The communicative context during and immediately after a televised debate is relatively fertile. These are long and complex events containing many policy statements and subtle behavioral cues, very few of which become salient in journalists’ reports and audience reactions. When they do become salient, it makes a difference to individuals’ responses on social media (Shah, et al., 2015). Older studies of traditional (not dual-screened) viewing of U.S. primary and presidential debates have shown that these events can affect individuals’ levels of information, attitudes toward the candidates, engagement, efficacy, and even vote choice (Benoit, et al., 2003; McKinney & Warner, 2013).
We should also consider the short-term “opportunity structure” (Chadwick, 2011b: 5–8) that now shapes engagement immediately after a broadcast event. Getting involved soon after a debate offers individuals the opportunity to influence others’ perceptions of the debate itself. Actions can be timed for when politicians, campaign workers, professional journalists, and political activists are involved in a struggle to define the candidates’ key strengths and weaknesses. However, post-debate actions like contributing to post-debate donation surges, voting in online petitions and polls, or following a party leader on Twitter are not narrowly instrumental; they are also indirect information signals designed to influence broader perceptions. These forms of engagement leave visible traces that can be read by others as signs of support for a candidate or cause. If they appear in sufficient numbers, they can influence perceptions of success or failure (Chadwick, et al., 2017).
But what if some of these social media–enabled dual screening roles are performed, not by humans, but by computers?
Bots are automated and semi-automated social media accounts that engage social media users to try to influence journalists’ and citizens’ perceptions of events, and, in the longer term, the formation of public opinion and behavior. When bots are instructed to act in concert, they produce botnets that can scale quickly, particularly if there are networks of humans organized and motivated to tweak bots’ workings as an event unfolds. Twitter is the social media platform most susceptible to bots, due to its relatively open application programming interface, which permits all manner of applications and automated scheduling services to work with a user account. Twitter is also the most popular platform for dual screening live events.
Many bots are straightforward and useful, such as automated news aggregators that push out tweets, or responders that answer customer queries sent to companies’ Twitter accounts. Bots’ actions may, however, become problematic. This depends on the specific context and the motivations of the humans who program them. For example, a bot programmed to tweet jokes 200 hundred times a day or automatically retweet all messages containing a specific word might seem like harmless fun. Yet if deployed and orchestrated in sufficiently large numbers during an important political event, we can quickly see how tweet and retweet bots might shape understanding, most obviously by disproportionately flooding public discourse with specific messages, themes, images, and links to websites. To some extent, this is what happened during the 2016 televised presidential debates.
The first debate was on September 26. Analysis of a sample of more than 9 million tweets using 52 hashtags related to the debate shows that, across four days (debate day and the three days that followed), 20 percent of tweets about the debate came from accounts with a high degree of automation (defined as accounts that tweeted more than 200 times during the four-day period). Given that just half a percent of accounts fell into this highly automated category, it is extraordinary to consider that these accounts together generated one-fifth of the Twitter content about the first debate. The top 100 accounts by number of tweets produced an average of 500 tweets per day. This is in stark contrast with the average account, which produced only one tweet per day (Kollanyi, et al., 2016a). Clearly a great deal of bot activity occurred during the first debate, but was this likely to have assisted Clinton or Trump?
If we examine the usage of pro-Trump and pro-Clinton hashtags, it is clear that Trump enjoyed far higher levels of support than Clinton on Twitter. For the first debate, fully 39.1 percent of the debate-related tweets carried pro-Trump hashtags. This dwarfed Clinton’s total of 13.6 percent. But about a third (32.7 percent) of tweets using pro-Trump hashtags like #CrookedHillary or #MakeAmericaGreatAgain came from identifiable bots or accounts with obvious automation (defined in this case as accounts that posted more than 50 times per day). Supportive bot and automated activity was much lower for Clinton, but still reached 22.3 percent of pro-Clinton hashtagged tweets (Kollanyi, et al., 2016a). And these patterns were broadly replicated during the second presidential debate on October 9 (Kollanyi, et al., 2016b). Trump had a much larger botnet army than Clinton.
Automated Twitter activity clearly exists, and in surprisingly large, and growing, quantities. Bot activity has also become more complex and difficult to discern. Early in Twitter’s development, fake and spam accounts were relatively easy to spot. They were what became known as “eggs” because they had incomplete descriptions and no profile picture, just the stock Twitter avatar—an egg. But, by 2016, many of the most significant bots contained profile photos (usually scraped from random websites) and profile descriptions that looked like they were written by humans. Perhaps this is because many of them were: there is a global market for astro-turfed Twitter followers, which can be bought for as little as $5 per 2,500 (MacArthur, 2016).
While the rise of debate bots presents a threat to democratic norms, we should also put this in context. For there is a second, less dramatic, set of findings from these studies. Twitter traffic peaked significantly during the debates and in the period immediately before and after each debate. Yet the vast majority of these tweets did not use obvious pro-Trump or pro-Clinton hashtags; instead, they used neutral hashtags (Kollanyi, et al., 2016a, 2016b). These tweets were still highly pertinent to the debate, they just did not contain the obvious partisan markers typical of bots and highly automated accounts. This is because these tweets were in fact posted by humans, particularly the kinds of people who generally tweet only a few times a day and, in this case, wanted to comment online during the debate. Bot-produced tweets were far outnumbered overall, but particularly in the periods closest to the live broadcast. This suggests that, while bots and automated posting of various kinds now form a key part of the systemic context of televised debates, the Twitter traffic that matters most—because it flows during the periods when publics and journalists are most attentive—is when bots play only minor roles.
Two of the most popular hashtags among Trump-supporting Twitter accounts (including bots) during the televised debates were #crookedhillary and #lockherup. These were references to one of Trump’s most outrageous campaign themes. The chant “Lock her up! Lock her up!” rang out at many of his rallies, a response to his pledge that, if elected, he would appoint a special prosecutor to investigate the claim that Clinton had contravened regulations governing conduct in office when she operated a private email server during her time as Secretary of State in the Obama administration (Hicks, 2016).
During the second television debate, Trump said, “If I win, I’m going to instruct my attorney general to get a special prosecutor to look into your situation, because there’s never been so many lies, so much deception.” Clinton responded with, “It’s just awfully good that someone with the temperament of Donald Trump is not in charge of the law in our country.” But, before Clinton could finish her point, Trump interrupted her with the line, “Because you’d be in jail” (Politico Staff, 2016). Unusually, Trump’s remark met with immediate applause from the studio audience.
Here was the spectacle of a candidate for president openly and publicly threatening his opponent with imprisonment, with the support of a section of the studio audience, and in front of 67 million television viewers (Serjeant & Richwine, 2016). Coming so close to election day, the potential damage to Clinton was obvious, and further emphasized the ongoing significance of televisi
on in the American campaign. And the situation was made much worse for Clinton when, on October 28, with early voting well underway, FBI director James Comey announced to Congress that he was investigating a new batch of emails relevant to a closed investigation into Clinton’s private email server. These had been found on the computer of Anthony D. Weiner, a former New York congressman and husband of a Clinton campaign staffer, who was under investigation for sending explicit images to a fifteen-year-old girl (Rosenberg, 2016).
In campaign post-mortems, Clinton’s head of polling, Navin Nayak, said that the email server controversy was decisive in Clinton’s defeat, most crucially because it depressed turnout among college-educated white voters (Pengelly, 2016). But how did the context within which Trump’s (and Comey’s) extraordinary interventions come to be established? This provides our final example of dysfunctional hybridity in 2016.
RUSSIA, WIKILEAKS, AND THE HACKING OF AMERICAN DEMOCRACY
On July 4, 2016 (Independence Day), with Julian Assange still hiding in the Ecuadorian embassy in London to escape extradition to Sweden (see chapter 5), WikiLeaks published a database of 1,258 emails about the Iraq conflict. These emails had been sent to or from the private server Clinton had operated during her time as Secretary of State from 2009 to 2013.7 WikiLeaks volunteers had combed through a larger database of emails that had been released in February 2016 by the State Department in response to a freedom of information request. Crucially, WikiLeaks presented the emails in the form of a searchable online database.
Less than three weeks later, WikiLeaks published another large email leak, this time of 19,252 emails from the Democratic National Committee’s (DNC) internal computer network. The messages, which were placed online in a searchable format, came from the accounts of seven senior DNC staff, including directors of finance Jordon Kaplan, Scott Comer, Daniel Parrish, and Allen Zachary, and communications director Luis Miranda. When journalists started searching the archive on the hunt for stories, they found evidence that the DNC had breached its regulations by attempting to undermine Bernie Sanders, Clinton’s rival in the primaries (Beech, 2016).
A month before election day, WikiLeaks began releasing yet another batch of emails. Taken from the hacked Gmail account of John Podesta, the chair of Clinton’s campaign and a former White House Chief of Staff, some of the approximately 20,000 pages of messages included transcripts from private, by-invitation, speeches Clinton had given at investment banks and other Wall Street financial institutions during the early 2010s. Clinton had received payment for these speeches, and their supposedly off-the-record format (her aides had recorded her remarks) meant that her comments were less guarded than if the meetings had been fully public. None of the revelations was hugely damaging, and hard facts were in short supply. But the emails did reveal the internal machinations of Clinton’s campaign, including its relationship with Wall Street donors and its attempts to undermine its opponents inside the Democrats (Stein, 2016).
The DNC and Podesta emails had been obtained by WikiLeaks, following a series of phishing attacks by hackers with alleged links to the Russian intelligence services. The hackers went under the names Fancy Bear and Guccifer 2.0, among others (Entous, et al., 2016). With Podesta’s Gmail account, the attack took the form of a fake email that directed Podesta to a webpage with a Google logo and a login box, into which he unwittingly typed his username and password. The webpage was a fake; Podesta had surrendered his login credentials to the hackers.
Coming, as it did, at the end of a long and bitter campaign that saw Clinton only narrowly defeat socialist Bernie Sanders in the primaries, the Podesta email leak gave journalists fresh evidence to write about Clinton’s personal character. An extended analysis of the emails for news site Vox argued, “more than anything, the Podesta emails show how Clinton is the transactional politician many have long suspected. That’s a dispiriting conclusion for some who may wish she was a pure progressive” (Stein, 2016). The message was clear enough: Clinton’s principles were in doubt. The 2016 Democratic primary campaign had opened up divisions between left and right. In the run-up to the July convention, a Pew survey found that 15 percent of those who voted for Sanders in the primaries would not be expected to vote for Clinton in the presidential election (Pew Research Center, 2016a: 24). The Podesta emails seemed to confirm these people’s suspicion, undermining the Democrats’ volunteer base and, perhaps, Clinton’s ability to turn out the vote. And, to make matters worse, in the second televised debate, Clinton was again presented with questions about how the emails revealed the contradictions between her public and private personas.
The 2016 campaign therefore witnessed a new role for WikiLeaks. Here was further evidence of its disruptive ability to morph from one repertoire to the next, all the while exploiting the power of large data leaks to destabilize and undermine the settled practices of both political and media elites. But the WikiLeaks of 2016 was a different beast from the WikiLeaks of 2010. Having mostly given up on its earlier hybrid strategy of collaborating with professional media organizations to promote transparency in the public interest, WikiLeaks had switched to dumping unredacted, often personal, information from a single individual’s email account, without any clearly defined transparency goal other than to undermine Clinton’s candidacy.
In recognition of the damage the leaks were doing to Clinton’s cause, Trump lost no opportunity to mention the hacks. At a Florida press conference on July 26, he appeared to encourage Russian hackers to hack Clinton’s emails: “I will tell you this, Russia: If you’re listening, I hope you’re able to find the 30,000 emails that are missing. . . . I think you will probably be rewarded mightily by our press” (Crowley, 2016). At a rally in Wilkes Barre, Pennsylvania, on October 10, Trump shouted, “I love WikiLeaks,” before proceeding to read from printouts of some of the hacked Podesta emails (ReasonReport, 2016). Trump mentioned WikiLeaks 164 times during the final month of the campaign (Legum, 2017). In December 2016, with Trump as president-elect, the CIA presented to Congress the classified results of an investigation into the hacks. Their conclusion was that they believed that the Russian government had played a role in supporting the hackers, as part of a broad plan to undermine Clinton and boost Trump. Meanwhile, Julian Assange repeatedly denied that the “Russian government” was WikiLeaks’ source (Entous, et al., 2016).
The Podesta Gmail hack featured heavily in news accounts during the closing weeks of the campaign, and it came on top of the long-running saga over Clinton’s other email problem: the reports of the potential security breaches from her private server. The origins of the Clinton email saga went back several years (Hicks, 2016). Before becoming Secretary of State in 2009, Clinton had established her own email server on a computer in her home in New York State, using the mailbox [email protected]. Clinton channeled all of her personal and professional correspondence through this account and chose not to use the state.gov email server that is traditionally provided to State Department personnel. Under the Federal Records Act, federal officials’ email accounts are, excepting classified materials, treated as government property, should official correspondence be requested by Congress.
Clinton stood down as Secretary of State in 2013 but did not hand over the archive of emails stored on her private server. In 2014, the State Department requested that Clinton submit these messages, as part of a push to bring the government into line with regulations. At that point, Clinton submitted 55,000 pages of work-related messages. But a few months later, in March 2015, an article in the New York Times raised questions about the security of her home server (Schmidt, 2015). Meanwhile, a House of Representatives committee investigating the 2012 terrorist attack on the US Consulate in Benghazi, Libya, in which two senior U.S. diplomatic staff were killed, had issued a subpoena requesting that all of Clinton’s emails related to Libya be made public (Schmidt, 2015). An extended period of uncertainly followed, as two investigations into Clinton’s emails—one by the State Department, the other by the FBI—kept the is
sue in the headlines. This tug-of-war between Republican congressional representatives and Clinton saw new batches of emails released every few months (Hicks, 2016). After fifteen months of back and forth, the conflict appeared to be over when, on July 5, 2016, FBI director James Comey ended the FBI’s investigation and cleared Clinton of any wrongdoing.
Yet, on October 28, Comey made an extraordinary intervention in the campaign. He issued a new statement informing Congress that the FBI was now investigating a new batch of emails that had been discovered as part of the Bureau’s investigation into the Anthony Weiner sex scandal. It was not until ten days later, on November 6, two days before election day, that Comey finally clarified that the new emails revealed nothing that indicated Clinton had acted improperly.
By that stage it was too late for the Clinton campaign to repair the damage. Comey’s announcements were a gift to Trump. They gave the veneer of official respectability to the email rumors that had rumbled through the campaign since the primaries. Those rumors had been fueled by the hacks that Trump had celebrated and even encouraged, keeping the issue in the headlines. Major news outlets, including the New York Times, were eager to report on the email affair, as were right-wing online news sites such as Breitbart (Benkler, et al., 2017). And WikiLeaks was equally happy to publish the leaks from the DNC and Podesta hacks, raising the salience of Clinton’s character during key moments of the campaign, and further increasing the likelihood that professional media would keep reporting on the issue. It was a vicious circle.
The Hybrid Media System Page 46