This is a revised edition of a final paper submitted for the class Social Innovation and Government at Sciences Po, Paris in the February of 2018.
There is little doubt that the election of Donald Trump as President in the 2016 United States (US) Presidential Election and the decision to Leave in the United Kingdom (UK) European Union (EU) Referendum was perceived as a turning point for democracy the world over. For many, it shattered the grand illusion that social media contributed to fairer, more open democracy.
A disturbing sentiment was surfacing that perhaps social media was beginning to disguise fissures in social reality. One need only look to the extreme polarisation of the US political landscape, made evident during the 2016 US Presidential Election. Vocal leftists in the US were quick to draw attention to the socioeconomic and educational profiles of Trump voters, the weaknesses of the electoral college , Russian intervention  and an epidemic of “fake news” . Meanwhile, even the Republicans who had kept themselves disguised from the public debate began to engage in unfettered discourse. These were the Americans who felt alienated by the changing global economy, or at least found a voice in this chorus, and they wanted to be made great again. The post-election mayhem was intensified by the grudging, confused analyses of fallen pollsters and pundits tasked with reclaiming authority on the subject.
Could the increasingly heavy investment of electoral campaigns into social media help decode the underestimation of alternative views by political factions? This raises a wealth of related questions. Among these, two are most pertinent for analysis:
- Does social media contribute to more or less polarised democracies?
- Has the commodification of data changed the quality of political discourse in democracies?
To conclude, I will explore the vulnerabilities inherent in relying on social media as a site for a democratic agora as it is now, discuss interventions to restrict the impact of such vulnerabilities on the political process, and raise other interventions to this end.
Online communities are empowered but segregated
One crucial element of an effective democratic system is the active, equitable engagement of participants in deliberation. German philosopher Jürgen Habermas proposed the need for a defined structure for such activity: the public sphere, a virtual or physical space in which people could engage in debate about societal needs and governance. At its inception in Ancient Greece, it was the agora that provided a site for such collective deliberation. At once a gathering place, a marketplace and a space for civic life, the Greek agora was the heart of a city’s culture and democracy. In Athens, it was as much a site for open exchanges of ideas between free-born citizens as it was a marketplace for material goods. It was a frequent haunt of Socrates and the Stoics who sought to question and enlighten the masses. Public decisions were taken by summoning all free-born citizens from the agora to the hill opposite the Acropolis, where any individual who wished to persuade the people to go to war, to make a peace treaty, or impose a tax had to present his case publicly.
Athenian democracy by no means constituted an inclusive form of representative democracy: foreign residents, slaves and women could not participate in decision-making, and women did not interact in the agora in the same ways as did men . For centuries, republics worldwide have made courageous attempts to construct a Habermasian public sphere to support their democratic processes.
Many were sold on the idea of social media platforms as low-cost, inclusive agora. They are instantaneous, real-time conduits for information with unprecedented accessibility. Anybody with a computer and internet access (now recognised as a human right in the UN Declaration of Human Rights) could engage in political discourse. Social groups who were underrepresented or marginalised could raise awareness about their experiences. Social movements gained massive traction with the aid of social media, including the Arab Spring in 2010, Occupy Wall Street in 2011, Black Lives Matter from 2013, and #MeToo from 2017. The impact of such movements is difficult to evaluate holistically, but we know that in some cases, policy decisions have been reversed or reoriented as a direct result of viral content on social media. It is certain that social media platforms can lead to more inclusive deliberation. They communication frictions and increase the plurality of ideas, leading to better decision-making—at least in theory.
Algorithms built to polarise
How, then, do we interpret evidence that social media increases naïveté about the views of others, and undermines harmony and negotiation in a democratic environment, even if not to undermine representation per se? A crutch for such high-speed, low-cost information exchanges is that information is so abundant that what is displayed to any given individual must be partitioned in some way. Algorithms decide what is and is not shown to a user, on the basis of retweets, likes, reacts, and any other information one has registered on the social media network including profile and post or tweet content. Bot activity and sponsored advertisements also impact algorithmic output—this will be discussed in the following section. The ultimate aim however is to display only what is likely to interest the user—hence, what is displayed to the user depends crucially on their activity on these platforms. Algorithm performance is recursively improved by testing success with data from the platform. For a social network and its users, this appears to be a reasonable reaction to address the problem created by a wealth of information and restricted capability to display it, and the limited tolerance of people to consume it.
However, journalists and academics argue that such algorithms have resulted in a phenomenon of the filter bubble or echo chamber. This is an unwitting effect of an algorithm that first appears benign, and as expressed fittingly by historian Melvin Kranzberg in his First Law of Technology, it constitutes “Technology [that] is neither good nor bad; nor is it neutral” . Critics decry the loss of serendipitous exchanges online. Social media platforms such as Facebook choose what to show to the user on the basis of its engagement—a deceptively innocuous measure that favours displaying content that is more surprising, sensational or with which the user’s values and interests are already aligned.  The effect is to emphasise the role of instantaneous emotional responses and increase polarisation, impairing the possibility of the mutual learning, empathy and desire to engage that would be productive for addressing complex problems . Empirically, evidence for such polarisation was found in a 2011 cluster analysis of the US Twitterspace by Conover et al.
According to Sunstein in #Republic, there are preconditions for maintaining a republic: a capacity to listen, a commitment to reason, a belief in the good faith of (most) people who disagree with you, a desire to participate, love of country, a concern for the direction in which your country is moving. These will be lost in an age where each of us lives in our own unique information universe.
Although the US and UK are more divided than at any other time in the past few decades, these segregations are not simply a starker reflection of segregations existent in analogue social life. The simultaneous mass adoption of onlife democratic discourse and the inherent features of social media networks has contributed to greater polarisation in society. If this new, increasingly dominant social space does not support a democracy that is harmonious and cooperative, perhaps such a mode of discourse would no longer be as desirable.
The commoditisation of political influence
Social media also functions as a marketplace, yet the marketplace found in this online public space is not related to its function in information exchange merely by proximity, as it was in Athens. Camouflaged behind other salient benefits, advertising on online and mobile applications is becoming increasingly efficacious. More time is spent on the Internet than ever, giving social media companies now access to an enormous amount and granularity of data about many individuals’ tendencies, from which it is possible to model their economic preferences. This constitutes a crucial element of online services’ business models. The mass of willingly volunteered data now available on individuals’ patterns of preferences can not only productively inform product strategy and advertising spending, but provide increasing returns on investment through onward data brokerage. In recognition of this, the share of retail advertising budgets dedicated to digital marketing  in the US, Australia and the UK has increased and will continue to increase, even if returns on the investment are not necessarily evident. 
Such methods have found customers in electoral campaigns. They boost the effectiveness of existing methods of voter profiling: traditional demographic analysis, field research and voter history databases. Keen to extract the same benefits as retailers in this contemporary public space, political campaigns have progressively increased their spending on digital advertising with in 2016 each presidential campaign invested at least 20% of their budget into their digital campaigning strategy. This included spending on digital analytics firms hired on an ad-hoc basis, of which some have been highly controversial, while others remain hidden.
It’s a weaponised AI propaganda machineAn online commentator on the Cambridge Analytica saga
Cambridge Analytica was responsible for such a service in the Ted Cruz and Trump 2016 US presidential campaigns, and such services are being offered to electoral campaigns in Europe by start-up Liegey Muller Pons. Resources were directed toward individuals that models predicted to be more “persuadable”, which naturally included many individuals in the swing states, where there was no clear majority in the polls. This informed Trump’s frequent visits to Wisconsin and Michigan in the final weeks of his presidential campaign. Cambridge Analytica also provided a Twitter sentiment analysis tool to determine what topics people in those states were discussing at the time. As a consequence, immigration featured heavily in the speeches he gave in these states. By acting like a perfectly opportunistic algorithm following audience reactions, some claim that Trump had an advantage in garnering the support he needed to tip the scales in his favour. 
More disturbing still is the alleged involvement of Cambridge Analytica in voter suppression projects in Nigeria and the Trump campaign. In both instances, it was suggested that investment in get-out-the-vote projects and traditional canvassing was insufficient to ensure success. Targeting supporters of the competition with negative press about their favoured candidates would enhance the cause. If true, while not denying people of their right to vote, they may be coerced into abstaining when they would not otherwise. This would constitute an instance of manipulation, not persuasion, which I discuss in the final section of this paper.
The risk created by digital analytics here is that theoretically any given individual with an effective digital analytics team could get elected in a democracy just from parroting views people already hold.  What would decide the result at the polls is not a debate of ideologies but populistic appeals to the electorate.  It might be said that in a liberal democracy, competing on the basis of better agreement with the electorate in this way constitutes an undemocratic phenomenon that contributes to the concentration of power, and through which the mass discourse upon which democracy relies might be suppressed. 
I believe this view is too alarmist for several reasons. To begin with, it is unclear that such segregations are reinforced by social media, as opposed to merely expressed through it. In addition, there is no compelling evidence that the psychometrics employed by Trump’s data analysts was effective. It is true that Trump attracted the highest voter turnout in history at the 2016 Republican primaries, coming up to14 million, however the economic conditions of Republicans at the time, Trump’s explosive but “refreshing” profile, and the state of his competitors, are confounding factors external to this digital marketing strategy. More evidence points to the importance of factors other than the Internet, though no conclusive studies are available at the time of writing. Only a small proportion of Trump supporters use social media , while support for Trump was same or higher among groups least likely to use the internet  which Allcott and Gentzkow use, among others, to show that no conclusion can yet be made in either direction. The abductive power of such studies is, disappointingly, limited at present due to the privatisation of personal data—that is, much of the online data of interest to academics is owned by one of the few social media giants, and not free or openly accessible for analysis. Thirdly, the advantage data analytics confers in the swing states is a short-term advantage if at all; once all candidates have adopted similar methods they will return to competing on grounds other than competence in data-driven strategies. Also noteworthy is that Hillary Clinton had access to more sophisticated data analytics services and voter databases, especially through the Democratic National Convention, which already had precise access to individual data in President Obama’s 2012 US Presidential re-election. She also had consistently higher numbers of tweets and sponsored posts on Facebook than Trump, even though he had consistently higher interactions  on both platforms . There is thus reason to believe such a competition on other grounds had already occurred of the 2016 battle in the swing states. The influence of the economic context and field of competitors in which Trump was elected, as well as the character of Trump himself, must be considered. Finally, there is reason to believe that outside of the US, social media networks are less important in election outcomes, and perhaps the democratic process as a whole. The outcome of the French presidential election in 2017 was well-predicted by traditional polls but not by social media engagement. Le Pen, the more emotive far-right candidate in that had the largest following of any candidate on social media platforms, but ultimately lost the election by a substantial margin (Macron won 65%–35%) .
The new social marketplace provided by social media has the potential to uproot traditional methods of conducting democracy. Yet equally important is that from the near future, candidates will be on similar footing in terms of their digital strategies.
“It’s not about being sinister. it’s not about tricking people into voting for a candidate who they wouldn’t otherwise support. It’s just about making marketing more efficient.”Alexander Tayler, Chief Data Officer at Cambridge Analytica
Politics is marketing, and marketing is now data science. Once the digital playing field levels, as before, democratic candidates will compete on more fundamental grounds.
Can fake news lead to legitimate rule?
Another hugely controversial subject was that of “fake news”: news articles that are intentionally and verifiably false. Debate over how new media technology would misinform the electorate is not new—similar fears were spawned after the proliferation of cheap newspapers in the 19th century (fear of compromising the press as a check on power), then radio and television in the 20th (fear of reducing political debate to soundbites, and privileging charismatic candidates).
These fears are not completely unfounded in the context of social media, where engagement is king: any given user can reach an audience wider than those of physical newspapers or television, with no significant fact-checking or editorial judgment. Coupled with the increasing number of people depend exclusively on social media for their news, many are rendered vulnerable to believing in fake news and any consequent distortion in political beliefs.
Empirical studies show that fake 2016 US presidential election news stories were shared widely on social media, tended to favour Trump, and most Americans who saw fake news stories reported that they believed them to be true.  Why might it be productive to engage in proliferating fake news? Allcott and Gentzkow offer a useful economic analysis: “Fake news arises in equilibrium because it is cheaper to provide than precise signals, because consumers cannot costlessly infer accuracy, and because consumers may enjoy partisan news. Fake news may generate utility for some consumers, but it also imposes private and social costs by making it more difficult for consumers to infer the true state of the world”. Alternatively, foreign powers might profit from manipulating elections to favour a candidate that supports their interests. Evidence is mounting for Russian interference in US elections mediated through political advertisements and online “trolls” funded by Russian groups and targeted to US residents.  Along with supporting the spread of fake news, these trolls also sponsored “dark ads” on Facebook. These ads are only visible to the people to whom they are targeted. Their presence destabilises the quality of political discourse by permitting unhinged misinformation to occur, with users being none the wiser.
Fortunately, there are probably limits to what people can be convinced to believe on social media. In investigating the effect of social media misinformation on the elections, Allcott and Gentzkow estimated that the average US adult read and remembered one or perhaps several fake news articles during the election period, with more being pro-Trump than pro-Clinton. How much this affected the election results depends on the effectiveness of fake news exposure in changing the way people vote.
They use a survey on news sources and a benchmark demonstrating that exposing voters to one additional television campaign ad changes vote shares by approximately 0.02 percentage points. Consequently, if one fake news article were as persuasive as one TV campaign advertisement, the fake news in their database would have changed vote shares by hundredths of a percentage point—much smaller than Trump’s margin of victory in the pivotal states on which the outcome depended.  This might be a reflection that preserving our ability to infer the true state of the world is still socially valuable enough for the legitimacy of our democratic process to depend on it. I would further argue that voting is not like consumer behaviour in that people are more deliberative about their political choices. This would act as a powerful check against dark ads and appeals to emotion. The problem of foreign intervention, however, still requires attention.
Toward a better agora
In closing, we return to the questions raised at the beginning of this paper:
- Does social media contribute to more or less polarised democracies? Unsatisfying as it is, the short answer is that we don’t know. It is difficult to determine the specific impact of social media on recent political events, which are by their nature singular. It can nonetheless be said that it has resoundingly expressed the extreme polarisation between political factions in the US.
- Has the commodification of data changed the quality of political discourse in democracies? Yes, it has promoted private interests in politics. Beyond the lobbying and demagoguery, the data science industry has unlocked the capability of campaigns to tap on mass surveillance, profiling, and targeted advertising—for the right price.
The social space provided by social media platforms can support greater empathy, information exchange and visibility of political discourse. However, this is not without the aberrations from its corollary roles as a data marketplace, advertiser, information funnel, and tool for personal gratification. There are several courses of action that can be taken, the implementation of which will depend on the particular conception of democracy we aspire to protect.
A first step is to acknowledge that one cannot rely on either their own social media accounts nor traditional polling to form an impression or expectation of the electorate, at least before social media networks change. In the wake of the Brexit result, I witnessed on my own right-leaning Facebook news feed (made up largely of students in Cambridge and London at the time) several waves of disappointment and shock roll over. It was a display both of sincere upset about the result, and of the naïve expectations my social sphere had developed on the public opinion. On the side of the Fourth Estate, pundits, pollsters and journalists can no longer will have to shed the pretence of impartiality and authority, at least in the realm of politics. The hope is that with an awareness of the distortions brought about by social media platforms, people will seek to be more accurate in their judgements and expectations. In addition, the knowledge of the ways algorithms and data analytics are informing how electoral campaigns and retailers target users could stimulate users to be more deliberative about the advertisements they are shown online.
Traditional or state news providers might once again come to the fore. The political impartiality of journalists and entire agencies are being called into question, leaving behind an unsatiated appetite for reliable, incisive and fact-checked journalism. Ideally, such news providers should have no interest in targeting only specific segments of the population. In the past few years, newspapers such as the The New York Times and The Guardian have been actively differentiating themselves from competitors by emphasising accuracy and thoroughness. The creation of a news watchdog to oversee the quality of journalism is a possible idea, but perhaps not viable or otherwise unappealing. A simpler option is having the state own, control or regulate news (such as in Singapore) or social media (such as in the People’s Republic of China). While unpalatable to some, it is a compelling way to confine news to facts that can be justified and established—with the condition that the state acts in the interest of the populace as a whole. What about crowdsourcing credibility ratings for content? When summoned to Capitol Hill for hearings at the Senate, Facebook raised its “Flag as Inappropriate” and the newer “Dispute” tools as mechanisms for combatting fake news. It is not clear if people are invested enough in credibility of posts on social media to use this tool often enough, or objectively enough (if that were even possible), to govern content. However, a start-up known as Datagora currently in incubation at Sciences Po, Paris has the ambitious aim of providing a database that unites news articles, the sources to which its claims defer (in a book, journal or report et cetera), analysis of the semantics or method used to make that claim, and crowdsourced contributions and ratings of accuracy. Which modes of journalism are favoured would depend on the desired social contract between the people, the state and social media corporations.
Moves to combat the opacity of dark ads and foreign intervention could involve policy or legal instruments aimed at increased transparency, or restricted access to social media user data, in the context of political campaigning. The aim is to dismantle the “darkness” in ads that at present allow misinformation to proliferate. At the House and Senate intelligence committees, congressmen and congresswomen pushed a bill known as the Honest Ads Act. The piece of legislation would require digital platforms with at least 50,000,000 monthly viewers to maintain a public file of all electioneering communications purchased by a person or group who spends more than USD$500 on ads published on their platform, and require online platforms to make all reasonable efforts to ensure that foreign individuals and entities are not purchasing political advertisements in order to influence the American electorate. If it passes, it would give US users what they need to catch and prevent instances of foreign intervention in US elections. Legal instruments can also powerful tools for disincentivising controversial uses of data. In the UK, all donations and expenses in electoral and referendum campaigns must be declared by law. It is illegal to accept foreign donations or services , while the Information Commissioner’s Office is still investigating whether any misuse of data occurred during Brexit campaigning.  Data protection regulations in the European Union mean that it would be illegal to store and use the individual-level data held by the Obama campaign in 2012 and by Cambridge Analytica now. Only data at the level of the balloting station is publicly available in France. Protections for personal data will be further strengthened in the General Data Protection Regulation for the European Union, which comes into force mid-2018.
As yet unresolved at this point is how to judge if echo chambers and dark ads are really manipulative, and thus lead to destructive political discourse. Semantically distinguishing rhetoric from manipulation is one of the oldest questions in Western philosophy. Someone who persuades may reach the same end as someone who manipulates, but the process and intention are different. Persuasion involves changing someone’s perceptions or behaviour by changing their beliefs. Manipulation reaches a similar outcome through underhanded, deceptive, or even abusive tactics. The outcome is also often in the interest of the manipulator, and at the expense of others. Following the same tangent, it is difficult to make the case that data-driven resource allocation, dark ads and increased segregation involve manipulation. In many cases, beliefs are not changed, only strengthened in one direction or another. However, persuasion is a term that rings too tame to fully describe what on social media contributes to political discourse. I believe that this dissonance can be resolved under the condition of greater transparency and awareness about the data-driven strategies being employed in electoral campaigns—which some of the interventions discussed above can address.
Finally, a solution might come from within the social media infrastructure itself. Facebook already plans to hire 200 new staff to conduct reality checks to moderate political content online. More can be done to promote serendipitous exchanges on social media, but also elsewhere on the Internet. Incorporating into the online realm the randomness of exchanges that still exists in the physical world would theoretically break down online “bubbles”, promote opportunities for social learning, and nurture mutual empathy.
To bring the agora fully up to date, there is a need for cooperation between the traders, the people, and the philosophers, through which better democracy can be fostered. Much can be done to address the lack of trustworthy arbiters of truth, structural faults in platforms, and the consequent distortions in the democratic process. However, the preponderance of data-fuelled digital electoral campaigns is not an aberration in the political realm, nor is it a death knell for democracy; it simply requires greater deliberation and adaptation in all of our myriad capacities as users, citizens, and the crowd from which data is being continually sourced for political gain.
‘2016 US Digital Marketing Budgets: Statistics and Trends’. Smart Insights (blog), 25 November 2016. https://www.smartinsights.com/internet-marketing-statistics/2016-us-digital-ad-spend-statistics-trends/.
ABC News. Hillary Clinton Full Speech at Harry Reid Portrait Unveiling, n.d. https://www.youtube.com/watch?v=7iXEIc0ClbQ.
Allcott, Hunt, and Matthew Gentzkow. ‘Social Media and Fake News in the 2016 Election’. Journal of Economic Perspectives 31, no. 2 (Spring 2017): 211–236.
‘Birth of Democracy: Women’. Accessed 30 November 2017. http://www.agathe.gr/democracy/women.html.
Boxell, Levi, Matthew Gentzkow, and Jesse M. Shapiro. ‘Greater Internet Use Is Not Associated with Faster Growth in Political Polarization among US Demographic Groups (Forthcoming)’. Proceedings of the National Academy of Sciences, n.d.
Doward, Jamie, Carole Cadwalladr, and Alice Gibbs. ‘Watchdog to Launch Inquiry into Misuse of Data in Politics’. The Observer, 4 March 2017, sec. Technology. http://www.theguardian.com/technology/2017/mar/04/cambridge-analytics-data-brexit-trump.
‘How Did the 2017 French Presidential Candidates Fare on Social Media?’ Myndset (blog), 9 May 2017. http://myndset.com/2017/05/french-presidential-social-media/.
Kranzberg, Melvin. ‘Technology and History: “Kranzberg’s Laws”’. Technology and Culture 27, no. 3 (1986): 544–60. https://doi.org/10.2307/3105385.
Kulat, Cathi, Keith N. Hampton, and Eszter Hargittai. ‘Stop Blaming Facebook for Trump’s Election Win’. Text. TheHill, 23 November 2016. http://thehill.com/blogs/pundits-blog/presidential-campaign/307438-stop-blaming-facebook-for-trumps-election-win.
Office of the Director of National Intelligence, National Intelligence Council. ‘Assessing Russian Activities and Intentions in Recent US Elections’, 6 January 2017.
O’Neil, Cathy. Weapons of Math Destruction, How Big Data Increases Inequality and Threatens Democracy. Penguin, 2017. https://www.penguin.co.uk/books/304513/weapons-of-math-destruction/.
Pariser, Eli. The Filter Bubble: How the New Personalized Web Is Changing What We Read and How We Think. Penguin Books, 2012.
Rotroff, Susan I., and Robert Lamberton. Women in the Athenian Agora. ASCSA, 2006.
Silverman, Craig. ‘This Analysis Shows How Fake Election News Stories Outperformed Real News on Facebook’. BuzzFeed News, 16 November 2016.
Silverman, Craig, and Jeremy Singer-Vine. ‘Most Americans Who See Fake News Believe It, New Survey Says’. BuzzFeed News, 12 June 2016.
Sunstein, Cass R. #Republic. Princeton, NJ: Princeton University Press, 2017. https://press.princeton.edu/titles/10935.html.
‘The Social Media Analysis of French Presidential Elections: Telescope or Microscope?’ Accessed 1 December 2017. http://matei.org/ithink/2017/04/24/the-social-media-analysis-of-french-presidential-elections-telescope-or-microscope/.
‘The State of Australian Digital Ad Spending and Why It Matters’. Smart Insights (blog), 22 April 2016. https://www.smartinsights.com/digital-marketing-around-the-world/digital-marketing-in-australia/state-australian-digital-ad-spending-matters/.
‘UK Online Ad Spend : Latest Statistics Released’. Smart Insights (blog), 11 May 2017. https://www.smartinsights.com/internet-advertising/internet-advertising-analytics/uk-online-ad-spend-latest-statistics-released/.
‘US Digital Ad Spend to Outstrip TV This Year, All Thanks to Mobile [#ChartoftheDay]’. Smart Insights (blog), 27 June 2016. https://www.smartinsights.com/internet-marketing-statistics/us-digital-ad-spend-outstrip-tv-year-thanks-mobile-chartoftheday/.
‘US Elections Social Media Report: Digging Deep into Data’. Quintly Blog (blog), 2 November 2016. https://www.quintly.com/blog/2016/11/us-elections-social-media-report/.
 This is although a central aim of the electoral college was to ensure more proportionate representation by state.
 Russian intervention is currently proving to be a probable, not merely plausible, hypothesis at the time of writing. However, there is little known about the real contribution of Russian activity toward Trump’s election.
 ABC News, Hillary Clinton Full Speech at Harry Reid Portrait Unveiling.
 The typical conception of the 4th and 5th century BC Athenian agora claimed the exclusion of women from public space, to the effect of “Women were not expected in the Agora” ‘Birth of Democracy: Women’.. However, this view has been challenged by archaeological finds depicting women conducting their affairs and business in public spaces (Rotroff and Lamberton, Women in the Athenian Agora, 3–10.).
 Kranzberg, ‘Technology and History’.
 Pariser, The Filter Bubble.
 Sunstein, #Republic.
 Advertising on online spaces as opposed to in physical space. This could include expenditure on dedicated staff for online advertising, tilting Facebook’s newsfeed algorithms in an advertisement’s favour, providing targeted conditions for such advertisements, purchasing advertising space on Google or Twitter, paying off “social media influencers”, sponsoring bloggers, et cetera.
 ‘2016 US Digital Marketing Budgets’; ‘US Digital Ad Spend to Outstrip TV This Year, All Thanks to Mobile [#ChartoftheDay]’; ‘The State of Australian Digital Ad Spending and Why It Matters’; ‘UK Online Ad Spend’.
 O’Neil, Weapons of Math Destruction, How Big Data Increases Inequality and Threatens Democracy.
 Populism is increasingly becoming a feature of global politics (consider Donald Trump, Brexit, Pauline Hanson and Rodrigo Duterte). However, due to the myriad contributing factors to the election of populist leaders other than social media networks, I will presume here that populism is neither necessary nor sufficient in a democracy.
 This judgement will vary depending on one’s conception of democracy.
 Kulat, Hampton, and Hargittai, ‘Stop Blaming Facebook for Trump’s Election Win’.
 Boxell, Gentzkow, and Shapiro, ‘Greater Internet Use Is Not Associated with Faster Growth in Political Polarization among US Demographic Groups (Forthcoming)’.
 Allcott and Gentzkow, ‘Social Media and Fake News in the 2016 Election’.
 A measure of engagement in terms of likes, reactions, comments and shares on Facebook, and replies, likes and re-tweets on Twitter.
 ‘US Elections Social Media Report’.
 ‘The Social Media Analysis of French Presidential Elections: Telescope or Microscope?’; ‘How Did the 2017 French Presidential Candidates Fare on Social Media?’
 Silverman, ‘This Analysis Shows How Fake Election News Stories Outperformed Real News on Facebook’; Silverman and Singer-Vine, ‘Most Americans Who See Fake News Believe It, New Survey Says’.
 Office of the Director of National Intelligence, National Intelligence Council, ‘Assessing Russian Activities and Intentions in Recent US Elections’.
 Allcott and Gentzkow, ‘Social Media and Fake News in the 2016 Election’.
Thus the controversy over the possible involvement of Cambridge Analytica, which is incorporated in the US.
 Doward, Cadwalladr, and Gibbs, ‘Watchdog to Launch Inquiry into Misuse of Data in Politics’.0