![The different aspects of "computerization" produce infrastructures that
"capture" (Agre 1994) an always larger number of practices, mediatizes
and "constitutes" (Burrows 2009, 451) them.
The "deep mediatization" (Couldry and Hepp 2016) that results is driven
by forms and function that flatten the differences between actors and
between types of content through standardized technical forms that are
exchanged in a very large market.
"In other words, by imposing a mathematically precise form upon previously unformalized
activities, capture standardizes those activities and their component elements and
thereby prepares them […] for an eventual transition to market-based relationships."
(Agre 1994, 120)
Computerization](https://image.slidesharecdn.com/riederlisbonfeb2019-191210133335/85/From-Algorithms-to-Diagrams-How-to-Study-Platforms-11-320.jpg)
![search
results for
[syria]](https://image.slidesharecdn.com/riederlisbonfeb2019-191210133335/85/From-Algorithms-to-Diagrams-How-to-Study-Platforms-20-320.jpg)
![[gamergate] avRBD 0.03
[syria] avRBD 0.47
[trump] avRBD 0.8](https://image.slidesharecdn.com/riederlisbonfeb2019-191210133335/85/From-Algorithms-to-Diagrams-How-to-Study-Platforms-24-320.jpg)
![After the phase of "flaunted neutrality", begins
that of "humanist values".
« [W]e're making a major change to how we build Facebook.
I'm changing the goal I give our product teams from focusing
on helping you find relevant content to helping you have
more meaningful social interactions. […] Now, I want to be
clear: by making these changes, I expect the time people
spend on Facebook and some measures of engagement will
go down. But I also expect the time you do spend on
Facebook will be more valuable. And if we do the right thing, I
believe that will be good for our community and our business
over the long term too. » (Zuckerberg 2018)
These normative questions are certainly not
simple.
5) Policies and "values"](https://image.slidesharecdn.com/riederlisbonfeb2019-191210133335/85/From-Algorithms-to-Diagrams-How-to-Study-Platforms-52-320.jpg)
From Algorithms to Diagrams: How to Study Platforms?
- 1. From Algorithms to Diagrams Bernhard Rieder Mediastudies Department Universiteit van Amsterdam Lisbon, February 28, 2019 How to Study Platforms?
- 2. Introduction Over the last twenty years, online "platforms" have been discussed intensely – under different names and from different angles. From search engines as "metamedia" (Winkler 1997) to the "web as platform" (O’Reilly 2005), to "walled gardens" (Zittrain 2008), "two- sided markets" (Rochet & Tirole 2003) or "custodians of the Internet" (Gillespie 2018), different functions of intermediation and their implications have been highlighted. Today, debates focus often on a limited number of large companies (GAFA, FANG, …), specialized in connecting (different groups of actors). The notion of "platformization" addresses the emergence of a "dominant economic and infrastructural model" (Helmond 2015).
- 3. Introduction The algorithms implicated in making connections are the last subject/aspect under scrutiny, particularly in the context of "infomediation" (Rebillard et Smyrnaios 2010). Words like distortion, opacity, fragmentation or polarization mark critiques and plurality, transparence, or accountability are proposed as solutions. Some even talk about domination and antitrust measures.
- 5. "New operators such as Google, Microsoft, Yahoo! and Apple, as well as the new, rising social media firms, such as Facebook or Twitter, should by now be included in the list of the most powerful media organisations worldwide." (Centre for Media Pluralism and Media Freedom 2013)
- 6. Introduction Different disciplines approach the question of the power of platforms in different ways and their conceptual and methodological differences mix with normative disagreement. In my own work, the technical dimension is central: how to think and study technical objects as "technologies of power" (Foucault 1994)? This includes historical and conceptual investigation of algorithmic techniques, the creation of digital tools for researchers, empirical research, questions concerning (software) design, et the study at the interstice of technology, politics, and economics. This presentation draws on these different lines of investigation.
- 7. The articulation focuses on a particular platform: YouTube. Introduction YouTube is at the center of a "hybrid media system" (Chadwick 2013) or a "new screen ecology" (Cunningham et al. 2016).
- 9. From algorithms to diagrams The goal is not to do a "platform biography" (Burgess 2016), but rather to retrace a "platform diagram". The term can be read in a technical sense: "At a more mechanical level, a platform is also a standardized diagram or technology." (Bratton 2015, 44) In Foucault's work, we find the concept of the diagram as a tool to think the connection between heterogeneous elements – between "discourses and architectures" (Foucault 1975, 276), between "programs and mechanisms" (Deleuze 1984, 46). These elements are not the emanation of a same logic, but rather an arrangement of "parts" that function as a whole.
- 10. platform (e.g. Facebook, Uber, App Stores, etc.) side 1 (e.g. users) side 2 (e.g. advertisers, sellers, etc.) platform-enabled transaction facilitates transaction by supporting offer, search, security, contracting, payment, etc. end-users YouTube (owned by Alphabet Inc.) advertisers content creators interfaces, ToS, etc.interfaces, ToS, etc.
- 11. The different aspects of "computerization" produce infrastructures that "capture" (Agre 1994) an always larger number of practices, mediatizes and "constitutes" (Burrows 2009, 451) them. The "deep mediatization" (Couldry and Hepp 2016) that results is driven by forms and function that flatten the differences between actors and between types of content through standardized technical forms that are exchanged in a very large market. "In other words, by imposing a mathematically precise form upon previously unformalized activities, capture standardizes those activities and their component elements and thereby prepares them […] for an eventual transition to market-based relationships." (Agre 1994, 120) Computerization
- 12. Building on the theory of Coase (1937) on the "nature of the firm", Ciborra develops an argument concerning transaction cost: "The costs of organizing, i.e. costs of coordination and control, are decreased by information technology which can streamline all or part of the information processing required in carrying out an exchange: information to search for partners, to develop a contract, to control the behavior of the parties during contract execution and so on." (Ciborra 1985, 63) The forms and functions that perform this "organizing" are constructed and imply many instances of decision-making and design. The mass of objects and context on offer in these very large markets seems both to demand and to valorize the delegation of sorting to algorithmic techniques that receive considerable power. Computerization
- 13. Institutional forms and governance Platforms operate as techno-institutional forms that combine decentralization (market) and centralization (state): "Platforms can be based on the global distribution of Interfaces and Users, and in this, platforms resemble markets. At the same time, the programmed coordination of that distribution reinforces their governance of the interactions that are exchanged and capitalized through them, and for this, platforms resemble states." (Bratton 2015, 41) How do distribution and governance operate, how to identify and describe the factors or "causes" behind actually observable phenomena or "outcomes"?
- 14. A platform diagram should allow for discussing together and juxtaposing elements that are often treated separately: The elements of this tentative diagram: 1) "Algorithms" 2) Constructed infrastructures 3) Participants, practices, and contents 4) Business models 5) Policies and "values" 6) The company and its environment Each element points toward many possibilities to study platforms: it highlights particular questions, specific approaches and methods, as well as specific historical trajectories. The YouTube diagram
- 16. The question of the power of algorithms is not new, but is now appears with urgency, in particular since machine learning has started to spread. The power often attributed to algorithms fuels demands for transparency and accountability: but the problem is complicated. Scholars in the social and human sciences are proposing empirical (Sandvig et al. 2014; Diakopoulos 2014) as well as conceptual approaches (Burrell 2016; Mackenzie 2015, 2017) to capture how algorithms operate or "think". 1) "Algorithms"
- 17. Experiments (here: Kosinski et al. 2013) show how e.g. Facebook "likes" can predict intimate or sensitive variables rather well. By specifying a target variable such as "time on site" or "click probability", these techniques represent an "interested" reading of reality (Rieder 2017). In this context, "the goal is not truth, but performativity, that is, the best input/output relation" (Lyotard 1997). 1) Thinking machine learning
- 18. Instead of specifying a model as a set formula, machine learning allows for deriving a model from specific "orchestrations" of feedback. Every signal receives a significance in relation to a target variable that defines a desired outcome. The model is, at the same time, complex (many variables and relationships between them) and dynamic (it changes in response to captured feedback).
- 19. Different initiatives for empirical observation have been appearing, here AlgoTransparency by Guillaume Chaslot.
- 22. YouTube Data Tools (Rieder 2015)
- 24. [gamergate] avRBD 0.03 [syria] avRBD 0.47 [trump] avRBD 0.8
- 25. We identified three types of "morphology": ⦿ stable over long periods of time (low avRBD); ⦿ stable with "newsy" interruptions (average avRBD); ⦿ "newsy" queries that change constantly (high avRDB); Stable periods are often organized around "explainer"-videos published by channels that present themselves as neutral or around well-known US actors (e.g. Stephen Colbert); during more agitated periods, "native" actors intervene more often. YouTube's "platform vernacular" (Gibbs et al. 2015) is clearly important, but intersects with the subjects sitting behind the queries, making each case different from the others. 1) YouTube, studied empirically
- 26. We can detect correlation between video production, search volume (via Google Trends), and the level of change; but a clear picture of causality remains elusive. 1) YouTube, studied empirically
- 27. Our approach is interested in the technical operation of search on YouTube, but situates this operation in a large system that is heavily affected by use practices. We move from ranking algorithms to ranking "cultures" and combine different qualitative and quantitative methods in a gesture of "descriptive assemblage" (Savage 2009). Technical factors, platform vernaculars and the particularities of individual subject are taken into account. Technical and non-technical factors, the "platform" and the "practices" become impossible to distinguish. 1) Beyond "algorithms"
- 28. "Concepts like ‘algorithm’ have become sloppy shorthands, slang terms for the act of mistaking multipart complex systems for simple, singular ones." (Bogost 2015, n.p.) "Programmed coordination" (Bratton 2015) is not limited to ordering algorithms, but includes a large set of forms and functions that operate on the level of interfaces or in the bowels of backends. Empirical approaches like the "walkthrough method" (Light et al. 2017) can take stock of affordances that link up to form "grammars of action" (Agre 1994), in order to understand, e.g. on Facebook, "the specific ways in which sociality is programmed (i.e., encoded, assembled, and organized)" (Bucher 2013, 480). The study of data flows, their "compression" into metrics and their operational uses constitute a second level of analysis. 2) Constructed infrastructures
- 29. end-users YouTube content creators search, recommendation, interaction advertisers interfaces interfaces interfaces Computerization implies that a large part of transaction are made through interfaces.
- 31. Interfaces can be seen as the creation of a communicative setting or infrastructure.
- 33. Platforms propose different "market information regimes" and "user information regimes" (Webster 2010). YouTube (Creator) Studio
- 34. Platforms are inhabited by participants, practices, contents and various relations between these elements; algorithms and infrastructures enter into dynamics with appropriations. Ethnographic studies provide interesting, but partial perspectives; the first attempts to describe YouTube on the whole (p.ex. Bärtl 2018) remain sketches – that confirm, however, familiar elements such as the strong presence of a "Matthew effect". Previously often described using the term "participatory culture" (Jenkins et al. 2015), we can now observe a broad variety of actors and vernaculars as well as important scale variation. The proliferation of "multi-channel networks" (MCN) introduces another layer of complexity. (cf. Lobato 2016) 3) Participants, practices, and contents
- 35. The techno-institutional format of the channel (and its subscription logic) is potentially the central "natively digital object" (Rogers 2013) that facilitates stabilization, professionalization, branding and audience accumulation. Channels render what is on offer readable and navigable. Who is on YouTube and why? Which themes or issues dominate? What "works"? What is the role of algorithms and of audience selection dynamics? Who will reign the new screen ecology? 3) Participants, practices, and contents
- 38. Network of "related channels" (YouTube Data Tools + Gephi) starting from Rubin Report and The Young Turks.
- 40. The already mentioned elements connect directly and deeply with the business models in use. YouTube's model still reposes in large part on advertisement, distributed through "keyword bidding" and shared (55%/45%) with more than a million "partners" (2016). The now classic mechanisms of "free" apply. 4) Business models
- 41. Certain actors (e.g. channels with more than 100k subscribers) have a right to "human contact".
- 43. Patreon has become a major source of financing.
- 44. Paid services such as YouTube Premium (previously RED) and YouTube Music remove ads and allow for additional uses. The join function integrates the Patreon model into the YouTube platform and sales function further broaden the set of monetization possibilities. 4) Business models
- 45. Can we automatically detect channels' business elements and intersect them with other metrics?
- 46. end-users YouTube (owned by Alphabet Inc.) advertisers content creators MCNs MCNs direct monetization Patreon
- 47. Justification discourses are (were?) often based on the idea of "revealed preference", which holds that "the individual guinea-pig, by his market behaviour, reveals his preference pattern" (Samuelson 1948). This fuels and justifies the use of feedback signals as "votes" and popularity measures as means of ordering. Due to commercial (nervous advertisers) and political pressure (public criticism, regulation menace), the situation has become much more complicated. The platform now has to engage in control and censorship beyond copyright (Content ID); we see a series of experiments that include new rules, controlled and policed by algorithms, editors and gamified users. 5) Policies and "values"
- 48. Demonetization has become YouTube's most visible instrument of discipline and governance.
- 49. Jack Dorsey and Alex Jones at the exit of the "Twitter- Facebook Senate hearings" in September 2018.
- 50. How can the effects of these interventions be studied empirically?
- 52. After the phase of "flaunted neutrality", begins that of "humanist values". « [W]e're making a major change to how we build Facebook. I'm changing the goal I give our product teams from focusing on helping you find relevant content to helping you have more meaningful social interactions. […] Now, I want to be clear: by making these changes, I expect the time people spend on Facebook and some measures of engagement will go down. But I also expect the time you do spend on Facebook will be more valuable. And if we do the right thing, I believe that will be good for our community and our business over the long term too. » (Zuckerberg 2018) These normative questions are certainly not simple. 5) Policies and "values"
- 53. Policies now include considerable financial initiatives.
- 54. Regulation and taxation attempts are picking up speed.
- 55. The large platforms follow a strategy of "concentric diversification", "moving into activities which mesh to some degree with the present product line, technological expertise, customer base, or distribution channels." (Thompson & Strickland 1978) 6) The company and its environment Do the synergies thus liberated favor the appearance and stabilization of cross-market oligopolies and monopolies? Core: established business, mastery of process and product, steady revenues; Extension layer: mastery of process and product, but not yet established and generating significant revenue; Expansion layer: experimental process and product, "competence testing";
- 56. YouTube is part of the Alphabet / Google product family and entertains various relationships with other products. How to study these relationships? What can resist the aspiration power of platforms? Large brands? Public service? Other platforms? Who are YouTube's competitors? Facebook + Instagram? Amazon + Twitch? Netflix? Disney? Traditional TV networks? Google Search 1998 2000 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 AdWords Google News Beta AdSense GmailIPO Picasa Google Maps YouTube Android Google Chrome AdMob Freebase Google Wallet Hangouts Play Store Android Wear Android Auto Google Print DoubleClick Google X Chrome OS Google+ Google Drive Google Cloud Platform Google Now 6) The company and its environment
- 57. Conclusions Platforms represent a reconfiguration of transaction modalities, following a "digital" mode, including these elements: ⦿ Normalization and standardization that facilitate the organization of transactions in market forms; a "flattening" of certain cultural differences through the notion of "content" and the extension of this principle to many different domains; ⦿ Forms of algorithmic coordination that automate continuous, adaptive and interested optimization and integrate it into infrastructures; ⦿ (Modulable) interfaces that capture participants, practices, and data, serving as coordinators and translators between participants and the techno-institutional modes of platforms; ⦿ Very large number of participants and contents that are constantly producing value for the platform by appropriating and feeding the forms and functions; ⦿ Business models that stimulate the production of content while broadly keeping the logic of "free" for users; ⦿ Policies, values, and justification discourses that seek to protect the model; ⦿ Modes of diversification and competition that favors the emergence of monopolies and cross-market domination;
- 58. Conclusions The emergent diagram describes a model that is deeply expansionist and feeds on the dialectics between decentralization and centralization. Platforms are "transversal network effect machines" that advance on a "North California" (Cunningham et al. 2016) model: experimentation, iteration, risk taking, etc. We under-estimate the complexity of the "platform problem" and the enormous energies liberated by the model if we treat it as a product of capitalism "as usual". Political responses to the challenges of platforms depends on the clarity of analyses and our capacity to thing different elements together.