterogeneous digital systems, and entrenched in everyday social contexts.33
Second, they “always work, no matter what the inputs” and operate automa-
tedly.34 Third, algorithms are extensive and encompassing, from reading present
data patterns to predicting future probabilities, as well as scaling personalized
outcomes to mass customization.35 Fourth, algorithms enact standards and pro-
cedures through quantification: transforming people, objects, and actions into
“computable abstractions.”36 Fifth, their material existence is built by someone
(e.g., developers) and is installed on existing structures (e.g., software) that set
the parameters of code.37 Sixth, it is intuitive for users to know how algorithms
work without having to understand their technicalities as they have become
“cultural” objects themselves.38 Seventh, as much as algorithms are self-sufficient,
they are entangled in the social conventions of their proprietors, developers,

This content downloaded from [Link] on Tue, 17 Mar 2026 18:48:30 UTC
All use subject to [Link]
84 . sensing , auto mating , mediating

and users.39 Eighth, for the reasons above, algorithms are generally reliable but
are also prone to malfunction, if not misinterpretation.40
By these definitions, algorithms fit Star and Ruhleder’s standards of what
makes an infrastructure, but they also have characteristics outside this archetype.
While infrastructures have a quality of permanence, algorithms are program-
mable such that part of their code can be altered that can affect their function-
ing.41 The NRS has been revised and updated through the years to produce more
“accurate” recommendations as Netflix expands its library and global audience.
Infrastructures also generally perform a facilitative function, where algorithms
also have the generative capacity by producing derivative data, optimized pro-
cedures, and new knowledge.42 As you will see later, the NRS has created new
data categories, optimization processes, and cultural products beyond what we
customarily consider as “recommendations.” This renders algorithms agential
properties, enacting their computational agency beyond their programming
and exceeding their logistical power as an infrastructure.

The Tale of the Netflix Recommender System

Netflix pioneered streaming in 2007, and every big media company followed
suit. While the competition like Disney+ leverages its stockpile of legacy media
franchises and Amazon repurposes its retail algorithms to Prime Video, Netflix
stands above the market through its global recommender system. The NRS lever-
ages its “global communities” to infer individual taste preferences from global
patterns of consumption, as country-level data might be limited in generating
“accurate” recommendations.43 This abundance of data enables the algorithms
to personalize their recommendations for users, both content they already like,
and those they are yet to discover.
Netflix has been experimenting with algorithms during its DVD-for-rent years
with Cinematch to generate recommendations for customers on which movie to
rent. In 2006, it launched the Netflix Prize, a three-year competition to improve its
algorithm’s ratings prediction. While the winning algorithm was not implemented
given excessive operational costs, the contest popularized the Netflix algorithms
in the tech industry. After 2009, the NRS becomes so prominent in media that
it is not only known as the platform’s operating system but as synonymous with
Netflix. Netflix has also been more forthcoming about how its algorithms oper-
ate, publishing explanations from developers in its Netflix-owned websites and
engaging media to translate the complexities of the NRS for popular consumption.
However, these remain either highly technical or too broad to explain the work-
ings of the NRS. The gap between these two ends is filled in by what Burrough
calls streaming lore, which creates our expectations of how the algorithms work
and shape our experiences when we encounter them.44

This content downloaded from [Link] on Tue, 17 Mar 2026 18:48:30 UTC
All use subject to [Link]
 coding and encoding stre amed media . 85

A prominent narrative around the NRS is its guarantee of personalized, qual-

ity recommendations. Personalization purportedly ensures that “no two Netflix
experiences are alike” and the more users spend time on the platform, the more
personalized the recommendations are.45 This intimate knowledge of users’
taste creates the impression that the NRS knows users better than they know
themselves, and thus it is trustworthy.46 The underlying assumption here is that
the algorithms are neutral and objective, relying wholly on the data to serve
users’ interests and not Netflix’s commercial ends. This relates to another well-
known streaming lore about the NRS deriving its recommendations exclusively
from users’ taste preferences and not from demographic data. The algorithms
are allegedly agnostic on class, gender, race, and other social identities, much
like the promise of the internet of disembodiment from traditional markers of
identity.47 This postdemographic profiling departs from conventional practices
in the industry, supposedly freeing the system of the biases and prejudices that
might box users into specific categories. Going beyond the user experience, it
is also commonly understood that Netflix uses its algorithms to inform busi-
ness and creative decisions—from what kinds of programs to produce, which
syndicated shows to keep in its library, and how to manage diversity in its cata-
log.48 The success of its debut Netflix Original, House of Cards, and its other big
bets was purportedly predicted by algorithms as a testament to their superior
computational judgment.
As I will illustrate later, some claims in streaming lore are contested, but they
also reveal how the NRS fortifies its influence beyond its technical capacities. It
establishes the algorithms as cultural objects, with both developers and users
having their algorithmic imaginaries of how the NRS works, what to expect
from it, and what it should do if it fails to perform accordingly.49 Users are also
incorporating algorithms in their everyday life and algorithmic recommen-
dations are cultivating users into ideal consumers, shaping each other in the
process.50 The relocation of algorithms from industry lore to media narratives
and social contexts helps articulate the materiality of the algorithms, as well
as their social construction through the ways people attach meanings to their
algorithmic encounters.

The Configuration of Algorithmic Infrastructures

Constructing the “configuration” of the Netflix Recommendation System in-
volved assembling technical, cultural, and social articulations of the logics of its
algorithms, deconstructing their underlying assumptions and procedures, and
articulating the cultural processes that animate the infrastructure. The follow-
ing discussion draws from the analysis of sixty technical documents primarily
from the Netflix Technology Blog, academic papers and corporate documents,

This content downloaded from [Link] on Tue, 17 Mar 2026 18:48:30 UTC
All use subject to [Link]
86 . sensing , auto mating , mediating

and 100 media reports from 2007 to 2019, as well as 990 unique Twitter posts
from February to May 2019 across countries.
My investigation has identified three key processes that underpin the cul-
tural infrastructure of the NRS: classification of data, signals, and actions into
predetermined types and criteria; categorization of data into emergent groups
and cultural constructions; and codification of cultural relations and identities
through recommendation. The protocols and procedures of these infrastruc-
tural processes, however, emerge not out of the consistencies but out of the
discrepancies and contradictions of their material and social articulations.

Classification
Pertinent Netflix documents and interviews with executives point out to the
NRS employing two classification schemes: first is to classify user data into
explicit and implicit taste preferences, and second is to evaluate titles based on
a set of criteria. Explicit taste preferences pertain to user actions that directly
inform the algorithms of their tastes and interests. These user activities include
answering a profile survey upon sign-up, adding titles to the watchlist, and rat-
ing content. Netflix used to have a five-star rating system, but it was replaced
with a binary thumbs-up or down system in 2017 to purportedly orient users
away from performative social standards such as box office results and critics’
reviews, and toward “intuitive” personal feedback.51 Implicit taste preferences
are data points that indirectly signify user interests, such as plays, watch time,
searches, and clicks. Between the two classifications, Netflix put a premium on
implicit over explicit taste preferences because the latter are deemed “aspira-
tional” while the former represent actual consumption.52
Titles are also classified based on a set of criteria that filters what appears in
the interface. Apart from tagging titles for “relevance” to users’ tastes, “popular-
ity” is used to identify which groups of titles are prominent in particular taste
communities, timeframes, or locations. While Netflix engineers recognize that
popularity might conflict with relevance, they assert that the algorithms indicate
that “a member is most likely to watch what most others are watching.”53 The
NRS also purportedly evaluates titles based on their “diversity” to respond to
the range of tastes and interests of users and households. Some other criteria
such as context, freshness, and novelty were identified briefly in the Netflix tech
documents but were not discussed.
These classification protocols are generally overlooked by users, except when
they were “failed” by the NRS, when it “misreads” their taste, or dismisses their
explicit taste preferences altogether. Titles they down-rated were allegedly re-
peatedly recommended to them, while those they up-rated or queued in their
watchlist were not showing up on their homepage. Users also criticize the al-
gorithms for confining them to a narrow set of interests despite the broad array

This content downloaded from [Link] on Tue, 17 Mar 2026 18:48:30 UTC
All use subject to [Link]
 coding and encoding stre amed media . 87

of genres they watched on the platform. They also complain that they needed
to search for content they like because the algorithms “bury” it deep into the
library, while Netflix Originals are always displayed in the “popular” row.

Categorization
Categorization includes the ways the NRS organizes its data into new and un-
conventional groups and clusters. These categories are publicized broadly in
media reports to definitively distinguish the NRS’s ascendancy among other
recommendation systems. In an article in The Atlantic, reporter Alexis Madrigal
(2014) has reverse engineered how Netflix categorized its content and discov-
ered altgenres—thousands of reimagined categories of content constructed by
fragmenting films and movies into granular attributes. Titles are deconstructed
based on specific aspects like the moral stance of the characters, the degree of
romance and level of gore, and then are graded on a scale. These “microtags”
are then used to formulate hyperspecific genres such as “Feel-Good Disney
Talking-Animal Animation,” or “Visually Striking Cerebral Fight-the-System
Movies.” Through altgenres, Netflix purportedly avoids “genre bias” and opens
new pathways for content discovery.54
Taste communities are another categorization convention of the NRS. They
are algorithmically generated groups that cluster individuals sharing similar
tastes, detached from demographic information used in traditional audience
targeting. Taste communities were Netflix’s way of integrating their national
audiences into a single global entity, drawing from geographically dispersed
users from the same taste communities to inform of the collective interests of
the group. In a report by Adage, there are at least 2,000 taste communities man-
aged by the NRS in 2020, but their actual categories and composition remain
obscure.55
While appearing to be cutting-edge, these categories present some issues
when translated into recommendations. Users narrate their experiences of
unrelated or inappropriate titles being recommended after watching particu-
lar shows. For instance, users were in disbelief to be served the true-crime
documentary Ted Bundy Tapes right after finishing the children’s show Peppa
Pig or the eighties classic The Breakfast Club as “actual suggestion Netflix’s al-
gorithm came up with.” It was also pointed out that several Netflix Originals
were “plastered everywhere” in multiple altgenres. Contrary to the promise of
postdemographic targeting, controversies on racial and gender profiling persist
in media and in everyday algorithmic encounters. In October 2018, several
African American users noticed that they are being served title artwork that
featured Black characters in films and shows with a predominantly White cast.
Podcaster Tolani Shoneye told The Guardian that she “noticed it a while ago
with a Zac Efron film that [she]’d already seen, but Netflix kept showing it as a

This content downloaded from [Link] on Tue, 17 Mar 2026 18:48:30 UTC
All use subject to [Link]
88 . sensing , auto mating , mediating

Michael B. Jordan movie.”56 Reporting indicates that viewers felt “manipulated”

by Netflix by reducing their taste to their race and manufacturing diversity in
an otherwise overwhelmingly White content catalog. Twitter users echo this
experience with Black users who were purportedly “read” as Caucasians by the
algorithms and were exclusively getting “White” films, while their peers were
being served with “Black” titles that were not in their recommendations. The
algorithms also were under fire in March 2019 for allegedly arranging the epi-
sodes of the animated anthology Love, Death, + Robots based on user’s sexual
orientation.57 Queer viewers report that the arrangement “changes based on
whether Netflix thinks you’re gay or straight” having been served the episode
with a sexually explicit lesbian episode, while their cisgender friends had the
heteronormative episode queued on top. Gay users in my Twitter data were
also allegedly being “queerbaited” by showing same-sex couples on recom-
mendation artwork. In both cases, Netflix insists that its targeting schemes
are randomly generated by the algorithms and it does not collect demographic
information from its users.58 By deflecting the blame on algorithms, Netflix
dismisses the possibility of the NRS recognizing demographic patterns in its
algorithmic categorization, even when media reports and Twitter users cor-
roborate being demographically targeted.

Codification
Codification is the process of materializing the algorithms’ cultural codes in the
form of recommendations that prescribe the orientation, choices, and actions
that users make in the platform. The personalized homepage lays out the rec-
ommended altgenres in each row, the selected titles per row, and their ranking
arranged in order of compatibility. All the recommendations have undergone
A/B testing, which evaluates the “effectiveness” of sets of recommendations and
selects the most compelling ones to be rolled out to a majority of the users. In
2016, Netflix also began personalizing artwork by algorithmically identifying
the images that purportedly “highlight the aspects of a title that are specifically
relevant to them.”59 This visual “evidence” is complemented by numerical evi-
dence of the algorithms’ accuracy called the match score, which calculates the
compatibility of titles to users’ tastes. Titles with scores below 50 percent are
allegedly not considered for recommendation altogether.
This codification of recommendations has become problematic in cases when
it misrepresents the films and shows they serve. For the show Grace and Frankie,
the artwork of actress Jane Fonda was said to perform poorly in the A/B testing
and her images were planned to be removed despite Fonda’s being the show’s lead
character.60 Another controversy involved the alleged whitewashing of artwork
for the show Nailed It! where images of the two White men who were only sup-
porting personalities were being served instead of its host Nicole Byer, who is a
Black woman.61 There were also tweets about users being frustrated about not

This content downloaded from [Link] on Tue, 17 Mar 2026 18:48:30 UTC
All use subject to [Link]
 coding and encoding stre amed media . 89

having the tools to correct the algorithms when the recommendations do not
match their tastes. Some users opted to “train” the algorithms to their liking,
while others request to have the ability to edit their viewing history or directly
send their feedback through reviews.

The Power of Cultural Infrastructures

The algorithmic processes of classification, categorization, and codification
are reminiscent of the protocols of conventional infrastructures such that they
intend to streamline “local” practices into “global” operations.62 Algorithms as
infrastructures, however, engage in both standardized operations and standard-
ized production, which in the case of the NRS is to generate bespoke outcomes
out of the perpetual intake of data, alterations in code, and changes in organiza-
tional structures. Thus, for the algorithms to function, they necessarily open up
the infrastructure to inquiry, dispute, and “repair” by developers, stakeholders,
and users that might undermine their stability and coherence.63 What I argue in
this chapter is that algorithms maintain their structural integrity despite being
subjected to human intervention by operating as cultural infrastructures. Cul-
tural infrastructures embody the dual capacity of governing social processes
within their built structures and generating the cultural standards from which
social processes are based and validated. Contrary to the conceptual distinction
made by Seaver, “cultural” here conveys that the algorithms are both located “in
culture,” such that they “affect culture and culture can affect algorithms because
they are distinct,” and exist “as culture” enacted by composite practices of hu-
mans and machines that produce their own cultural assumptions, standards,
and values.64 The cultural constructs produced out of the social relations among
various actors mutually constitute the cultural construction performed by the
materiality of the algorithm, and together they comprise the “control and value
. . . indissolubly linked to the machine ensembles.”65 Below, I discuss the ways
the NRS enacts the power of the cultural infrastructure of algorithms.
The underlying protocol of any algorithmic system is to translate data into
quantifiable attributes. Quantification enables the algorithms to both manage in-
formation and create new models that define how information is deconstructed
and decoded, which categories and hierarchies they are classified and assigned,
and what new compositions and structures are reconstructed from them. More
than a recipe of instructions, these models are calculated relationships that
establish the rules that determine all decisions made by the algorithms. They
are if-then statements that prescribe “ways of seeing” the relations between
cultural constructs, artifacts, and identities (if “gender” is “gay” then “artwork”
is “same-sex couple”) and they constitute the complex cultural schema of the
algorithmic infrastructure.66 However, rules are universal, and recommenda-
tions are supposedly personal. Where there is a conflict between the two, the

This content downloaded from [Link] on Tue, 17 Mar 2026 18:48:30 UTC
All use subject to [Link]
90 . sensing , auto mating , mediating

universal supersedes the personal. In the NRS, subjective expressions of taste

are necessarily reduced into binary choices and discredited as aspirational and
performative if they do not translate to consumption. The popularity criterion
homogenizes recommendations based on the preferences of an invisible, pre-
sumably Western cultural majority. Demographic targeting surfaces despite its
supposed absence in the data, because race, gender, and sexual orientation are
categories that can accommodate generalized assumptions about people. For
Netflix, its algorithmic models always work even if that means neglecting the
nuances, ambiguities, and even dissonances of individual subjectivities. Fur-
ther, the recursive process of the algorithms means the same rules govern the
generation of new rules. These algorithmic ontological models fundamentally
become self-affirming, not only enacting rules but also normalizing constructs
that are enduring, encompassing, and resistant to change.67
This intellectual structure of the algorithm is cemented not only by its com-
putational logic, but also by its social relations with its users. The streaming lore
about the NRS, among other myths about algorithms, has socialized us to trust
them, sometimes more than our own judgment. In particular, the promise of
personalization “hails” users into identifying with the characters, narratives, and
other reflexive images projected in the algorithmic recommendations.68 This
process of interpellation is performed through semantic codes embedded in
objects like altgenres and artwork, as well as through the spatial arrangements
through ranking and sequencing. Normative representations of race, gender,
and sexual orientation in the recommendations were particularly compelling,
despite users’ ambivalence about being demographically targeted. The subjec-
tification of users unequivocally manifests when they become hyperaware of
and involved in constructing their algorithmic identities.69 The frustrations and
anxieties over whether the algorithms are accurately reading their behavior or
responding to their feedback are rooted in the “sense of ownership, agency,
and presence” over their algorithmic identities.70 In an attempt to “correct”
or “train” the algorithms, users regulate their behavior and actions. This self-
regulation following the logics of the algorithms sustains the persistence of
ontological models, as well as providing users with a semblance of control over
the algorithms.
The NRS’s cultural schemas are reified in all parts of the infrastructure, manu-
facturing the norms, standards, and values of “quality,” “popular,” and “person-
alized” cultural experiences. The repackaging of a White-majority cast film by
using a Black character artwork stand-in reframes the film as relevant to Black
viewers, and relocates it to a marginalized cultural category, albeit temporarily.
Titles with a match score below 50, although arbitrarily assigned, disappear in
the recommendations completely. These algorithmic outcomes not only direct
users toward particular choices, but create the sphere of influence where present
and future actions are made. The NRS interface may look vast, but it conceals

This content downloaded from [Link] on Tue, 17 Mar 2026 18:48:30 UTC
All use subject to [Link]
 coding and encoding stre amed media . 91

a horizon of choices available to its users from its various infrastructural pro-
tocols. How the sphere is constructed allows but also constrains user agency
through “complicit forms of resistance” and those outside the parameters of its
prevailing models are dismissed as “unrecommendable.”71

Conclusion
Algorithms represent a new kind of infrastructure assembled from code, data,
and practices and they introduce a new “grammar,” a “language-like set of pro-
tocols for arranging the world and the organs of sensation.”72 I identified a few
of those procedures and protocols, most of which characterize what we know
of infrastructures and some present prospective dimensions. These distinguish-
ing capacities of algorithms as cultural infrastructures render to it the power
to both govern cultural processes and generate new standards and protocols
that animate those processes. This chapter underlines algorithms as perform-
ing the “politics of ‘as-if ’” by setting the conditions that define the possibilities
and boundaries of cultural experiences.73

Notes
1. Reed Hastings, “Netflix Keynote 2016,” (Lecture at Consumer Electronics Show,
Las Vegas, NV, January 6, 2016), [Link]
2. Ashley Rodriguez, “Netflix Divides Its 93 Million Users around the World into
1,300 ‘Taste Communities,’” Quartz, March 23, 2017, [Link]
-divides-its-93-million-users-around-the-world-not-by-geography-but-into-1300-taste
-commu.
3. Google, “How Search Works: How Results Are Automatically Generated,” https://
[Link]/search/howsearchworks/how-search-works/ranking-results/
4. Tarleton Gillespie, “#trendingistrending: When Algorithms Become Culture,” in
Algorithmic Cultures: Essays on Meaning, Performance and New Technologies, ed. Robert
Seyfert and Jonathan Roberge (London: Routledge, 2018), 52–75.
5. Maria Eriksson and Anna Johansson, “‘Keep Smiling!’: Time, Functionality, and
Intimacy in Spotify’s Featured Playlists,” Cultural Analysis 16, no. 1 (2017): 67–82.
6. Frank Pasquale, The Black Box Society: The Secret Algorithms That Control Money
and Information (Cambridge, MA: Harvard University Press, 2015), 12.
7. Christian Sandvig, Kevin Hamilton, Karrie Karahalios, and Cedric Langbort, “An
Algorithm Audit,” in Data and Discrimination: Collected Essays, eds. Seeta Peña Gan-
gadharan, Virginia Eubanks, and Solon Barocas (Washington, DC: New America’s Open
Technology Institute, 2014), 6–10; Rob Kitchin and Martin Dodge, Code/Space: Software
and Everyday Life (Cambridge, MA: MIT Press, 2011); Nicholas Diakopoulos, “Algorith-
mic Accountability: Journalistic Investigation of Computational Power Structures,” Digital
Journalism 3, no. 3 (2015): 398–415, [Link]
8. Nick Seaver, “Algorithms as Culture: Some Tactics for the Ethnography of Algo-
rithmic Systems,” Big Data & Society 4, no. 2 (2017).

This content downloaded from [Link] on Tue, 17 Mar 2026 18:48:30 UTC
All use subject to [Link]
92 . sensing , auto mating , mediating

9. José van Dijck and Thomas Poell, “Understanding Social Media Logic,” Media and
Communication 1, no. 1 (2013): 2–14; Jean-Christophe Plantin, Carl Lagoze, Paul N. Ed-
wards, and Christian Sandvig, “Infrastructure Studies Meet Platform Studies in the Age
of Google and Facebook,” New Media & Society 20, no. 1 (2018): 293–310; Taina Bucher,
If . . . Then: Algorithmic Power and Politics (New York: Oxford University Press, 2018).
10. Robert Prey, “Nothing Personal: Algorithmic Individuation on Music Streaming
Platforms,” Media, Culture & Society 40, no. 7 (2018): 1086–1100; David, ed., The Social
Power of Algorithms (London: Routledge, 2020); Bucher, If . . . Then.
11. Gilles Deleuze and Félix Guattari, A Thousand Plateaus: Capitalism and Schizo-
phrenia (Minneapolis: University of Minnesota Press, 1987).
12. Susan Leigh Star and Geoffrey Bowker, “How to Infrastructure,” in Handbook of
New Media: Social Shaping and Consequences of ICTs, ed. Leah A. Lievrouw and Sonia
Livingstone (London: Sage, 2004), 230.
13. Seaver, “Algorithms as Culture.”
14. Bucher, If . . . Then, 63.
15. Ibid., 72.
16. Benjamin Burroughs, “House of Netflix: Streaming Media and Digital Lore,” Pop-
ular Communication 17, no. 1 (2019): 1–17.
17. Bucher, If . . . Then; Julia Velkova and Anne Kaun, “Algorithmic Resistance: Media
Practices and the Politics of Repair,” Information, Communication & Society 24, no. 4
(2019): 523–40.
18. Kitchin and Dodge, Code/Space.
19. Robert Kowalski, “Algorithm= Logic+ Control,” Communications of the ACM 22,
no. 7 (1979): 424–36, [Link]
20. T. Striphas, “Algorithmic Culture,” European Journal of Cultural Studies 18, no.
4–5 (2015): 404.
21. Ed Finn, What Algorithms Want: Imagination in the Age of Computing (Cambridge,
MA: MIT Press, 2017), 18.
22. Robert Seyfert and Jonathan Roberge, “What Are Algorithmic Cultures?” in Al-
gorithmic Cultures: Essays on Meaning, Performance, and New Technologies, ed. Robert
Seyfert and Jonathan Roberge (London: Routledge, 2018), 12.
23. Kitchin and Dodge, Code/Space, 30.
24. Van Dijck and Poell, “Understanding Social Media Logic”; G. Bolin and J. Anders-
son Schwarz, “Heuristics of the Algorithm: Big Data, User Interpretation, and Institu-
tional Translation,” Big Data & Society 2, no. 2 (2015); Gillespie, “#trendingistrending:
When Algorithms Become Culture.”
25. Tarleton Gillespie, “The Relevance of Algorithms,” in Media Technologies : Essays
on Communication, Materiality, and Society, ed. Tarleton Gillespie, Pablo Boczkowski,
and Kirsten Foot (Cambridge, MA: MIT Press, 2014), 167–94; Tarleton Gillespie, Cus-
todians of the Internet: Platforms, Content Moderation, and the Hidden Decisions That
Shape Social Media (New Haven: Yale University Press, 2018); John Cheney-Lippold,
We Are Data: Algorithms and the Making of Our Digital Selves (New York: New York
University Press, 2017); David, “Envisioning the Power of Data Analytics,” Information,
Communication & Society 21, no. 3 (2018): 465–79; Bucher, If . . . Then.

This content downloaded from [Link] on Tue, 17 Mar 2026 18:48:30 UTC
All use subject to [Link]
 coding and encoding stre amed media . 93

26. Brian Larkin, “The Politics and Poetics of Infrastructure,” Annual Review of An-
thropology 42, no. 1 (2013): 328.
27. Susan Leigh Star and Karen Ruhleder, “Steps toward an Ecology of Infrastructure:
Design and Access for Large Information Spaces,” Information Systems Research 7, no. 1
(1996): 111–35; Susan Leigh Star, “The Ethnography of Infrastructure,” American Behav-
ioural Scientist 43, no. 3 (1999): 377–91; Geoffrey Bowker and Susan Leigh Star, Sorting
Things Out: Classification and Its Consequences (Cambridge, MA: MIT Press, 2000).
28. John Durham Peters, The Marvelous Clouds: Toward a Philosophy of Elemental
Media (Chicago: University of Chicago Press, 2015).
29. Lisa Parks and Nicole Starosielski, eds., Signal Traffic: Critical Studies of Media
Infrastructures (Urbana: University of Illinois Press, 2015); Aswin Punathambekar and
Sriram Mohan, eds., Global Digital Cultures: Perspectives from South Asia (Ann Arbor:
University of Michigan Press, 2019).
30. Star, “The Ethnography of Infrastructure,” 380.
31. Star and Ruhleder, “Steps toward an Ecology of Infrastructure,” 112.
32. Ibid.
33. David Beer, “Algorithms: Shaping Tastes and Manipulating the Circulations of
Popular Culture,” in Popular Culture and New Media: The Politics of Circulation (Bas-
ingstoke: Palgrave Macmillan, 2013), 69.
34. John MacCormick, Nine Algorithms That Changed the Future: The Ingenious Ideas
That Drive Today’s Computers (Princeton, NJ: Princeton University Press, 2012), 3.
35. Gillespie, “The Relevance of Algorithms”; Popular Culture and New Media.
36. Finn, What Algorithms Want, 23.
37. Bucher, If . . . Then, 22.
38. Gillespie, “#trendingistrending: When Algorithms Become Culture,” 53.
39. Gillespie, “The Relevance of Algorithms,” 183.
40. Jonathan Cohn, The Burden of Choice: Recommendations, Subversion, and Algo-
rithmic Culture (New Brunswick, NJ: Rutgers University Press, 2019).
41. Peters, The Marvelous Clouds; Bucher, If . . . Then.
42., Popular Culture and New Media.
43. Carlos Gomez-Uribe, “A Global Approach to Recommendations,” Netflix, February
17, 2016, [Link]
44. Burroughs, “House of Netflix.”
45. Netflix, “There’s Never Enough TV on Netflix,” Huffington Post, February 9, 2017,
[Link]
46. Cohn, The Burden of Choice.
47. Ibid., 26.
48. Ramon Lobato, Netflix Nations: The Geography of Digital Distribution (New York:
New York University Press, 2019).
49. Bucher, If . . . Then; Velkova and Kaun, “Algorithmic Resistance.”
50. Ignacio Siles, Johan Espinosa-Rojas, Adrián Naranjo, and María Fernanda Tristán,
“The Mutual Domestication of Users and Algorithmic Recommendations on Netflix,”
Communication, Culture, and Critique 12, no. 4 (2019): 499–518; Eriksson and Johans-
son, “‘Keep Smiling!’”

This content downloaded from [Link] on Tue, 17 Mar 2026 18:48:30 UTC
All use subject to [Link]
94 . sensing , auto mating , mediating

51. Cameron Johnson, “Goodbye Stars, Hello Thumbs,” Netflix Media, April 5, 2017,
[Link]
52. Tom Vanderbilt, “The Science behind the Netflix Algorithms That Decide What
You’ll Watch Next,” Wired, August 7, 2013, [Link]
-algorithm.
53. Xavier Amatriain and Justin Basilico, “Netflix Recommendations: Beyond the
5 Stars (Part 1),” Netflix Tech Blog, April 6, 2012, [Link]
-recommendations-beyond-the-5-stars-part-1–55838468f429.
54. Nick Lucchesi, “Netflix Says Its Algorithm Is Helping to Kill ‘Genre Bias,’” Inverse,
August 22, 2017, [Link]
55. Ana Andjelic, “Opinion: Why Taste Communities Are the Future of Marketing,” Adage,
January 16, 2020, [Link]
-are-future-marketing/2223986.
56. Nosheen Iqbal, “Film Fans See Red over Netflix ‘Targeted’ Posters for Black View-
ers,” The Guardian, October 20, 2018, [Link]
netflix-film-black-viewers-personalised-marketing-target.
57. Anthony Ha, “Netflix Is Experimenting with Different Episode Orders for ‘Love,
Death & Robots,’” TechCrunch, March 19, 2019, [Link]
love-death-robots-experiment/.
58. Iqbal, “Film Fans See Red over Netflix ‘Targeted’ Posters for Black Viewers”; Ha,
“Netflix Is Experimenting with Different Episode Orders for ‘Love, Death & Robots.’”
59. Ashok Chandrashekar, Fernando Amat, Justin Basilico, and Tony Jebara, “Artwork
Personalization at Netflix,” Netflix Tech Blog, December 8, 2018, [Link]
artwork-personalization-c589f074ad76.
60. Shalini Ramachandran and Joe Flint, “Hollywood versus the Algorithm at Netflix,”
Wall Street Journal, November 12, 2018, [Link]
-when-its-hollywood-vs-the-algorithm-1541826015.
61. Rachel McGrath, “Netflix Accused of Whitewashing by Nailed It! Host Nicole Byer,”
Huffington Post, May 29, 2019, [Link]
-whitewashing_uk_5cee50ebe4b0975ccf5dada5.
62. Star and Ruhleder, “Steps Toward an Ecology of Infrastructure.”
63. Velkova and Kaun, “Algorithmic Resistance.”
64. Seaver, “Algorithms as Culture,” 4–5.
65. Larkin, “The Politics and Poetics of Infrastructure,” 339.
66. Prey, “Nothing Personal.”
67. Gillespie, “The Relevance of Algorithms,” 187.
68. Louis Althusser, “Ideology and Ideological State Apparatuses (Notes towards an
Investigation),” in Lenin and Philosophy, and Other Essays (New York: Monthly Review
Press, 2001), 76–87.
69. Cheney-Lippold, We Are Data.
70. Cohn, The Burden of Choice, 89.
71. Velkova and Kaun, “Algorithmic Resistance”; Cohn, The Burden of Choice, 108.
72. Peters, The Marvelous Clouds, 15–16.
73. Larkin, “The Politics and Poetics of Infrastructure,” 335.

This content downloaded from [Link] on Tue, 17 Mar 2026 18:48:30 UTC
All use subject to [Link]