A new algorithmic imaginary

Abstract

The algorithmic imaginary as a theoretical concept has received increasing attention in recent years as it aims at users’ appropriation of algorithmic processes operating in opacity. But the concept originally only starts from the users’ point of view, while the processes on the platforms’ side are largely left out. In contrast, this paper argues that what is true for users is also valid for algorithmic processes and the designers behind. On the one hand, the algorithm imagines users’ future behavior via machine learning, which is supposed to predict all their future actions. On the other hand, the designers anticipate different actions that could potentially performed by users with every new implementation of features such as social media feeds. In order to bring into view this permanently reciprocal interplay coupled to the imaginary, in which not only the users are involved, I will argue for a more comprehensive and theoretically precise algorithmic imaginary referring to the theory of Cornelius Castoriadis. In such a perspective, an important contribution can be formulated for a theory of social media platforms that goes beyond praxeocentrism or structural determinism.

Introduction

Algorithms are a central organizing principle of social media platforms. Since 2011, when Facebook, following the introduction of the Like button, organized its News Feed according to such algorithmic organizational principles, a spread of algorithmic feeds can be observed across almost all commercial social media platforms. For example, both Twitter (in late 2015) and Instagram (in early 2016) have aligned their timeline and explore feed according to algorithmically curated content based on the parameters of popularity and relevance.

For research on social media platforms, this circumstance means that methodological-theoretical approaches are urgently needed that, on the one hand, reach beyond a currently prevailing practice-theoretical and media-ethnographic paradigm of describing single platforms (e.g., Leaver et al., 2020; Miller, 2011), but on the other hand, do not get lost in structural determinisms that are often shaped by cultural criticism (e.g., Fuchs, 2014; Zuboff, 2019). Where the former usually provide very precise descriptions of user practices, but in doing so rarely allow for theoretically robust conclusions beyond the described social media logics of the respective platforms, the latter engage in an ideology critique often sweepingly aligned with the concept of “production” or deal with an exploitation of data subjects in surveillance or platform capitalism (Srnicek, 2017; Zuboff, 2019), but without really being able to say anything about the actual practices of users.

This paper aims to address this tension by taking the widely received concept of “algorithmic imaginaries” (Bucher, 2018) as a point of departure and tries to think theoretically further with the help of Castoriadis’ theory of the imaginary. By conceptualizing the algorithmic imaginary in this way, it is not only possible to take a look at the permanently interrelated interplay between users and algorithms as well as practices and platform infrastructure, rather it becomes possible to simultaneously take a theoretical scope at the cross-platform institutionalization of infrastructure, as is characteristic of the spread of social media feeds in recent years.

The algorithmic imaginary and its significance for a theory of social media platforms

The concept of the “algorithmic imaginary” (Bucher, 2018) has received attention in discourses around social media research, as it brings into focus for the first time users’ appropriations of algorithmic processes operating in opacity and their imaginaries of these operations. As such, the concept has been highly influential in media and communication studies, triggering approaches to “decoding” (Lomborg and Kapsch, 2020) and “understanding” or “interpreting” (Andersen, 2020) algorithms from a users’ perspective. Moreover, a very similar discussion is taking place in the field of human–computer interaction under the notion of “folk theories” (DeVito et al., 2018). Bucher (2018) describes this concept as

[. . .] ways of thinking about what algorithms are, what they should be, how they function, and what these imaginations, in turn, make possible. While there is no way of knowing for sure how algorithms work, the personal algorithm stories illuminate how knowing algorithms might involve other kinds of registers than code. [. . .]In other words, the algorithmic imaginary emerges in the public’s beliefs, experiences, and expectations of what an algorithm is and should be. (p. 113f)

Thus, this approach is primarily concerned with the users’ perspective, which indeed marks an important starting point for a comprehensive theoretical understanding of social media platforms, because without a precise description of the users’ practices and imaginaries of how social media platforms work, one would inevitably fall into a (structural determinist) reductionism in one way or another. However, in this perspective, the processes and practices on the side of the platforms and their developers are largely left out, which would already make a precise theorization of social media platforms impossible on a micro level.

Therefore, it is important to include the algorithmic processes, which are commonly situated in the so-called “backend,” that is, the invisible part of the interface, in a concept of the algorithmic imaginary. Here, too, the algorithm “imagines” the future behavior of the users via so-called “predictor modules” of machine learning, which are supposed to predict the future behavior of the users from all their actions. Although it must be noted that the algorithm does not “imagine” in the human sense and the metaphorical part must always be taken into account, it is worth extending the concept of the imaginary to non-human entities. The algorithm not only computes according to predefined parameters, but also constantly plays through transforming and supposedly fitting models in the backend depending on the user’s behavior. In this respect, the approach of Fisher and Mehozay (2019), according to which the algorithm on the one hand observes the behavior of the users and derives imaginaries from this, but on the other hand the designers and developers also rely on “imaginary interlocutors” (Fisher and Mehozay, 2019: 1179), must be extended to include the user perspective, just as conversely the perspective of the developers and algorithms must be integrated in Bucher’s concept of an algorithmic imaginary.

Indeed, the users’ imaginaries are concretely part of the platform infrastructure and precisely because of this, the algorithmic imaginaries also have an effect on the behavior of those very users via the loop of the backend and change it. In this regard, we are already dealing with a permanently interrelated interplay on the micro level, coupled to the imaginary, which makes it necessary to conceive a theoretically well-founded algorithmic imaginary that includes both sides – algorithms and users. Nevertheless, there are instituted principles that emerge within designers’ imaginaries, and in doing so, build on historically established algorithmic techniques (Rieder, 2020). On this basis, an important part can be formulated for a theory of social media platforms that is neither exhausted in praxeocentrism nor in structural determinism, but is able to analyze in particular the cross-platform institutionalization processes that paradigmatically characterize the proliferation of social media feeds.

Attempt of a symmetry between “design imaginaries” and “algorithmic imaginaries”

Recent scholarly appropriations of an imaginary in the field of social media research rely almost exclusively on science and technology studies, in line with the so-called “practice turn” (Schüttpelz et al., 2021).

Significantly, in this context Lucy Suchman, following Keith Grint and Steve Woolgar, has already pointed out that “design imaginaries” (Suchman, 2007: 187–205) on the part of the developers play a central role in the “user configuration” (Grint and Woolgar, 1997: 92) in the course of technology developments, which is concretely reflected in the development of software in the anticipation of potential user behavior in “mental models” (Greca and Moreira, 2000). This means that users and their anticipated use of technologies already play a central role at the software development level in the form of imaginings that designers and programmers make of them (Fisher and Mehozay, 2019).

With reference to social media platforms, which are constantly changing and in a state of constant “flux” due to the permanently interrelated interplay of modes of use and platform infrastructure, so-called A/B tests are carried out not least for this reason before each introduction of functions. This applies equally to algorithmic feeds as well as to other functions now implemented across platforms, such as the story format originally introduced by Snapchat, which can be traced back to the practice of selfie photography. In order to minimize the proportion of the imaginary that can never be completely resolved into certainty in relation to the users (not least for economic reasons), the platforms regularly carry out such tests when introducing functions.

This approach shows that it is not only the behavior of users that is changed by the technologies. With regard to the algorithmized feeds, depending on the different modes of use, fundamental parameters within these technologies themselves also change situationally, as they are fed by constantly changing data layers. After all, each feed looks different depending on the usage of the respective user and the “predictor modules” played out.

Despite this permanent interplay between algorithms and users, the feeds of social media platforms fall back on institutionalized algorithmic principles, which are largely based on the reputation principle set by Google with PageRank and its “authority rank” (Kleinberg, 1999). With PageRank, for the first time the status (or with regard to the topology of the network: the centrality) of the respective page was prominently included in the ranking (cf. Brin and Page, 1998) and thus prestige or authority was attributed to certain pages. These procedures can be traced back to the origins of graph theory and sociometry (Rieder, 2020: 81–141), as behind the calculations and imaginations of the algorithmized feeds are “social graphs,” that is, the mapping of the relationship networks of monodirectional (“followings”) or bidirectional (“friendships”) connections between users, which becomes tangible, for example, in friendship suggestions on Facebook (“people you might know”). On the same basis, the prestige of different users within social media platforms is calculated, where, in a very similar way, a like of a user with many followers tends to be worth more than a like of a user with few followers (Constine, 2018).

In short, the feeds of social media platforms are based on institutionalized “algorithmic techniques” (Rieder, 2020) that can be traced back to the beginnings of graph theory. At the same time, on the platforms’ side, there is an indeterminacy deeply rooted in the technology that feeds on the intertwining of practices and infrastructures. This is precisely why imaginings are necessary both on the part of the developers about possible ways of using the technology by the users (regarding to the implementation of new functions) and on the algorithmic level (regarding to the playout of suitable content). Of course, it is by no means the case that a newly introduced (algorithmic) technology or function always leads to the way users use it that the developers had previously anticipated, although certain incentives can be set, for example in the interface design (such as push notifications) to exert an influence. Conversely, it is also not the case that users would entirely determine the algorithms of the feeds, although they may try to influence the algorithm by selectively giving likes.

While the latter practices stem from users’ ideas about how the algorithms work and correspond quite closely to what Bucher meant by her suggestion of an “algorithmic imaginary,” it is nonetheless essential to consider the other side of the “design imaginaries” and algorithms for a comprehensive understanding of how social media platforms work, without playing one side off against the other. The question of how institutionalized algorithmic techniques and a constantly changing infrastructure relate to each other as a result of the permanently interrelated interplay between algorithms and users remains open. And it is precisely this problem that needs to be addressed theoretically.

Bucher’s concept, which focuses on the imaginaries of the users, provides a useful starting point, which, however, has to be extended by the perspective of the imaginaries of developers and algorithms. In the following, the notion of the “interface” will be used for this purpose, because the imaginaries of both sides, users and developers or algorithms, converge on the level of interfaces. The interface thus represents a “threshold,” which not only connects the “design imaginaries” of the developers with the algorithmic imaginary of the users and is thereby constantly and potentially infinitely traversed during platform use (Schulz and Matzner, 2020). Even more, starting from this microperspectively conceived feed-interface relation, it can also be scaled to the structural or institutionalized level and thus trace how the macroperspectively institutionalized level of cross-platform infrastructure relates to the constantly emerging, instituting level of the permanent interplay between algorithms and users. And this is where Cornelius Castoriadis’ theory of the imaginary comes into play.

Ontological indeterminacy

Castoriadis starts from society in its entirety as an institution that is first brought forth by the imaginary. Accordingly, the social imaginary is for him precisely the ability to be able to imagine something at all and “to see in one thing what it is not,” or rather still “to see it other than it is” (Castoriadis, 1997: 127). In this respect, for Castoriadis, the imaginary is not simply a reflection of reality and/or a phantasm of futures, but a fundamental dimension from which all social forms and institutions spring. In this sense, the concept is essential for the formation of sociality. In particular, the radical imaginary takes a central place, which is described as a “productive” and “creative” starting point, “manifested in indissolubly in both historical doing and in the constitution, before any explicit rationality, of a universe of significations” (Castoriadis, 1997: 146). The “radical” thing for Castoriadis about the imaginary is that it thus precedes the symbolic and is consequently fundamentally indeterminate, with which it as such appears precisely “radically” open and stands for permanent change (Gertenbach, 2011: 282). According to Castoriadis, institutions exist only in the symbolic, which provide for a certain form of stabilization, which is why something like sociality can emerge from them in the first place. At the same time, however, this symbolic itself is subject to permanent change. Therefore, in addition to the concept of “institution,” he also introduces the concept of the “instituting” which refers to this “perpetuation of otherness” (Castoriadis, 1999: 369) in the radical imaginary and describes the moments in which the instituting society breaks into the instituted and that creates itself as another (instituted) society (Castoridias, 1990: 369). For Castoriadis (1997), then, the (radical) imaginary exists on two levels: first, on a social-historical level, and second, on an individual-subjective level, which he calls “psyche/soma” (p. 369). He describes this relation as follows:

The radical imaginary exists as the social-historical and as psyche/ soma. As social-historical, it is an open stream of the anonymous collective; as psyche/soma, it is representative/affective/intentional flux. That which in the social-historical is positing, creating, bringing into being we call social imaginary in the primary sense of the term, or instituting society. That which in the psyche/soma is positing, creating, bringing-into-being for the psyche/soma, we call radical imagination (Castoriadis, 1997: 369).

Castoriadis here basically thematizes a central tension between constantly instituting and already instituted society, which is fed by the radical imaginary as the origin of becoming different. Such a conception of the imaginary resonates with the maxim of a “practice that ceaselessly changes the world and is changed by it” (Gertenbach, 2011: 280). These levels of theory of the social-historical and the individual-subjective thus also address the connection between practice theory and social ontology (Gertenbach, 2011: 278). For an algorithmic imaginary, this offers two crucial advantages:

First, with a conception of the algorithmic imaginary aligned with Castoriadis, it becomes possible to extend its role beyond the concrete interplay of algorithms and practices to the processes of institutionalization, without therefore having to assume a dialectic or a cybernetic feedback loop. This level is exemplified by cross-platform functions, such as algorithmic feeds, but also software plug-ins as the like button (in the form of little hearts) or the story format originally introduced by Snapchat.

And secondly, and almost more crucially, within these two connected levels, of algorithms and practices on the one hand and institutionalization processes of platforms on the other, the stabilizing as well as liquefying moments of this arrangement can thus be simultaneously brought into view. Referring to Castoriadis’ theory, this would mean to analyze the near and vanishing points of institution and instituent within both levels.

Here, the instituent and instituted behave in a way analogous to the difference between radical and social (or actual) imaginary (Ernst and Schröter, 2021: 33). Within this framework, the radical imaginary is responsible for the productive or creative impulses, which includes, as it were, the production of affects, while the social or actual imaginary is then directed toward the concrete shaping, and that is also stabilization, of the imaginary (Gertenbach, 2011: 286). Specifically related to the constantly constituting infrastructure of social media platforms, this means that a function such as the “feed” is implemented by the designers, but can only stabilize in the (social) interaction of the users with each other and mediated by the algorithm.

In between, Castoriadis locates two additional layers with the peripheral imaginary and the central imaginary (Ernst and Schröter, 2021: 33). Compared to the radical and social or actual imaginary, these are of secondary importance. Nevertheless, these two intermediate levels again provide a more accurate understanding of the relationship between the radical and the social imaginary and the workings of algorithm-based social media platforms. Despite the abstractness of these terms, with the help of this structuring of the imaginary, it can also be scaled quite concretely to a micro level, which parallels the notion of an “algorithm,” which also describes an abstraction as well as concrete applications in equal parts (Matzner, 2022). This surplus of the imaginary can now be exemplified by the feed-interface relation.

The imaginary and the feed interface relation

As pointed out, for Castoriadis (1997), the radical imaginary is the “perpetually orientation of otherness” (p. 369), which means that it stands for the constant change of meanings (Ernst and Schröter, 2021: 33). With reference to the algorithmic, which itself always represents a human–machine assemblage, this productive-creative aspect of the radical imaginary, as Castoriadis emphasizes it, then also offers an access to get a view of the interplay between the instituting and the instituted. In this respect, the basic principles of this algorithmic technique of feeds are instituted, respectively, they have a history that can be traced back to sociometry but is also radically indeterminate, because without the actions of the users, these systems would simply not work.

It is precisely here that Castoriadis theory offers a crucial advantage with his more fine-grained perspective of the imaginary, which locates between the two poles of the radical imaginary and the social or actual imaginary two further forms of the imaginary: the peripheral imaginary and the central imaginary. The radical imaginary on the one hand functions, as already shown, as a productive-creative act and makes the symbolic possible in the first place. Ernst and Schröter (2021) call it the “outer limit of the universe of significations” (p. 33). However, this outer limit shapes itself differently depending on the point of view and the actor. With regard to the functioning of a feed, users imagine its parameters as an external and for them invisible level of meaning, while, on the other hand, programmers and designers also have to imagine the user behavior (Fisher and Mehozay, 2019). The latter means, for example, that the programmers assume that the users want to see more of the same content. This assumption arises from the principle of homophily (Chun, 2021), which is implemented as an institutionalized algorithmic technique in the functioning of the feeds. In addition, the imaginations on the part of the interface designers, the ad team, the product managers, and the algorithms itself play crucial roles. Their individual agendas may follow an overarching capitalist principle, but can still conflict with each other. Typical examples can be found when rolled-out applications are not well received by users or are used with pleasure, but may be ethically questionable in certain respects, such as Facebook’s “emotional contagion” experiments regarding the feed in 2012. Here it becomes clear that the algorithmic imaginary cannot be limited to one of the involved sides, the heterogeneous user groups or the different platform actors, but has to be thought multi-perspectively. The approach to the imaginary can only ever be situated and dependent on the respective position (Natale and Guzman, 2022: 632). At this level, agency on the platform side is mainly tied to designers and their imaginings of various use cases, while at the same time user imaginaries are triggered by, for example, blog posts that typically accompany the launch of a new feature.

However, once the system with the different ideas of the actors involved has been established – that is, once an application or function has been rolled out (this also includes A/B tests) – something like a peripheral imaginary emerges. Castoriadis (1997) characterizes this peripheral imaginary as follows: “It corresponds to this second or an nth imaginary development of symbols, to the successive layers of sedimentation around a center” (p. 131).

If we understand these symbols as data that are algorithmically processed as a result of the actions of users, then the resulting profiling of these users, which here corresponds to the center, also constantly creates new meanings of this data image, while older, chronologically preceding meanings are constantly updated and supplemented. Thus, a constant recontextualization of the actions and the resulting data takes place, which also addresses the processual constitution of the imaginary. However, the peripheral imaginary remains, as it were, coupled to the constitutive dimension of the radical imaginary of the actors involved. At this level, this means that the data collection in the backend relies on the actions of the users.

The framework of this peripheral imaginary, which is also fed by the productive-creative radical imaginary, is then provided by the third form of imaginary in Castoriadis, the central imaginary. This central imaginary corresponds in Castoriadis’s (1997) philosophy to an “organizing principle” and also feeds from the “elementary symbols,” respectively, it assigns a “global meaning” to the human being within a system with it (p. 131). From the platform’s point of view, this corresponds to a categorization logic that is both institutionalized and individualized (with regard to the users), which emerges from the extracted patterns of constant recontextualization. So, at this stage, agency is initially on the side of the platforms, or more precisely, the algorithms recontextualize depending on the instituted principles and the user actions. Accordingly, the central imaginary can also be described as a “standardized point of departure” (Ernst and Schröter, 2021: 33). This arrangement is not to be understood as a stable structure that solves certain “real problems” as Castoriadis (1997) writes and, in the case of social media platforms, would probably correspond to a data image of the users that is as exact as possible (pp. 133–134). Rather, this problem constellation and thus the so-called “real problems” only exist through a central imaginary that is extended to the object of the like button, for instance, as a central actor in the feed and only functions through an underlying motivation such as recognition. The central imaginary is thus by no means “only” the purely technical algorithm that is imagined on the part of the users. That would only be one side of the story. Rather in the context of social media, the central imaginary is constitutive of the principle of recognition, because the motives of the users can rarely be justified by the fact that they want to game the algorithm in particular, but in the vast majority of cases they are concerned with the socialization performance conveyed by it. In relation to the data created in the process, or more precisely: the data to be extracted as a result of the user’s actions, the imaginary part of it is thus always part of the entire imaginary (Castoriadis, 1997: 131). In this respect, the central imaginary can also be described as a tipping point between constantly instituting and instituted infrastructure, since it is here that the normalized point of departure emerges in situ and with recourse to the algorithmic techniques within the social imaginary as the period of use progresses.

And this total or social imaginary then finally marks the fourth form of the imaginary in Castoriadis. This level, also called the actual imaginary, is characterized by the fact that it constitutes the meanings that have emerged in the three forms of the imaginary previously passed through (Castoriadis, 1997: 146). In this respect, it not only directly draws on the meanings of the central imaginary, but furthermore establishes the structure and division of the social world and consequently can be called the social imaginary (cf. Castoriadis, 1997: 146).

Again referring to algorithms and the relation of feed and interface, this level thus marks, in a sense, the reflux of the imaginary endowed with meaning from the invisible backend of the interface into the visible frontend. In the case of feeds, this is done by playing out machine learning-based “predictor modules,” that is, various feeds available in the backend, which use the random forest algorithm to imagine various interaction parameters and decision trees for different users. This means how likely it is that a user will interact with a certain image or video that has already been played, whereupon the feed intended for this action pattern will be played out. At this level, then, agency is primarily algorithmic, but according to the three forms of imaginary passed before, it is, again, also based on the actions of users.

Although this is a supposedly circular model from the front end, it is not a cybernetic feedback loop, as one might be inclined to assume, but rather more complex due to the imaginations in the back end and the existence of various algorithmic models that exist in parallel. And it is precisely this small-scale interplay that can be examined with the help of the four forms of the imaginary highlighted by Castoriadis. In this way, the micro-perspective of the constantly instituting interaction between users and algorithms can be scaled to the macro-level of the instituted infrastructure, without losing sight of the normative aspects inscribed (and constantly inscribing) in this process.

Conclusion

Such a conceptualization of an algorithmic imaginary makes it possible, above all, to adopt a more sophisticated perspective within this arrangement, which is usually described as circular, and which, in contrast to previous conceptions of “algorithmic” or “design imaginaries,” includes both perspectives, users and algorithms, as well as their designers at the same time. In addition, it becomes possible to focus on the theoretically elusive relationship between static and dynamic, between solid and fluid, with the help of the complementary relationship between the instituting and the instituted. For the study of social media platforms, this includes the gain of overcoming the difference between system and practices and dissolving the dichotomy of praxeocentrism and structural determinism.