TWO
Echo Chambers? Filter Bubbles? What Even Are They?

To determine whether echo chambers and filter bubbles are real, we must first develop a clearer definition of what these terms mean. This is surprisingly difficult: these concepts are rarely explicitly defined and the few definitions we do have vary widely over the years and across the phenomena they seek to describe.

At their core, though, they describe the properties of networks: offline and online; personal, social and professional. They describe how individuals and institutions (news outlets, political parties, activist groups, etc.) in those networks connect with each other, and how they communicate and share information with one another. Echo chambers and filter bubbles come into being when there are significant disconnects in the network: when specific groups and communities separate from the rest of the network (by their own choice or through the actions of others) and can therefore no longer be reached by new information from the outside.

In principle, such occasional disconnects from society are nothing new: they precede social media by several millennia. Echo chambers and filter bubbles are therefore not limited to online environments. However, modern platforms have made enclosures and disconnections more visible, more measurable, and potentially more powerful. Given the increasing role that online and social media play as information conduits and news sources (Newman et al. 2016), we must look especially to echo chambers and filter bubbles in online environments as we try to understand their spread and impact.

Conventional media – including print and broadcast as well as their online extensions – also play an important role here, however: as active participants in the networks themselves or as sources of the information that users circulate through social media. News outlets, influential individuals (politicians, journalists, celebrities, activists) within social networks, and platform providers such as Google, Facebook, and Twitter, serve as ‘information intermediaries’ (Helberger 2018: 154): in a complex and multifaceted information environment, users rely on them as gatewatchers and news curators (Bruns 2018) to reduce information overload. These intermediaries channel attention and help users find relevant information. Some intermediaries have even assumed quasi-monopolistic roles: this is certainly true for Google, whose search engine dominates the market.

Critically, these intermediaries do not limit the ability for diverse ideas and perspectives to be expressed, online or offline: they do not curtail the diversity of information supply, but potentially affect the diversity of information that users trusting in such intermediaries are exposed to (Helberger 2018: 158). If the information selection mechanisms built into intermediary platforms – directly in the form of search and filtering algorithms, or more indirectly in the user interactions they promote or discourage – privilege the availability and circulation of certain types of information over others, then users’ exposure diversity suffers. Ultimately, this assumes that such processes occur in concert across all the channels through which people access information: for example, if a user’s online information exposure is limited by the search engine they consult, but they maintain a diverse information diet in their offline news media consumption, then this does not pose a particularly significant concern (cf. Schmidt et al. 2017: 7).

Notwithstanding this broader, multi-platform perspective, much of the early discussion of echo chambers and filter bubbles focused on their potential impacts on single platforms, with definitions changing as the concepts were applied to ever new contexts. Let us now attempt to trace this shape-shifting across several key stages.

Early concerns about our online information diets predate even Google itself. In his 1995 book Being Digital, Nicholas Negroponte envisaged the Daily Me, a highly personalised online newspaper that selects news stories according to the explicit or implicit interests of each user. Still futuristic then, these predictions have been realised at least in part by now, with news websites employing user profiling and algorithmic content selection to such an extent that ‘gatekeeping no longer belongs to journalists or humans exclusively’ (Nechushtai and Lewis 2019: 299). Although Negroponte did not necessarily present such developments as negative, subsequent commentators highlighted especially the dismantling of a shared knowledge base for all citizens in society that would result from a heavy personalisation and algorithmic curation of news offerings (Zuiderveen Borgesius et al. 2016: 2). For example, President Obama’s farewell speech describes the ‘splintering of our media into a channel for every taste’ (2017: n.p.), threatening democracy itself.

In turn, Negroponte’s Daily Me is referenced explicitly by the legal scholar Cass Sunstein, the leading proponent of the echo chamber thesis, in his 2001 book Echo Chambers. As Sunstein puts it,

a well-functioning democracy – a republic – depends not just on freedom from censorship, but also on a set of common experiences and on unsought, unanticipated, and even unwanted exposures to diverse topics, people, and ideas. A system of ‘gated communities’ is as unhealthy for cyberspace as it is for the real world. (Sunstein 2001a: 2)

Sunstein would go on to further develop this argument across a series of books: Republic.com (2001b), Republic. com 2.0 (2009), and #Republic (2017), updating his warnings of the deleterious impact of such echo chambers for an evolving online media environment in which the role of personalised news websites diminished as personalisation in social media spaces increased.

Mid-1990s concerns about information personalisation generally remained at the level of hypothetical thought experiments and extrapolations, largely because news outlets’ technological capabilities for user profiling and content personalisation systems were still very poor; the Daily Me as foreseen by Negroponte and denounced by Sunstein never eventuated at the time. A more credible threat of informational fragmentation arose with the advent of modern search engines in the late 1990s: now, it did seem possible and even plausible that different users would be offered diverging results by the (undisclosed) search algorithm, customised perhaps based on the user’s search history.

Eventually emerging from these concerns is our second key concept: the filter bubble. This term was introduced by political activist and tech entrepreneur Eli Pariser, whose 2011 book The Filter Bubble: What the Internet Is Hiding from You opens with a reflection on Google’s efforts to personalise search results to address a given user’s likely interests. Pariser begins with an anecdote:

in the spring of 2010, while the remains of the Deepwater Horizon oil rig were spewing crude oil into the Gulf of Mexico, I asked two friends to search for the term ‘BP’. They’re pretty similar – educated white left-leaning women who live in the Northeast. But the results they saw were quite different. One of my friends saw investment information about BP. The other saw news. For one, the first page of results contained links about the oil spill; for the other, there was nothing about it except for a promotional ad from BP. (Pariser 2011: 2)

If such patterns are widespread and systematic, and result from algorithmic selection based on the individual user’s interest profile – as opposed to more benign factors, such as breaking news or even built-in randomisation designed to promote source diversity – they would indeed point to the possibility of filter bubbles. More specifically, these processes might lock users into highly idiosyncratic filter bubbles based purely on their personal interests, or could give rise to collective filter bubbles that enclose groups with broadly similar interests and ideologies in a ‘unique information universe’ that could facilitate the ‘hardening of their own political position’ (Krafft et al. 2018: 6; my translation). As Pariser puts it, such algorithms would ‘narrow what we know, surrounding us in information that tends to support what we already believe’ (Pariser 2015: n.p.).

If such early visions largely explore what is done to users by the personalisation algorithms of news sites or search engines, a more recent stream of discussion focuses more closely on what users do to one another, independently or with the support of platforms and their algorithms. Here, our attention shifts to social media in their early and contemporary forms.

This part of the story begins with the emergence of blogs in the early 2000s: at the time, it was expected that this new publishing format would bring about a shift from the consumption of mainstream news to more active online engagement and discussion, especially of political issues. This raised concerns about a ‘cyberbalkanization’ of the Internet into networks of blogs with shared ideological views (Adamic and Glance 2005: 37). Here, it is no longer the external force of search and personalisation systems that pushes users into echo chambers or filter bubbles; instead, users actively seek out those sources that best represent their...