Viral contagion and collective speculation: lessons of marketized finance for the networked public sphere Péter Csigó (PhD) research fellow Budapest University of Technology and Economics, Dept. of Sociology and Communications Budapest, Egry József u. 1, 1111 E building, room 715 Tel: 0036703832356 email: csigo@mokk.bme.hu Introduction Today, when the connective utopia of social networks sinks into bitter controversy at all levels, blame falls from all directions on the ultimate „popularity principle underpinning the online economy of social media”. The popularity principle has been claimed to incite the aggressive viral spread of sensationalist, tribal, clickbait stories through mechanisms of „friending, following and trending” (van Dijck 2013:13). „Sharing” has been commonly regarded as the essential fuel of this threefold popularity engine. Accordingly, users would share contents they „like” and let them spread quickly (trending) through networks of peers, who „like” each other (friending) and are keen to follow other peers whom they find „like” themselves (following). Social networks’ algoritmed news curation services – with their alleged ability to serve each user’s personal news preferences with contents shared by peers – have widely been thought to consummate a long-standing process of „media popularization” in which media and information services move closer and closer to popular audiences and seek to maximally cater to popular tastes. Scholarly research has grasped social media platforms like Facebook or Twitter as highly effective popular media services. Research has used the established vocabularies of popular media research. Indeed, the most discussed aspects of social networks and algorithmed news curation have all been „issues that have been central to the analysis of traditional media institutions” (Napoli 2014:354). Critics have discovered in news algorithms the same bias that formerly had been attributed to earlier popular media forms: the immediate gratification of popular taste at the expense of democratic values. Defenders have also mobilized long-known arguments of media as a force of popular emancipation from elite control.1 The above „import” of former scholarly concepts has followed a similar pattern to the way web1.0 debates twenty years ago replicated former popular media debates (Miller 2000:2- 3, Spigel 2004:12, Livingstone 2006:19). Popular media-inspired standard critique has a key inner limitation. The fiercer the critique of the „tyranny of popularity”, the less space is left to critically examine whether social media algorithms are indeed able to detect and serve popular news interests. Standard critique of the „popularity principle” has a tendency to take for granted social networks’ self-presentation as algorithmic superpowers of news personalization. This is a serious limitation. Today, we are clearly in a moment of transition when the first wave of the Social Network revolution ends and a new arrangement is being formed. This moment can hardly be seized solely with a critique that is complicite with its object in such an essential question. 1 Critics have resuscitated the long-standing popular media criticism that the immediate gratification of popular taste degrades the quality of public discourse and feeds a „tyranny of the popular”. Critics have referred to: sensationalism (algorithm driven journalism is „twerking” [Tandoc 2014]), dumbing down and commercialism (see Rogers 2009; Granka 2010, Foster 2012), spectacular impulses („hackers of attention” see boyd), segmentation-tribalization (algorithms enclose us into „filter bubbles” [Pariser 2011]), or manipulation (fake news). By contrast, defenders have pointed to choice, diversity and complexity: selective exposure (it’s „individuals’ choice” and not algorithmic filtering that causes ideological „bubbbles” – Science Bakshy), the higher diversity of non-elitist news curation (information diversity may trump filter bubbles Helberger et al 2015), and the very complexity of popular media (there is evidence equally for mainstreaming, polarization and diversification, but effects are modest – Flaxman). Following the hint that standard popular media related concepts may not represent in themselves „a sufficient vocabulary for assessing the intervention of … algorithms” (Gillespie 2011), the following paper takes financial speculation rather than popular media as the model by analogy of which to analyze and criticize the algorithmed news curation platforms of social media. I will apply on news algorithms (crowdsourcing based services of „implicite personalization” [Thurman-Schifferes 2012]) a new vocabulary of brokerage and speculation, inspired by political economic and ethnographic analyses of financial markets. In contrast to their standard image as popular media services gratifying users’ personal tastes and preferences2, the proposed approach identifies algorithmed news recommendation platforms as brokerage services which accelerate the viral spread of stories that conform with users’ „speculative” expectations about each others’ tastes and preferences. Social networks, in this view, create a viral contagion of contents which are commonly believed to be popular. The proposed approach will rethink the key concept of „sharing” as an inherently „speculative” activity and reconceive the phenomenon of „viral contagion” by the analogy of financial „bubble-blowing”. My starting point is that news traffic in social media is greatly driven by users’ speculative expectations about each others’ preferences. Social media platforms accelerate the viral spread of contents that many users believe to be popular in the crowd and therefore are willing to share further, to „like”, to click and engage with. Similarly to the speculative logic of financial trade which urges the investor to buy assets that she expects will be bought by other investors, social media users are urged to share stories they expect to be liked by their peers and like stories they expect to truly matter for peers who have shared them. Users of social media platforms who share stories and follow their peers’ shares are speculating on the „popular value” of these stories in the same way as investors try to read out the „market value” of financial assets from other investors’ past and expected trading activity (in a process to be called „collective valuation”). 2 There are good reasons to argue that the algorithmic personalization of news is not even possible. In a complex information environment where a very high number of potentially relevant information is produced every day, a practically infinite number of news packages can be constructed which do not significantly differ from each other in the personal „relevance” they represent for the user. In 2014, a FB product management director revealed that in any moment, 1500 potential stories could appear in the News Feed of the average user, and even ten times more for those with many friends and Page likes. We can assume the number of potentially relevant stories are much higher today than in 2014, as it depends from the number of friends and Pages liked (both growing quickly every year, for example, the average user in 2014 liked 50% more Pages than the year before). Out of thousands of possibly relevant stories, the News Feed reportedly displays 300. This 300 items large personalized news package is FB’s prime product. The above numbers give us a hint of the impossibly complex challenge FB takes up when it promises to compile the most personally relevant News Feed for each user. There is an almost infinite number of ways to select a hundreds-large news sample from a thousands-large news pool. The possible combinations of selecting a 100 element news sample from a 1000 element news pool amount to a mind-blowingly high, 140 digit number. There is, thus, an infinite number of possible News Feeds out of which FB’ recommendation service is supposed to select for the user the one that she would find the most possibly relevant and engaging for herself. This seems to be quite an impossible task, even for a powerful algorithm that reportedly „looks at 100000 highly personalized factors”. In a situation when thousands of alternative News Feeds can be compiled that the same user would find equally engaging or relevant for herself, a strategy of credibly imitating a personalized news service pays off better than attempting to actually provide it. If „sharing” is reconceived as an engine of speculation, the most apparent problems of viral contagion on social media platforms – clickbait contents, fake news, hysteric outrage, tribally ideological contents (Pariser 2011, Tandoc 2014) – cannot simply be drawn back to users being genuinely and viscerally interested in such contents (as it is often suggested by critics of the „attention economy”). This is an illuion maintained by social networks which justify their power by claiming that the stories they recommend do indeed gratify the personal information needs of users. Social networks justify this claim with hard statistics of „user engagement”, which veil the fact that users are not aware of their preferences and their engagements can neither be seen as an expression of a genuine personal interest. As long as users take peers’ preferences as a proxy to their own personal preferences (as they certainly do in social media) they will be willing to „engage” with whichever story as long as they believe their peers find them relevant for themselves. Users’ peer-dependence can be exploited by social networks, whose main preoccupation lies less in gratifying users’ personal tastes but recommending contents that suit users’ expectations about their peers’ taste. Social networks fill users’ News Feeds with stories that users see as popular among peers, and urge users to learn to accept these as personally highly relevant. Social networks are able to concentrate more audience engagement on recommended news stories than these stories’ genuine personal value is relative to other contents. The viral contagion of recommended stories makes users overestimate the popular value of these stories, and the overvaluation process runs independently from the genuine personal preferences of users – reminding to the way in which speculative „bubbles” depart from real-world „fundamentals” and overvalue the market value of assets in finance. In a similar way as fraudsters can trigger artificial market demand for assets that investors commonly expect to be found worthy to buy by other investors, social networks can trigger artificial demand for stories that users expect to be popular. The proposed model of speculation resonates well with the emerging scholarly interest in algorithmic failure and algorithmic stereotyping. It has been recognized recently that the project of personalized news recommendation may end up in „corrupt personalization, [a] process by which your attention is drawn to interests that are not your own” (Sandvig 2014). This highly controversial process can be explained with the proposed model that grasps the viral contagion of personally recommended news by analogy of „speculative bubble blowing”. This self-driven process runs independently of whether social networks do indeed assess the user’s personal taste correctly or – as it is too often the case – their assessments are based on „seriously wrong and misdirected assumptions”. (Halpern 2016) The proposed speculation model can explain the seeming paradox that while social networks are „using [their] raw data to come to strange and wildly erroneous assumptions”, they are also „cashing in on those mistakes” (ibid). This can happen because users are caught in a „speculative bubble” in which they are keen to engage with almost any commonsensically popular story if it is displayed in the News Feed and portrayed as if it was genuinely found most relevant by peers. This speculative „overvaluation” makes users engage with average, commonsensically and stereotypically popular stories, in the belief they are personally more relevant than other stories. The proposed speculation approach fully resonates with another strand of research which demonstrates – in the fields of labour and insurance markets, among others – that algorithmic attempts to detect personal profiles inevitably result in stereotyping the personal profiles they try to identify. The algorithmization of social control leads to coding bigot social stereotypes into algorithmic models. (O’Neil 2016, Goodman-Flaxman 2016, Sandvig et al. 2016) Research also documents that expertly and lay observers and users equally are keen to use stereotyped hints and truisms when making sense of what algorithms do. Such popular „folk theories” (Eslami et al. 2016), or commentator „guesswork” (Gillespie 2001), popular but unfalsifiable „bullshit” concepts (Mendelson 2012, Nielsen 2015) revolve all over the social media space, and are established as „social representations” (Moscovici 2001) that actors – who code, operate, analyze, criticize and use news algorithms – routinely rely on and use in their speculation about how algorithms feed us with „popular” stories and how they predict our taste. The argument of the article, to be developed in in four parts, is as follows. In the first part I will compare the „liquid intermediation infrastructures” that have emerged both in the financial sphere and in the networked public sphere through their „post-bureaucratic” transformation in the past decades. (Csigó 2016)3 These infrastructures have promised to support actors by two essential features: „algorithmed brokerage” and „collective valuation”. The key goal of „brokerage” in finance and social networks was the „diversification of portfolios”: new brokerage infrastructures connected investors with a diverse variety of investment opportunities, and media users with news from various sources. Key diversified products in the two spheres have been derivatives and News Feeds. Meanwhile, processes of „collective valuation” have been established with the aim of supporting individual actors in extracting individual „profits” from these new diversified products. Collective valuation processes have produced plausible representations of the collective verdict of all actors („the markets” or „the crowd” of peers) about the value of diversified products, and urged actors to plan their own personal pursuits against the background of these collective verdicts (that is, to like contents or buy assets that are liked or bought by crowds of actors). All investors and users have been urged to take active part in the collective evaluation process. The essence of collective valuation lies in actors’ active participation in the exchange of products: trading assets in finance and sharing stories in social media. The second part of the paper presents two widely recognized analytical models from the study of finance. The models have allowed scholars to grasp the basic „speculative” controversies that undermine the collective valuation process in deregulated financial trade. These controversies are evocatively described by Keynes’ famous „beauty contest” model (Keynes 1936) and André Orléan’s model on the „autonomy of collective valuation” (Orléan 2004) which represents a Durkheim inspired French school of political economy called „économie des conventions” . Both models are important for understanding the speculative nature of sharing on social media platforms, given that free sharing in the networked public sphere is the functional equivalent of free trade in the sphere of marketized finance. The third part of the paper takes inspiration from the two above models to describe how practices of brokerage and collective valuation have triggered a ranking-driven process of 3 The argument of the article broadly follows the four-part model that I have proposed (Csigó 2016:360-4) to grasp how liquidization transfigures into speculation in the spheres of mediatized politics and marketized finance. artificial overvaluation – or „speculative bubble blowing” – in both spheres. Caught in a collective speculation process, actors have been lured into heavily overvaluing the „market value” of overhyped and „overranked” assets in finance and the „popular value” of recommended news stories in the networked public sphere. The concluding part of the article argues that the problems of viral contagion can hardly be resolved without fully considering the logic of speculation which artificially overvalues contents that are commonsensically believed to resonate with the „popular soul”. 1. Liquid systems of brokerage and collective valuation in finance and the public sphere The analogy between the „networked public sphere” and the „marketized financial sphere” is far from arbitrary. Financial (and especially derivative) markets should be regarded as key models to understand the new public sphere created by social networks, because finance has been the first sphere of late capitalist society to undergo those fundamental structural transformations in the 1990es that later have shaped other fields as well, among them social media. As the rise and glory of deregulated finance clearly prefigures the triumph of social media platforms, the financial breakdown in 2008 should also be seen as an essential orienting anchor to handle the present crisis of the networked public sphere. Speculative tendencies make integral part of the „post-bureaucratic” structural transformations that in the past decades that have parallelly reshaped the financial sphere and the public sphere, together with many other fields like politics or cultural industries. Post-bureaucratic processes of „liquidization” (Bauman 2000) have emerged in reaction to the crisis of the bureaucratic model (Beniger 1986) of ordering and controlling society. Since the 1970s, worries have intensified that the bureaucratic “big institutions” of high modern society – like the mass party, high journalism, mass media, the Fordist factory, and the welfare state itself – are losing their power to connect social actors. A new conviction emerged that „liquid” – flexible, individualized, mobile, adaptive – mechanisms are able to satisfy and control the diverse and fragmented conglomerate of consumers-audiences in late capitalist society. Since the 1970s and 80s, liquid intermediation infrastructures – free markets, popular media, social media – have been widely advocated as the most adequate answer to the rising age of abundance and „plenty”, when scarcity ends and its corresponding model of social ordering, the model of bureaucratic control over scarce resources, is increasingly discredited (see Csigó 2016: ch.1.5. and p360-82). In the imagined, abundant world of „plenty” (affluent consumer society plenty of borrowers and investors, and multichannel media plenty of contents and active users), investment opportunities and media contents of possible interest are being felt to lay „out there” in abundance, but unseen and unexploited, kept behind bureaucratic fences. Gatekeepers like banks or media, accordingly, would constantly fail to reflect the very diversity of options that would be open for actors had these bureaucratic „middlemen” been left out of the picture. The new „liquid intermediation infrastructures” have challenged the rigid hierarchies of bureaucratic institutions, with the promise of fostering a more effective control, while also opening the space for a more autonomous self-realization and more horizontal self-organization of social and economic actors. The emerging liquid intermediation infrastructures have promised to bring flexible and effective new solutions to the problems capital allocation (hence the liberalization of financial markets), political representation (hence the rise of popular media- centered, mediatized politics) and news distribution (hence the rise of social networks and their „networked public sphere”). The above programme of liberating actors from bureaucratic rule via new „liquid intermediary infrastructures” has fostered a double creation process: the parallel construction of new systems of computer-assisted „brokerage” that would connect each actor with a diverse multitude of opportunities and highly diversified products on the one hand, and systems of „collective valuation” that would allow actors to rely on collectively judgements in assessing the value of the „intangible” products created in the process of diversification. (Brokerage systems: diversification through „repackaging” and „ranking”) Algorithmed brokerage systems play a key role in „liquidized” spheres, from marketized finance to the networked public sphere. I define here „brokerage” in the broad terms of network science, as a flexible „connection making” between so far unconnected entities, a professional practice of intermediation between diverse sets of actors and items. (Gould – Fernandez 1989)4 In media and in finance, the practice of information brokerage has been focused on „repackaging” and „ranking”. These practices have promised investors and users to diversify their investment portfolio and their media diet in a highly complex environmnent. „Repackaging”, the creation of new and diverse combinations from existing but hardly accessible products that conventionally had been kept apart by bureaucratic fences has been the main diversification technique of brokerage services. In the financial sector, the metaphor of „sausage” has been widely used (O’Neil 2016) as an analogy of „repackaged” financial instruments. The key „sausages” in the financial sphere have been derivatives, and their equivalent in the public sphere have been news listings or news feeds. In finance, derivatives are instruments which combine other, more basic financial assets, like loans and mortgages, that for long had been „illiquid” (immobile) forms of invested capital in traditionally regulated banking. During the decades before bank liberalization, bank loans (mortgages, consumer debt, corporate debt) were objects of a two-side agreement between creditor banks and debtors, which agreement has been based on the bureaucratic rules of the bank setting the terms of the credit. (My argument follows Orléan’s [2014:259-61], see also Csigó [2016:69-77, and chapter 3.1.]). Loans, thus, were immobile forms of capital, “illiquid and immune to market valuation” (Orléan 2014:259) or in other terms, „collective valuation” (Orléan 2004). The deregulation of banking, however, has allowed banks to “liquidize” this capital. In the process of securitization, banks have bundled vast numbers of loans (so far 4 To state the obvious, „media brokerage” is not to be conflated with the work of „data brokers” selling personal informations. illiquid assets) into a pool, transferred this pool to an new „intermediary” called a „special purpose vehicle”. This institution has chunked the pool into new, „repackaged” instruments, that combined hundreds or thousands of different loans and could be sold directly to investors as new financial products: derivatives. (see MacKeenzie 2011) Crowds of investors have been offered the whole ocean of consumer and corporate debt for investment, in the form of derivatives. In the creation and trade of derivatives, new financial intermediary institutions have played a key role. A whole new structure of financial intermediation has been built in which key roles are played less by traditional banks and depositors but new intermediaries that form a „shadow banking system” (see repowatch.org). „…traditional “relationship” banking (on balance sheet) was increasingly complemented by “transactional” banking (largely off balance sheet) based on securitization of traditional bank assets financed through wholesale funding. A long chain of intermediaries thus developed (the “shadow banking” system) to link ultimate borrowers and ultimate lenders.” (White 2013:4) In deregulated finance, „lightly regulated nonbank financial institutions" offer bank-like services without being submitted to bank regulation. (Volcker 2016:5). In the networked public sphere, the corresponding new intermediaries are social media platforms (polls, PR or multichannel media being their predecessors in the analogue world), and corresponding repackaged products are the News Feeds (equivalents of recommendation listings in commercial social media) which recombine the news and informations produced by traditional media actors and gatekeepers. A news feed is a recombination of media contents that algorithmic „brokers” claim to liberate from the closed platforms of traditional media companies. Media brokers have regarded contents enclosed in the „walled gardens” of media channels in a similar way as brokers in finance have seen the illiquid capital enclosed into the old-style bank: a huge reserve of yet unexploited value. Algorithmed „brokerage” systems of social media platforms have promised to enable every media user to select from this reserve the combination of news that best suits her needs. The other essential brokerage strategy, „ranking”, helps actors to find the most appropriate items by ranging them into hierarchical order according to their estimated worth (market value in finance or popular value in media). Most importantly, rankings are supposed to be universally accepted objective representations of value aggregating all available information that is scattered across the field. In the financial sphere, investors buy and sell repackaged securities typically by ranking them according to their risk and their profitability (negatively, but not deterministically correlated with risk). In terms of their risk, securities are formally ranked by credit rating agencies who use complex algorithmic models to range assets into risk categories from excellent to poor, designated by letters (A, B, C). Super-secure category (AA or AAA rated) securities offer lower yields than the ones ranged into riskier categories. Meanwhile, the yields of securities in the same risk category have not been entirely identical. Each security being composed of many different assets, their issuers have had some space to offer specific yields (called „spreads”) based on their estimation of how other securities offering similar combinations fare on the market. Investors have typically ranked securities by their „spreads” and by their risk category, and used this comparison in deciding which asset to buy (typically as we will see, the most profitable asset in the super-safe category). This „spreads-ratings nexus” (spread for a given risk rating) has been established as the very ground on which the market value of securities has been defined in the market process. (MacKenzie 2011:1786) In networked news recommendation services, „ranking” and „repackaging” have been closely integrated from the beginning. Repackaged News Feeds are themselves rankings of news items which are ranged by their estimated relevance and the observed engagement they are able to trigger. Here, what is „ranked” is not repackaged products (the News Feed), but the singular stories of which the News Feed is composed: ranking is inherent and essential to the work of repackaging itself. The integration of repackaging and ranking, which is a key source of corrosion, is an open and fundamental attribute of social media platforms, whereas the same integration (as we wil see) had worked more latently in the financial sphere. (Collective valuation systems: the empowerment of subordinated roles) Brokerage systems produce diversified products the value of which is „intangible” and hard to judge. Thus, liquid infrastructures have fostered processes of „collective valuation” (Orléan 2004) that have been meant to support each actor with instant information about how other actors evaluate products. In these collective valuation systems, each actor is empowered to judge the value of products by engaging in their free exchange (trading or sharing) and each actor is provided with prompt information about how the value of the same products is judged by similarly positioned peer actors (traders or sharers). Collective valuation systems in finance and social media have been based on the empowerment of actors. The clients (investors, audiences) of former bureaucratic institutions (banks, media organizations) have been liberated from their former dependence on these institutions and their fixed, predefined products (fixed illiquid assets, edited news products). Empowered clients have been involved into new and more flexible forms of distribution of products, through freely buying and selling „repackaged” assets, or freely sharing news and accessing news shared by others, and taking part in the work of „repackaging”. The above „empowerment” has been based on the liquification of formerly clear bureaucratic boundaries between institutions (commercial bank vs. investment bank, bank vs. non-bank in finance; media company vs tech company in media), between actors’ roles (buyer vs. seller, lender vs. borrower in finance, „gatekeeper” and „gated” (Shoemaker and Vos, 2009, Barzilai-Nahon 2008), performer and audience in media), and forms of capital in the field (liquid vs. illiquid in finance, proprietary vs, free-to-share in media). Investors and users have been urged to continuously monitor and speculate about the present and possible future movements of other actors who have been impersonated in third-person collective figures: „the markets” in finance (Lee and LiPuma 2002:196, Orléan 2004:200) and „the crowd” (of „people” or „peers” or „friends”) in the media and public sphere. These assemblies’ „collective wisdom” has been supposed to serve individual actors with more reliable guidance than what old „gatekeepers” (like banks or media channels) could offer. Collective judgment has been celebrated as more effective and democratic than bureaucratic reason, that has been denounced as narrow, biased, elitist, exclusive and sectarian. The best way to surpass the barriers of bureaucratic coordination, according to the liquid utopia, would lie in letting the markets themselves set the genuine market value of assets and let the crowd of peers or people themselves „define what’s news” and what’s the genuine value of stories. The collective valuation of repackaged products has been expected to help the personal pursuits of individual actors both in finance and social media. To achieve this, all actors have been provided with instant information about each others’ moves (instant market reports, instant „engagement” reports) and have been urged to form their personal strategies against the background of the moves of their peers („the markets”, „the crowd”). The promise of making personal gains from collective wisdom (much typical to all speculative thinking) has appeared in two variations in the two fields. In finance, the diversity of personal choice is secured by the variety of complex repackaged derivatives which are not personalized products themselves. All investors trade with the same derivatives, however, they are allowed to make personal decisions about how to buy and sell them. Individual investors relate to the movements of „the markets” strategically: they try to forecast these moves, to buy early when prices turn to rise, and sell early when they start to fall. These profit-maximizing strategies are believed to be shaped by well-informed individual considerations, which would all be built into aggregate market demand, making it a genuine collective judgment, according to the system’s ideology. In contrast to derivatives (being repackaged by system operators and being freely sold and bought by individual investors who monitor each others’ behaviour), the News Feed is a heavily personalized product which is repackaged partly by the crowd itself, bears in itself the wisdom of the crowd, but cannot be abandoned. The social media user is not allowed to opt out: she cannot choose between different News Feeds, she has to loyally accept the one she has been offered. The user, however, is still allowed to choose simply by „engaging”, which personal engagements are algorithmically monitored and are allowed to shape future news recommendations. In the end, the user is claimed to have the final word to define, through the choices she makes between the stories displayed, how her News Feed is compiled. The algorithm is claimed to learn from the user’s personal choices, to further enhance its recommendations by detecting how the user selectively engages with the stories that are „liked” in the crowd of peers. Similarly to brokerage, the system of collective valuation is much more concentrated in the social media sphere than in the financial sphere: while in the latter, investors make individual choices and market price aggregates these independent acts of trade, in the social media system, there is a personal news recommendation „coach” which choses instead of the user and the user merely approves or reject the algorithm’s choice. Collective valuation is inherent to the work of repackaging, and is openly integrated with the system of brokerage. The same integration has been a key cause of corrosion in finance, but it worked more latently than in social media, as we will see. 3. Two models from finance to grasp „sharing” as „speculation” Having compared the structural homologies between marketized financial sphere and the networked public sphere, I will turn now to models that reveal the vulnerability of the collective valuation process in finance, and I will transfer the same critique to the sphere of social media. The chronic instability of the financial sphere has made observers recognize long ago that aggregate trends of market demand cannot be read as a genuine collective vote of investors who would express their informed personal opinion about the value of assets by freely buying and selling them. This ideology of the collective valuation system (often called as the Efficient Market Hypothesis) can be attacked on various grounds. The most known objections have been made by behavioural economists claiming that personal judgments are not well-informed as they are shaped by collective manias and „irrational exuberance”. (Shiller) In the followings I will proceed along another criticism of collective valuation, which associates speculation less with irrationality than with a specific form of rationality. Speculative reason is based on the assumption that an actor can effectively estimate the personal value of an intangible (hard-to- evaluate) product (eg. a financial asset or a news story) by observing the spontaneous verdicts of „collective valuation” – like the supposed judgment of „the markets” in finance or the alleged „wisdom of the crowd” in social media – that aggregate other actors’ judgments on the product’s value. In the analysis of finance, an important critical approach argues that aggregate market demand an inadequate tool for genuine collective valuation because it is shaped not simply by actors’ judgement on the value of assets but inevitably also their expectations regarding how the majority of other investors judges this value. The basic rationale behind this approach may be grasped with the well-known case of the investor who is willing to buy an overvalued asset above the price she personally thinks would be its real value, as long as she expects that „the markets” will continue to buy it and its price will climb further. This simple case is just the entry point to the hardly controllable spiral of speculative and stereotyped reasoning that is unleashed in the very moment when the investor’s expectations about other investors’ future behaviour enter the picture. A key model to grasp the spiral of speculative expectations in finance is the famous „Keynesian beauty contest”. Keynes has compared financial speculation with a beauty contest where each spectator votes on three candidates, and those who have voted on the three actual winners get a reward. Keynes’ quesion is what spectators should do if they vote not just out of pure enthusiasm but with an eye on the reward? In this case, they cannot simply vote by their own taste, instead, they have to guess which candidate will get the most votes. This guess, however, cannot be simply aimed at genuine popular taste, either, as such a guess would naively assume that other spectators are voting by their own taste and are not speculating on the future winner. For a rational voter, the only option is to vote for the contestant whom the voter estimates would be perceived by the most voters as the most probable winner of the votes. Thus, in the end, it is not a case of choosing those [faces] which, to the best of one’s judgment, are really the prettiest, nor even those which average opinion genuinely thinks the prettiest. We have reached the third degree where we devote our intelligences to anticipating what average opinion expects the average opinion to be. And there are some, I believe, who practice the fourth, fifth and higher degrees. (Keynes 2007, 140) The above three (and higher) degree reflexivity problem jotted down by Keynes points to the very core of the inner instability of „liquid” systems that causes their immune deficiency to collective speculation. In the liquid systems of finance and social media, actors are similar to Keynes’ spectators who do not simply vote for the „prettiest face”: investors do not simply buy assets they personally find the most worthy, users do not simply share or engage with those contents they personally find the most relevant. Actors are wary of selecting solely on their own the most worthy assets or most relevant news, given the complexity of liquid systems and the abundance of choice. In lack of direct individual experience to assess products’ value, actors in both spheres turn to the crowd’s wisdom for orientation. They step up to a „second-degree” level of reflexivity where they try to find out what crowd members (similarly positioned as the actor herself) genuinely believe to be the most worthy products. For utopists of collective intelligence – call it „efficient markets” or „wisdom of the crowd” – this second-level aggregation of first-degree preferences represents a good enough, democratic and efficient solution. Keynes’ model shows why such a trust invested in second-degree solutions is illusory. Actors are inevitable doomed to grope in the dark at this „second-degree” level of reflexivity, as well, as they have no more direct personal knowledge about the „genuine” opinion of the crowd than about the „genuine” worth of derivatives/contents. Actors who cannot assess the value of assets or stories on their own (thus, cannot make „first degree” judgments) are neither capable of directly assessing the verdict of the crowd, either, in a „second-degree” judgment. Actors who look for collective (second-degree) anchor points for assessing the value of assets or news will inevitably rely on collective (third-degree) intelligence that helps them assess what exactly the crowd’s value judgment is and whether this judgment can be trusted. Investors will trust the crowd’s opinion only as long as they think that others do equally trust the crowd’s opinion (that a third-degree average opinion justifies the second-degree average opinion). In consequence, the spiral of speculation does not stand still here, at the third degree, either. On the financial market, actors cannot fully trust the average opinion of the crowd that bets on the further rise or fall of asset prices. The individual investor has to monitor how unanimously the crowd behaves. She monitors whether there exist a stable-enough „average opinion” that the markets’ actual „average opinion” about an asset’s value (that is, the asset’s actual price) is right and sustainable in the future. Accordingly, actors are checking whether and when a significant number of investors start betting against the prevailing „average opinion” about the sustainability of market price. Actors try to assess significant shifts in „(today’s) average opinion on (future) average opinion”. These judgments about the correctness of the crowd’s „third-degree” opinion have partly been algorithmed in the financial sphere: trading algorithms have been developed to detect those instances that reasonably represent a shift in the above „average opinion on average opinion”. The very fact that actors use algorithms to detect such instances is a clear proof of Keynes’ suggestion that the spiral of speculation does not stand still on the third degree, either. Investors do not trust their individual capacities are enough, so they code all the conventional signs of such a shift into algorithms and let the software make the work. This is already „fourth-degree” speculation, as shifts in actual „average opinion” about future „average opinion” are detected with the help of conventional, sedimented forms of expertly experience (fourth degree average opinion) on how such shifts happen (an „average opinion about average opinion on average opinion”). Even a fifth degree may exist where trading algorithms are trained to follow the moves of those outstanding trading algorithms that have (commonsensically, that is, approved by a higher, fifth-degree average opinion) proved to be the most successful in predicting such shifts. Before turning back to social media, I propose a second model that supplements Keynes’ and that has been offered by political economist André Orléan (2004, 2014), a key figure in the French school of thought called „économie des conventions”. Thinking further the process addressed by Keynes, Orléan argues that the above self-exploding spiral of speculation is stabilized by the fact that actors base their higher-degree assessments about „average opinion” not on individual considerations, but on collective conventions of thought that they all accept and that regulate their common self-understanding. These autonomous, Durkheimian collective representations – originating in the „shared, historical and cultural points of reference that define the identity of the group” (2004:211) – shape all actors’ expectations about other actors’ average opinion. Accordingly, market price in finance can hardly be regarded as a collective vote that aggregates genuine („exogeneous”) individual opinions (as predicted in the efficient market hypothesis). This view is right only as long as we are dealing with „first degree” judgments, strictly personal assessments of assets’ real value. By contrast, the higher we climb on Keynes’ speculative ladder, the more actors rely on second-, third- and higher degree speculative expectations on each others’ judgments, the more these considerations will be shaped by autonomous collective representations, stereotyped communal self-understandings which are evoked as common orienting points that drive individual actors’ pursuits, in an „endogeneous” (Orléan 2014:65) process of collective mutual adaptation. Recognizing that on financial markets (where actors’ expectations on each others’ value judgments are essential and more important than their personal vaue judgments) market demand is shaped by autonomous collective conventions of field actors’ self-understanding „leads us to modify profoundly the way in which we understand economic interactions. It brings to light a new type of reasoning, at odds with the classical individualist model that treats collective representation as the ’sum’ of individual opinions. Consequently, two levels and two ways of reasoning coexist, and the articulation between them needs to be analyzed.” (Orléan 2011:208) Keynes’ and Orléan’s models prefigure this paper’s endeavour to reveal how „popularization” (the process of detecting and gratifying each user’s personal information needs) feeds inevitably higher degree processes of „speculation on the popular” which result in the viral contagion of contents that are not genuinely popular but are collectively expected to be popular. Distinguishing these two levels and exploring their tensions is essential for understanding the popular and social media space in which we live. In my recent monograph, I have established the general claim that „collective speculation on popularity” has grown into an autonomous cultural structure in „liquid modern” society (Csigó 2016). In the followings I will present, with the help of Keynes and Orléan, how the networked public sphere and its sharing-based news distribution model are shaped by self-propelled speculative forces that spread contents which conform with collective stereotypes on popularity and popular taste. Social networks’ official ideology presents „sharing” as a „first-degree” activity and veils its inevitably speculative character. In the social network utopia, sharing is regarded as the prime vehicle of genuine self-expression. Accordingly, social networks would enable users to express their personal tastes and opinions through „sharing” and connect users with vital orienting information about their peers’ sharing activities that would also express peers’ genuine tastes and opinions. In this vision, algorithmed platformes would seamlessly aggregate users’ genuine first-degree personal preferences into genuine, unproblematic, transparent second-degree collective votes: News Feeds that would represent the genuine „average opinion” of the crowd of peers about what is relevant news today. However, social media forms a reflexive universe where social media users do not simply express what they need but also perform themselves in front of their „imagined audiences”. (Arif et al. 2016; Litt 2012, Abercrombie-Langhurst 1998) Accordingly, an act of „sharing” is shaped not just by the user’s personal preferences but also her expectations about others’. Social media users who “like”, “share”, “retweet” or “comment” do not simply express their personal opinion: they also make speculative bids in a collective popularity hunting game in which all users „share” and „like” in the hope of gaining affirmation from peers. (see Luckerson 2015, Eslami et al. 2016 about „liking” toddlers) The majority of users would certainly not share contents they personally agree with but estimate to be „unpopular” among their peers and risk to incite negative reactions. Users share stories they expect to be popular, and the stories they share will also be liked and shared further because those peers themselves expect them to be „popular” among their own peers, and so on. The more a news curation algorithm is focused on „sharing” activities, the more its recommendations will be contaminated by users’ collectively held (self-)stereotyping beliefs about what is “popular” in the crowd. For, sharing is not just a „first degree” expression of genuine personal taste („listen, this is relevant for me”) but also inevitably a „second-degree” estimation of the average taste of the crowd of peers with whom the content is shared („listen, this may be relevant for all of you”). When social networks’ algorithms track users sharings (and their engagements with shared contents) and aggregate them into news recommendations, they do something different from simply aggregating the „first-degree” personal opinions of users into second-degree collective judgments (the verdict of the crowd). The sharings that algorithms detect and aggregate bear the mark of users’ second-degree estimations about the genuine average taste/opinion of the crowd. Algorithms aggregate these estimations into higher-degree judgements: when they recommend a content that has been shared by many peers, they construct a (third-degree) average opinion out of peers’ opinions about their own peers’ taste/opinion. And the process moves beyond the „third degree”. This is because users’ second-degree judgments about popular taste are in fact never individual. Users can hardly make personal judgments on „what is news for others”. Their belief that a certain content is „popular” in the crowd of peers can hardly be imagined as a personal judgement that is formed against the grain. It is hard to imagine a user taking the following position: „I share a story that I know will be popular, though I know that most of my peers do not believe it is popular”. In the very moment when the user activates her expectations about genuine (second-degree) popular taste, she finds herself being spiralled on the third degree: her expectations about what is popular have to conform almost necessarily with established, commonsensical collective beliefs on what is popular („average opinion on average opinion”). Users cannot access individually the genuine „second-degree” opinion of the crowd, therefore they follow users’ commonsensical average opinion about popular opinion. Users inevitably rely on a higher, „third degree” collective wisdom in anticipating „what is relevant news for others” in the same way as, one degree lower on the ladder of speculation, they depended on „second-degree” collective wisdom in judging „what is relevant news for me”. News algorithms aggregare „sharings” which are themselves combinations of users’ second- degree judgements about the opinion of the crowd and third-degree judgements about the „average opinion on average opinion” that each user tries to follow when judging which content is „popular enough” to be shared. In fact, when news algorithms recommend the most shared/commented/liked contents, they aggregate users second- and third-degree judgements into an even higher level „average opinion”. These third- and fourth-degree „average opinions” represent what „the most people believe to be judged by the most people as popular”. Algorithms recommend contents which fit the most with the most collectively approved, most commonsensical beliefs about what is popular. 4. Ranking-driven bubble blowing in finance and social media Social networks’ news algorithms recommend contents that conform with actors’ commonly held stereotypes on popularity, and their main effect lies in teaching users to accept as personally relevant contents that are commonsensically popular. Pretending that „sharings” express genuine relevance judgments („first-degree” personal opinions), social networks urge users to learn to accept as personally relevant those stories that they expect to be relevant for their peers („second-degree” expectations). The conflation can be made on the ground that users and peers share the same common truisms about what is popular (third-degree expectations, collective beliefs on popular taste, „average opinion on average opinion”), thus, stories that sound for peers „popular” and thus shareable will sound „popular” for users, too, guaranteeing the frictionless operation of the speculative contagion engine. 5 The last element of the puzzle is to show that the infrastructure of Facebook and Twitter is able to mechanically concentrate user engagement on stories that are widely believed to be popular and are thus widely shared by peers. The mechanical concentration of actors’ engagement is vital in the mechanism of portraying commonsensically popular contents more relevant than their actual personal worth relative to other contents. The above mechanism of „artificial overvaluing” drives viral contagion on social media platforms, and the process called „bubble blowing” in finance. The common „speculative” defect of liquid brokerage systems that leads to bubble blowing in finance and the networked public sphere is that they cannot hold in balance their key constitutive components: repackaging, ranking and collective valuation. „Liquid” systems would be effective and functional only if they could guarantee the autonomy of these key constitutive elements. However, one of the three elements, ranking, tends to grow into a self- 5 Most important is to note here an essential aspect of news algorithms, which is that they construct third-degree stereotypes about popularity in a personalized form. News algorithms feed each user with her favourite stereotypes on popularity. The process of personalized stereotyping, serving the user with the personally most attractive combination of stereotypically popular stories, may well be analyzed by the model of „cold reading”, the well-known process of fake personalization in fortune telling (Forer 1949). This essential aspect of algorithms’ operation, unfortunately, will have to be explained in another paper. driven process that can colonize the systems of algorithmed brokerage and of collective valuation. In the networked public sphere just as at derivative markets, „ratings become rules” [MacKenzie 2011:84] on which all actors depend. This allows system operators to trick actors into overestimating the value of highly ranked products. The above „ranking-based overvaluation” lies in the core of the speculation bias of „liquid intermediation systems”. Research of derivative markets before 2008 shows that individual investors have by far too mechanically followed „the markets” and credit ratings to find out what is the most profitable and least risky offer. Investors mechanically followed the herd that went after the sexiest offer, and never questioned whether credit rating agencies estimated the risk of the asset accurately. This mechanical acceptance has allowed powerful insiders – who knew how to game (trick, influence, bribe) credit ratings agencies – to mechanically orient the crowd to highly risky assets. It was enough to hide toxic loans into fancy new derivatives properly designed in a way that they would be routinely rated as super-safe by agencies, and „the markets” predictably went crazy (falsely believing that their relatively higher spreads represented a fantastic investment opportunity in the super-safe category, while in fact, much higher spreads should have been offered, given the extreme risk of default which had not been properly expressed in credit ratings). Fraudsters could make arbitrage profits by pulling the strings of the crowd and mechanically triggering a fake market demand for valueless assets. As a derivative issuer expressed, “the whole market is rating-agency-driven at some level. . . . It’s just that there are investors who are constrained by ratings. . . and that creates value for everyone else and we’re in the business of exploiting that.” (MacKenzie 2011:1786) Since the infrastructure of social networks – as I have shown – is much more concentrated than derivative markets are, and it is arguable that system operators can drive users’ engagements even more mechanically than fraudsters could drive demand for fake financial products in pre- 2008 finance. On social media platforms, the News Feed directly establishes the mechanically driven market demand that fraudsters in finance could create only with indirect means through the hidden manipulation of brokerage and of collective valuation. As recommendation algorithms serve passive users who are not willing to actively personalize their news diet for themselves (Thurman-Schifferes 2012), operators like Facebook or Twitter can expect with all reason that users will mechanically approve news stories ranked in the higher strata of their News Feeds as of high personal value. The mechanistic attitude of users to accept the highest ranked offer has been documented long ago in relation to search engine rankings where users have been shown to „have high confidence in the search engine’s ranking” and „try out the top ranked results even if these results are perceived as less relevant for the task”. (Guan- Cutrell 2007) Social media users also tend to mechanically follow what the News Feed offers, and operators of Facebook or Twitter have good reasons to expect that, within the limits of common sense, they will engage with whichever content is recommended to them. Facebook and Twitter fully exploit users’ opportunistic stance of settling with „good enough” results. This allows for the companies to create artificial demand and pretend that users’ engagement marks the satisfaction of their genuine personal needs. The News Feed is believed to correctly represent (Eslami et al 2015)6 the genuine verdict of the crowd of peers about „what is news today” in a similar way as price movement is believed to express the genuine verdict of the markets or as credit ratings are believed to represent the best possible estimation of risk. Users’ mechanically triggered engagements with stories „shared” by others are immediately detected and fed back into the system, in the form of recommendations instantly displayed in the personal News Feeds of other users who will also mechanically „engage” with these contents, and so on. The system lubricates the quick viral spread of contents that crowd- dependent actors believe to matter to the crowd. Social networks, in a word, can trigger „chain reactions” of sharing and liking and seemingly „make popular” what actors believe to be popular, in a fraudulent self-fulfilling prophecy that is typical to the delusive effects of speculative bubble blowing that artificially increases the value of products that are expected to be highly appreciated by others (independently of its „fundamental” value). Seen from the above vantage point, the expansive viral spread of clickbait or hysterical or tribal stories reads less as a popularity-hunting „race to the bottom” than a self-propelled process of „speculative bubble blowing” that departs from the „fundamentals” of popular opinion formation, because it falsely depicts „shared” contents more popular than they actually are. Conclusion Social networks are not simply popular media infrastructures that immediately gratifiy popular taste. They are equally „speculation infrastructures” that accelerate and massify the flow of stereotyped information about popular taste. „Popularization” and „speculation on popularity” are heavily interwined structural processes which are opposed and coexistent at the same time, as are the positive and negative poles of a magnetic field. In the schizophrenic, „popular- speculative” media universe where social networks belong, wherever an attempt – algorithmed or non-algorithmed – is made to suit public discourse to popular taste, speculative forces are inevitably unleashed by actors who mobilize collective beliefs to make sense of the alchemy of popular success (Csigó 2016). It is vital to distinguish the above twin processes analytically and to address news recommendation algorithms as potential culmination points of both. Understanding the double character of news recommendation algorithms, how they maintain a „magnetic field” stabilized by contrary forces of „popularization” and of „speculation on popularity”, is vital for a more effective challenge of their too often dysfunctional operation. The corrosive effects of social networks on the public sphere will have to be located in the tensionful space that opens up between the two above contrasting logics: the popular race to the bottom which closes in to the popular ground and the speculative bubble blowing which departs from the popular ground. The very problem with Facebook or Twitter is that social networks exaggerate the popular value of stories that many users think to be popular, and maintain a speculative opinion climate, a „bubble”, where it becomes impossible to assess the real, fundamental value of stories for users and the these latters’ real, fundamental news preferences. The key indeces that they use to 6 Many users of social networks believe that their News Feeds precisely report on everything their peers have found important for themselves in a given day (many users are not aware even of the existence of a news curation algorithm). document the popularity of contents – like click numbers, likes, shares, or other forms of „engagement” – lose their very meaning in the bubble-like opinion climate of the networked public sphere. In a „speculative” environment such as Facebook’s or Twitter’s, it becomes impossible to tell, when users are found to „engage” with ideologically consonant, or sensational, or simplistic, or hysterical, or fake etc. contents, this is whether because these contents actually resonate with their genuine personal taste, or because they believe these contents are popular, attractive, relevant for the crowd. The key structural tension between genuine personalization (closing in to the popular ground) and speculative bubble blowing (departing from the popular ground) reveals the very „counterperformative” nature of liquidization processes which actively undermine the world they promise to create. (MacKenzie 2006, Csigó 2016:142) In finance, the liquidization of capital allocation has promised more safety and risk control, a „Great Moderation”, a more inclusive bank system, more loans for less solvent borrowers, higher profits for capital investors – and the process ended up in concentrating risk, in erasing profits, in chaos and breakdown. In the networked public sphere, the utopian promise has been to „give people the power to share” and „give everyone voice” in a new, „connected world” (see Carmody 2012), and the process has established a speculative opinion climate, a „bubble” in which the most stereotypically „popular” – scandalous, hysterical, aggressive, spectacular – contents are allowed to spread, aggressively silencing other voices. The bubble-like acceleration and expansion of stereotypically „popular” contents is extremely dangereous to the democratic public sphere, because it makes processes of popular opinion formation the pray of predatory trolls and manipulators, in a similar process as in finance, where the „liquid” capital allocation system has allowed for „predatory lending”, the deliberate production of trash assets for realizing arbitrage profits. In the networked public sphere, the bubble blowing process aggressively silences users’ real-world affinities and opinions that are much more varied than what the viral contagion of allegedly „popular” contents suggests. The post-bureaucratic revolution that has sought to replace large media organizations with self- regulating collectivities and liquid brokerage infrastructures has created its own news ecology which fosters the quick spread of news that are commonly and stereotypically believed to be popular. The proposed critique of post-bureaucratic reason and collective speculation connects with important efforts to deconstruct the liquid utopia and to rehabilitate the values of newsmaking by bureaucratic press organizations. Critics of the „peer production consensus” (Kreiss et al. 2011) and of the „Future of News consensus” (Starkman 2011) have persuasively pointed out the flaws of a public information ecosystem that is too strongly shaped by the connective, liquid utopia of networked communities „defining what’s news” for themselves. "When you take away bureaucracy and hierarchy ... you take away the ability to negotiate … on explicit terms. And you replace it with charisma, with cool, with shared but unspoken perceptions of power. You replace it with the cultural forces that guide our behavior in the absence of rules.” (Turner 2017) Among these cultural forces, a key role is played by the common stereotypes we maintain on each other, collective representations and myths that have grown into rule-setting structural forces as they have occupied the place of vaning bureaucratic (press and also representative political) institutions.  The controversies of the social network endeavour have made it clear that the undeniable distorsions of the former, centralized media system cannot be resolved simply by sidelining gatekeepers and letting the „liberated” information flow freely on algorithmed platforms from users to users. Seen from the proposed speculation viewpoint, the sobering conclusion is that the traffic of public information is inevitably forced behind fences: if not by old gatekeepers, then by users themselves who take their place in „defining what’s news” and submit news traffic to their common speculative stereotypes about what matters to „people like me” and to „people” in general (this contourlessness of the „imagined audience” [Litt 2012] is documented in Bortree 2005). Algorithmed news services, in spite of (or in fact, because of)7 all their efforts to articulate the genuine preferences of the popular audience in their real diversity, have become the prime medium expressing and vindicating – not simply popular voice, but – the unchallengeable stereotypes that our popular media-saturated societies collectively maintain about themselves. Without profound systemic reform similar in spirit to what has been proposed to cure the „systemic excesses” of the new financial intermediation system (Pozsar 2013), social networks and news algorithms will continue to glorify, under the pseudonyms of „sharing”, „giving voice” or the „wisdom of the crowd”, our societies’ most commonsensical and too often fatalistic and debilitating understandings of their own popular media-immersed lives. References Arif, A., Robinson, J. J., Stanek, S. A., Fichet, E. S., Townsend, P., Worku, Z., & Starbird, K. (2017). A Closer Look at the Self-Correcting Crowd: Examining Corrections in Online Rumors. In Proceedings of the 2017 ACM Conference on Computer Supported Cooperative Work and Social Computing (pp. 155– 168). New York, NY, USA: ACM. doi:10.1145/2998181.2998294 Barzilai-Nahon, K. (2008). Toward a Theory of Network Gatekeeping: A Framework for Exploring Information Control. Journal of the American Society for Information Science and Technology, 59(9), 1493–1512. Bauman, Z. (2000). Liquid modernity. Cambridge, UK; Malden, MA: Polity Press ; Blackwell. Beniger, J. R. (1986). The control revolution: technological and economic origins of the information society. Cambridge, Mass.: Harvard University Press. 7 This nice formulation comes from Justas Patkauskas’ review of my book. (Patkauskas 2018) Thank you for it. Bessy, C., & Chauvin, P.-M. (2013). The Power of Market Intermediaries: From Information to Valuation Processes. Valuation Studies, 1(1), 83–117. doi:10.3384/vs.2001-5992.131183 Bortree, D. S. (2005). Presentation of self on the Web: an ethnographic study of teenage girls’ weblogs. Education, Communication & Information, 5(1), 25–39. doi:10.1080/14636310500061102 Carmody, T. (2012). Facebook’s “Letter from Zuckerberg”: The Annotated Version. WIRED, 02. January. Csigó, P. (2016). The neopopular bubble: speculating on „the people” in late capitalist democracy. Budapest: CEU Press. Dijck, J. van. (2013). The culture of connectivity: a critical history of social media. Oxford ; New York: Oxford University Press. Eslami, M., Karahalios, K., Sandvig, C., Vaccaro, K., Rickman, A., Hamilton, K., & Kirlik, A. (2016). First I “like” it, then I hide it: Folk Theories of Social Feeds. In Proceedings of the 34rd Annual SIGCHI Conference on Human Factors in Computing Systems, Association for Computing Machinery (ACM) (pp. 2371–2382). /paper/First-I-like-it-then-I-hide-it-Folk-Theories-of-Eslami- Karahalios/3411f97b3ed4761f3d20fffaa0cbcdd62e76e650. Accessed 9 April 2017 Eslami, M., Rickman, A., Vaccaro, K., Aleyasen, A., Vuong, A., Karahalios, K., et al. (2015). “I Always Assumed That I Wasn’T Really That Close to [Her]”: Reasoning About Invisible Algorithms in News Feeds. In Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems (pp. 153–162). New York, NY, USA: ACM. doi:10.1145/2702123.2702556 Flaxman, S., Goel, S., & Rao, J. M. (2016). Filter Bubbles, Echo Chambers, and Online News Consumption. Public Opinion Quarterly, 80(S1), 298–320. doi:10.1093/poq/nfw006 Foster, R. (2012). News Plurality in a Digital World. London: Reuters Institute for Journalism. Gillespie, T. (2012). Can an Algorithm be Wrong? Limn. http://limn.it/can-an-algorithm-be-wrong/. Accessed 9 April 2017 Goodman, B., & Flaxman, S. (2016). European Union regulations on algorithmic decision-making and a “right to explanation.” In arXiv:1606.08813 [cs, stat]. Presented at the ICML Workshop on Human Interpretability in Machine Learning (WHI 2016), New York, NY. http://arxiv.org/abs/1606.08813. Accessed 13 February 2018 Gould, R., & Fernandez, R. (1989). Structures of Mediation: A Formal Approach to Brokerage in Transaction Networks. Sociological Methodology, 19, 89–126. Granka, L. A. (2010). The Politics of Search: A Decade Retrospective. The Information Society, 26(5), 364– 374. doi:10.1080/01972243.2010.511560 Guan, Z., & Cutrell, E. (2007). An Eye Tracking Study of the Effect of Target Rank on Web Search. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (pp. 417–420). New York, NY, USA: ACM. doi:10.1145/1240624.1240691 Halpern, S. (2016). They Have, Right Now, Another You. The New York Review of Books, (December 22). https://www.nybooks.com/articles/2016/12/22/they-have-right-now-another-you/. Accessed 9 April 2017 Helberger, N., Karppinen, K., & D’Acunto, L. (2016). Exposure diversity as a design principle for recommender systems. Information, Communication & Society, 1–17. doi:10.1080/1369118X.2016.1271900 Helberger, N., Kleinen-von Königslöw, K., & van der Noll, R. (2015). Regulating the new information intermediaries as gatekeepers of information diversity. info, 17(6), 50–71. Keynes, J. M. (2007). The general theory of employment, interest, and money. Houndmills, Basingstoke, Hamshire ; New York, NY: Palgrave Macmillan. Kreiss, D., Finn, M., & Turner, F. (2011). The limits of peer production: Some reminders from Max Weber for the network society. New Media & Society, 13(2), 243–259. doi:10.1177/1461444810370951 Lee, B., & LiPuma, E. (2002). Cultures of Circulation: The Imaginations of Modernity. Public Culture, 14(1), 191–213. Leung, S. Y., Corral, R., Smith, G., Pink, R., & Roper, G. (2014). Tweet for Democracy: Examining the Monitorial Citizen. In iConference 2014 Proceedings (pp. 1149–1156). doi:10.9776/14407 Litt, E. (2012). Knock, Knock. Who’s There? The Imagined Audience. Journal of Broadcasting & Electronic Media, 56(3), 330–345. doi:10.1080/08838151.2012.705195 Livingstone, S. (2006). The changing social landscape. In L. A. Lievrouw & S. Livingstone (Eds.), Handbook of New Media: Social Shaping and Social Consequences of ICTs (pp. 17–21). London ; Thousand Oaks, Calif; New Delhi: SAGE. Luckerson, V. (2015, July 9). Here’s How Your Facebook News Feed Actually Works. Time. http://time.com/collection-post/3950525/facebook-news-feed-algorithm/. Accessed 13 February 2018 MacKenzie, D. (2005). Opening the Black Boxes of Global Finance. Review of International Political Economy, 12(4), 555–576. MacKenzie, D. (2011). The Credit Crisis as a Problem in the Sociology of Knowledge. American Journal of Sociology, 116(6), 1778–1841. doi:10.1086/659639 MacKenzie, D. (2014). A Sociology of Algorithms: High-Frequency Trading and the Shaping of Markets. MacKenzie, D. A. (2006). An Engine, Not a Camera: How Financial Models Shape Markets. Cambridge, Mass: MIT Press. Maguire, J. S., & Matthews, J. (2010). Cultural Intermediaries and the Media. Sociology Compass, 4(7), 405–416. doi:10.1111/j.1751-9020.2010.00285.x Mendelson, B. (2012). Social media is bullshit (1st ed.). New York,N.Y: St. Martin’s Press. Miller, T. (2000). Hullo Television Studies, Bye-Bye Television? Television and New Media, 1(1), 3–8. Moor, L. (2008). Branding consultants as cultural intermediaries. The Sociological Review, 56(3), 408–428. doi:10.1111/j.1467-954X.2008.00797.x Moscovici, S. (2001). Social representations: explorations in social psychology. New York: New York University Press. Napoli, P. (2014). Digital intermediaries and the public interest standard in algorithm governance. Media Policy Project. http://blogs.lse.ac.uk/mediapolicyproject/2014/11/07/digital-intermediaries-and-the- public-interest-standard-in-algorithm-governance/. Accessed 9 April 2017 Napoli, P. M. (2014). Automated Media: An Institutional Theory Perspective on Algorithmic Media Production and Consumption. Communication Theory, 24(3), 340–360. doi:10.1111/comt.12039 Nielsen, R. K. (2015). Social media and bullshit. Social Media + Society, (April-June), 1–3. Noelle-Neumann, E. (1974). The Spiral of Silence A Theory of Public Opinion. Journal of Communication, 24(2), 43–51. doi:10.1111/j.1460-2466.1974.tb00367.x O’Neil, C. (2016). Weapons of math destruction: how big data increases inequality and threatens democracy (First edition.). New York: Crown. Orléan, A. (2004). What is a Collective Belief? In Cognitive Economics (pp. 199–212). Springer, Berlin, Heidelberg. doi:10.1007/978-3-540-24708-1_12 Orléan, A. (2014). The empire of value: a new foundation for economics. Pariser, E. (2011). The Filter Bubble: What the Internet is Hiding From You. London: Viking. Patkauskas, J. (2018). Book Review: The Neopopular Bubble: Speculating on ‘the People’ in Late Modern Democracy by Péter Csigó | LSE Review of Books. LSE Review of Books. http://blogs.lse.ac.uk/lsereviewofbooks/2018/02/07/book-review-the-neopopular-bubble- speculating-on-the-people-in-late-modern-democracy-by-peter-csigo/. Accessed 20 February 2018 Perse, E. M., & Lambe, J. L. (2017). Media effects and society (Second edition.). New York: Routledge. Peters, J. D. (1993). Distrust of representation: Habermas on the public sphere. Media, culture & society, 15(4), 541–571. Pozsar, Z. (2013). Shadow banking and the global financial ecosystem. VoxEU.org. http://voxeu.org/article/global-financial-ecosystem-0. Accessed 10 April 2017 RepoWatch. (n.d.). RepoWatch. RepoWatch. https://repowatch.org/. Accessed 10 April 2017 Rogers, R. (2009). The Googlization question, and the inculpable engine. In K. Becker & F. Stalder (Eds.), Deep Search: The Politics of Search Engines beyond Google (pp. 173–184). Edison, NJ: Transaction Publishers. Sandvig, C. (2014, June 26). Corrupt Personalization. multicast. http://blogs.harvard.edu/niftyc/archives/1003. Accessed 13 February 2018 Sandvig, C., Hamilton, K., Karahalios, K., & Langbort, C. (2016). Automation, Algorithms, and Politics | When the Algorithm Itself is a Racist: Diagnosing Ethical Harm in the Basic Components of Software. International Journal of Communication, 10(0), 19. Shoemaker, P. J., & Vos, T. P. (2009). Gatekeeping theory. New York: Routledge. Spigel, L. (2004). Introduction. In L. Spigel & J. Olsson (Eds.), Television after TV: essays on a medium in transition (pp. 1–34). Durham: Duke University Press. Starkman, D. (2011). Confidence Game. Columbia Journalism Review, 2011(November/December). http://www.cjr.org/essay/confidence_game.php. Accessed 13 February 2018 Straka, J. W., Zorn, P. M., Winkler, K. P., Markowitz, J., & Mahoney, P. E. (2011). A shift in the mortgage landscape: The 1990s move to automated credit evaluations. Journal of Housing Research, 11(2), 207–232. Tandoc, E. C. (2014). Journalism is twerking? How web analytics is changing the process of gatekeeping. New Media & Society, 16(4), 559–575. doi:10.1177/1461444814530541 Thurman, N., & Schifferes, S. (2012). The ’Social’ and Reflexive Turn of ’Adaptive’ News. In E. Siapera & A. Veglis (Eds.), The handbook of global online journalism (pp. 373–393). Malden, MA: Wiley-Blackwell. Turner, F. (2018, January 3). Don’t Be Evil. Logic Magazine, (03). https://logicmag.io/03-dont-be-evil/. Accessed 20 February 2018