Posted in Essays, Media Theory, Politics on November 14, 2020 18 min read
Introduction
The pillar of all hitherto existing societies has been the acquisition and distribution of knowledge. The most fundamental always remained the impressions of our senses and various technologies of distribution. Uncountable such technologies have been used during humanities history: Speech, Tales, Quips, Telegraphs, Writing, Typography, Hieroglyphs, Music, Magazines, Radio, Homing Pigeons, Television, Smoke Signals, and the Internet. Each with its own logic in structuring knowledge and disseminating it. Nevertheless, the individual still had to make sense of them on the basis of its sensual impressions. In 1989, media scholar Marshall McLuhan disavowed the idea of a separation between the techne and the senses for we merely develop the techne to make certain aspects of the real world intelligible to our senses. In effect, he proposed we have merely extended our senses to grasp a different, potentially more useful, part of reality. Perhaps to bend it to our will? No extension superior, but certainly different with its “its own principles and lines of force” (McLuhan, 1964, p. 6) in its preparation of knowledge.
When the blinding man acquires the gift of enhanced hearing as a consequence of his deteriorating sight. It is not his ears that have improved. Instead, his shifting abilities of perception have influenced his “whole psychic and social complex” (McLuhan, 1968, p.11) to condense more knowledge from the auditory world. In our case equally a change in our senses constitutes a change in our knowledge acquisition, and need it really be said that: knowledge guides action?
Tell me who gives up their sight willingly? Who wants not to forget, but never know? No musician in their will to excellence, has given up their sense of sight. Not even the most devoted of music lovers will trade their eyes for a heightened enjoyment of a melody. For when we develop a sense, we develop a need to use it (McLuhan, 1964, p. 15).
Turn it around. Give a myopian man glasses. Extend his vision, and he will be transfixed by the colours, light, shade & shapes he can now perceive. He will not mull over the lost contrast and splendour in his music. McLuhan proposed that we are no different, we too become transfixed with new extensions of our senses. Overwhelmed with a new medium, the social body turns “numb, deaf, blind, and mute” (McLuhan, 1964, p. 8) to the changes that had occurred within it. Yet, we ought to question, what have we lost, what did it cost, what has atrophied in our hypnosis?
McLuhan thought we had not even grappled with the effects of the televisual medium itself, which was merely hosting content scheduled along the work life cycles of consumers. Meanwhile, our brightest minds have come up with a new technology, already inserted into the social body . Silently, the matters of deciding what should be perceived, what constitutes knowledge, and which of it is worthy to acquire, once the endeavour of experts and journalistic institutions has been dethroned by the algorithm. The supposedly most unbiased of technologies, therefore best suited to navigating the wealth of public knowledge (Gillespie, 2014, p. 191).
The most impactful of this kind may be YouTube recommendation algorithm, in connection with the platform itself it is used by more than a billion people per day (About YouTube, n.d.). From its beginning it has put us under its spell, and we are much too late to scrutinize its effects on the social body. To give an impetus towards such an endeavour is the aspiration of this essay.
From Base to Superstructure
I believe that any enquiry into the knowledge logic and consequently social impact of a media must not look past the circumstances of its creation. For Marx, the economic base of a society acted as a constraint on certain developments whilst enabling others. People create culture and technology, but not necessarily under conditions of their own volition. Indeed, the creators of YouTube built the platform on the back of a medium that had only formed because it was suitable to satisfying market needs – the Internet. The founders’ idea was simply to create an online video sharing site for dating purposes, but after a lack of uploads, they decided to allow videos of any type (Dredge, 2016). Even then the platform remained inconsequential concerning people’s daily conduct and media consumption. Then, how can we explain it becoming into a media giant?
The gist of the matter is that the social significance of an invention within a capitalist system, usually only occurs after it has been selected for investment and production (Williams, 1989, p. 120). Following Williams pointed finger, we may see that it was venture capitalist’s investment into the platform that both lifted into social significance but simultaneously infected it with a “perspective of capitalist reproduction” (Williams, 1980, p. 59; Helft, 2006). But if capital is to be reproduced an opportunity for profit is needed, luckily the progress of technology had supplied just that.
Algorithms and Profit from the Long Tail
The emergence of YouTube was only possible because of progress in digital technologies and digital storage capabilities. It was these that opened up the long tail market.
To remain profitable, a retail store must always be certain that any goods it carries, “can generate sufficient demand to earn [their] keep” (Andersen, 2018). Yet, demand can only extend as far as the limited local population, and therefore the store will only sell so called “hits”. Products or content that generate a lot of demand, often precisely because of their generalized and inoffensive nature. Yet, for a Content Distribution Network such as YouTube a different economic structure applies. The costs of storing a video on a server had become relatively inexpensive, and the emergence of the internet meant that demand wasn’t limited to a local population but could extend as far as national and international audiences. Precisely because of this even relatively obscure and specialized content could generate profit, as with the larger audience it became easier to reach the critical threshold of sales needed to drive profit. Not withholding that the low storage costs of digital media, made it relatively cheap to keep an almost unlimited amount of content. This obscure and specialized content when taken as a whole, far outmatched the “hits” in terms of total demand and therefore was a sum total much more profitable business. The remaining problem was that consumers only cared about a certain range of this obscure content and were heavily disinterested in the rest (Andersen, 2018). The most inexpensive solution to match consumers with their preferred segment of long tail was the recommendation algorithm. A technology much faster than humans at navigating the wealth of content available on a platform. On this economic foundation it made sense for a platform to allow the free uploading of all sorts of content that could then be explored without restrictions by anyone. Fast forward to today and we can revel in all kinds of content that our heart desires. YouTube is endless and has something for everyone: I can watch my favourite Fortnite videos, get advice on how to up my dating game, tune into the life of vloggers that share my experiences, and watch lecturers unravel the workings of capitalism.
But, waking up from exactly this trance induced by the new medium is imperative. As McLuhan states: it the bias of the operators and consumers of a media, to be concerned about its content. The owners or investors themselves are ‘are much more concerned about the media as such’ (McLuhan, 1968, p. 62).
Thus, while YouTube may state and cultivate the image that its mission is to give “everyone a voice” and that it believes that “people – not gatekeepers [should] decide what’s popular” (About YouTube, n.d.). It is merely a marketing ploy within which it continually has to emphasize the neutrality of its algorithms. Yet, inevitably, behind any “cold” algorithm there lie “warm institutional choices” (Gillespie, 2014, p. 169). For example, deciding which datapoints are isolated, its structure, and the specified outcome are all choices made by humans, and in the case of YouTube again we must suspect under a “perspective of capitalist reproduction” (Williams, 1980, p. 59). So while most of us are still entranced with the wealth of content, we must avert a critical eye to the medium itself.
The YouTube Recommendation Algorithm.
Unpacking the warm choices behind the algorithm is hard in itself, if not for the fact that the YouTube Recommendation Algorithm is confidential and inaccessible to the public, except for a singular eight page paper released by Google in 2016, which alludes to its inner processes but refrains from giving too much insight.
Principally, it is useful that the paper states that the ‘desired objective function’, or end-result of evaluation, of the recommendation algorithm is “expected watch time” (Covington et al., 2016, p. 8). This function is sensible for an advertising corporation like YouTube since it maximizes the time users spend on its platform, consequently they see more ads and the company generates more profit.
Turning to the question of how the algorithm functions the paper states that it is separated into two parts. The first one, a candidate generation network, which notably only relies on a “broad personalization” via collaborative filtering using “coarse features such as […] video watches, search query tokens and demographics.” (Covington et al., 2016, p. 2). To reiterate this indicates that the algorithm selects videos from a corpus of billion, merely on the basis of rough demographic features and past watches. The use of collaborative filtering insinuates that videos will be recommended to you, on the basis of similarity to other users and their preferences i.e. age, gender, area of interest. It is then assumed, that if some of your previous datapoints align your other preferences will equally match. Although, there is explicit feedback mechanisms on YouTube such as thumbs up and down and flagging of videos, the paper states that these are excluded from the algorithm (Covington et al., 2016, p. 2).
After this first stage of candidate generation, which proposes “hundreds” of possible videos, the algorithm enters a second stage where-in it uses more datapoints to consider the user at hand and the previously generated videos to evaluate which videos lead the user to the longest expected watch time. Consequently, these are presented to the user on the website (Covington et al., 2016, p.2).
Encoding/Decoding
In order to judge the effects of the Recommendation Algorithm on the social body, I will rely on a theoretical framework of a media that is reasonably close to YouTube, that of Television. In “Encoding and Decoding in the Television Discourse” Stuart Hall offered a theoretical approach of message circulation for that medium. I believe it reasonable to lean on the four points in the production process he isolated to structure my account of changes that heed from the YouTube recommendation algorithm versus Television. To summarize: Hall thinks that the four stages are Production, Circulation, Use, and Reproduction. In production a message is encoded by media producers drawing upon a society’s dominant ideology and certain televisual production structures. Circulation concerns the way audience members receive the message. Use is the stage at which a message is being decoded by an audience which can happen in either a dominant-hegemonic stance, in which they accept the meaning of the message. A negotiated stance, where the message is received but adapted to local conditions. An oppositional decoding, which concerns a complete de-totalization of the message in its preferred framework and re-totalization within an alternative framework of reference (Hall, 2006, p. 508-511). Conceptualizing YouTube as a sense extension different from Television leads to the following implications on the social body:
A Sense that Maximizes its Use (Consumption)
As we have seen the YouTube recommendation algorithms function is to maximize the use time of the platform. Using McLuhan’s terms, this would make it an extension of the senses whose objective is to increase its own usage. To this end, it uses continuous A/B testing and shifting weights inside the algorithm to ensure that the viewing pleasure of the subject is maintained above a certain threshold to keep it engaged with the content. Consequently, the algorithm skews the sense-ratio of an individual towards itself. While the individual’s explicit valuations of content are ignored in the calculation. I think the problem is most succinctly summarized when we consider the fact that between two CPR instruction videos the recommendation algorithm would rank the longer one higher. It’s own usage is more important than the value it provides.
Other Media Change in the Pursuit of Increased Watch Time (Production)
“The content of a medium is always another medium” (McLuhan, 1964, p. 1)
Just like the algorithm is sensitive to the content created by producers, so are they sensitive to and aware of the algorithm and create their content accordingly (Gillespie, 2014, p.183). Content creators notice that certain qualities of videos i.e. duration, thumbnails, patterns of aural-visual presentation are more extensively promoted by the algorithm and are therefore incentivized to adopt these qualities. Moreover, they are make their content ‘algorithmically recognizable’ (Gillespie, 2014, p. 184), to be more widely distributed by the algorithm. For example, the use of captions may help those with hearing issues, but captions also form a pattern of data legible for the algorithm to categorize data. To summarize whether we call it the medium like McLuhan or encodings for Stuart Hall of messages that align with the algorithms goal become more prominent on the platform, and possibly even in other media. So far, we may notice the different patterns of speech on, presentation that are entirely YouTube specific due to the existence of the recommendation algorithm.
The Algorithm Invites us to Become its Pre-figured Identities (Circulation)
As seen in the candidate generation stage of the YouTube recommendation algorithm, it does not concern itself with perfect predictionof our interests, but merely sufficient approximation as it generates content only according to “broad personalizations”. This means that our algorithmic identity only ever approximates our real identity, with the data points available to the algorithm, yet directed towards purposes of capital reproduction. Therefore, the platform is content to offer us content according to pre-figured categories constructed along its commercial interests. Consider, the experience of a new user to the platform. The user arrives to it with a certain set of interests and markers of identity. Yet, for candidate generation they must be assorted to one of the pre-existing algorithmic categories of identities. As aforementioned, the latter is only a sufficient approximation of the real identity, and consequently so is the recommended content. Recurring exposure to content aligning with the interests of our sufficiently approximated identity invites us to become that caricature of ourselves directed towards advertising needs.
The Algorithm Constructs the Publics (Circulation)
Landsberg argues that the cultural hit economy and its mass cultural representations are both commodified, but also construct a shared past across racial and socio-economic lines. This means they can serve as a ‘prosthetic memory’, a common framework and shared public experience that serves as a starting point for public discourse, a public past that transcends backgrounds, races, places, and classes. These shared memories then become the basis of identification as a collective and may produce counterhegemonic public spheres. (Landsberg, 2018, p. 149).
This, however, cannot apply to YouTube’s recommendation algorithm which navigates the long tail. Here, the warm choices in the algorithm determine when publics form, how they interact, and when they ought to fall apart, all on the basis of expected watch time (Gillespie, 2014, p. 188). This implies, that meaningful disagreement within a public could be stifled at the point that the algorithm determines that the differences within that public merit different content recommendations. The danger lies in the fact that, due to the different content presented to each user a common ground of public dialogue disappears, and barriers of understanding arise between subjects that otherwise share a physicality or class.
The Algorithm Minimizes Oppositional Decoding (Consumption)
Assuming that an oppositional reading of content decreases engagement and therefore watch time, the algorithm will adapt and cease recommending content encoded in that particular way. In effect, it will aim to match content creators and viewers based on their shared frameworks of meaning and totalized views of the world. Therefore, the subject will to a large extent not receive any content offensive to its mental horizon and mostly consume apolitical content with reference to its already held opinions. This means that the algorithm itself aims to minimize oppositional and negotiated readings within the subject. Persons may still position themselves in opposition to the dominant hegemonic ideology as such, but they themselves will rarely decode a message in an oppositional way. While the content of their recommendations may well oppose the hegemonic ideology as such, the producer and consumer are matched in such a manner that the message will be read as intended by the producer.
The YouTube recommendation algorithm is but a reflection of the capitalist search for profit, which has most recently found it in supplying audiences with content from the long tail. Whilst the company may state that the goal of their algorithm is to maximize engagement or expected watch time, we may well say it is to get audiences addicted. This alone would not be as concerning as the effects it has on the backbone of political conduct as such. As we have seen, it changes our patterns of speech and how we perceive reality. It changes our identities and constructs public spheres along its commercial goals. At last, it may be responsible for a withering away of our capacity to oppose a hegemonic ideology for it minimizes the moments where oppositional and negotiated readings could happen, taking away our practice grounds. Its assault is on all of the pillars of our political social body at once. Yet, since we are still caught up in the spell of the content it lays at our feet, discourse around the perils of this newest sense extension is but faint and far away to most of us.
References
About YouTube – YouTube. (n.d.). Retrieved October 20, 2019, from https://www.youtube.com/about/.
Anderson, C. (2018, October 2). The Long Tail. Retrieved October 20, 2019, from https://www.wired.com/2004/10/tail/.
Dredge, S. (2016, March 16). YouTube was meant to be a video-dating website. Retrieved October 20, 2019, from https://www.theguardian.com/technology/2016/mar/16/youtube-past-video-dating-website..
Covington, P., Adams, J., & Sargin, E. (2016). Deep Neural Networks for YouTube Recommendations. Proceedings of the 10th ACM Conference on Recommender Systems – RecSys 16. doi: 10.1145/2959100.2959190Gillespie, T., Boczkowski, P. J., & Foot, K. A. (2014). Media technologies: essays on communication, materiality, and society. Cambridge, MA: The MIT Press.
Hall, S. (2006). Encoding, Decoding. The cultural studies reader. (S. During, Ed.) (2nd ed.). London: Routledge
Helft, M. (2006, October 12). San Francisco Hedge Fund Invested in YouTube. Retrieved October 20, 2019, from https://www.nytimes.com/2006/10/12/technology/12hedges.html.
Landsberg, A. (2018). Prosthetic memory: the ethics and politics of memory in an age of mass culture. In Memory and popular film. Manchester University Press
Lister, M. (2009). New media: a critical introduction. London, NY: Routledge.
McLuhan, M. (1964). Understanding media: The extension of Man. London: Sphere Books.
McLuhan, M. (1968). Understanding Media. London: Sphere
Williams, R. (1980). Problems in materialism and culture: selected essays. London, England: Verso.
Williams, R. (1989). The Politics of Modernism, ed. Tony Pinkney (Vol. 65). London: Verso
Recent Comments