Automated or semi-automated software agents, better known as bots, have become an integral part of social media platforms. Reportedly, bots now generate twenty-four per cent of all posts on Twitter (Orlean “Man”), yet we know very little about who these bots are, what they do, or how to attend to these bots. This article examines one particular prominent exemplar: @Horse_ebooks, a much beloved Twitter bot that turned out not to be a “proper” bot after all. By examining how people responded to the revelations that the @Horse_ebooks account was in fact a human and not an automated software program, the intention here is not only to nuance some of the more common discourses around Twitter bots as spam, but more directly and significantly, to use the concept of persona as a useful analytical framework for understanding the relationships people forge with bots.
Twitter bots tend to be portrayed as annoying parasites that generate “fake traffic” and “steal identities” (Hill; Love; Perlroth; Slotkin). According to such news media presentations, bots are part of an “ethically-questionable industry,” where they operate to provide an (false) impression of popularity (Hill). In a similar vein, much of the existing academic research on bots, especially from a computer science standpoint, tends to focus on the destructive nature of bots in an attempt to design better spam detection systems (Laboreiro et.al; Weiss and Tscheligi; Zangerle and Specht). While some notable exceptions exist (Gehl; Hwang et al; Mowbray), there is still an obvious lack of research on Twitter bots within Media Studies.
By examining a case of “bot fakeness”—albeit in a somewhat different manner—this article contributes an understanding of Twitter bots as medium-specific personas. The case of @Horse_ebooks does show how people treat it as having a distinct personality. More importantly, this case study shows how the relationship people forge with an alleged bot differs from how they would relate to a human. To understand the ambiguity of the concept of persona as it applies to bots, this article relies on para-social interaction theory as developed by Horton and Wohl.
In their seminal article first published in 1956, Horton and Wohl understood para-social interaction as a “simulacrum of conversational give and take” that takes place particularly between mass media users and performers (215). The relationship was termed para-social because, despite of the nonreciprocal exposure situation, the viewer would feel as if the relationship was real and intimate. Like theater, an illusory relationship would be created between what they called the persona—an “indigenous figure” presented and created by the mass media—and the viewer (Horton and Wohl 216). Like the “new types of performers” created by the mass media—”the quizmasters, announcers or ‘interviewers’” —bots too, seem to represent a “special category of ‘personalities’ whose existence is a function of the media themselves” (Horton and Wohl 216). In what follows, I revisit the concept of para-social interaction using the case of @Horse_ebooks, to show that there is potential to expand an understanding of persona to include non-human actors as well.
Everything Happens So Much: The Case of @Horse_ebooks
The case of the now debunked Twitter account @Horse_ebooks is interesting for a number of reasons, not least because it highlights the presence of what we might call botness, the belief that bots possess distinct personalities or personas that are specific to algorithms. In the context of Twitter, bots are pieces of software or scripts that are designed to automatically or semi-automatically publish tweets or make and accept friend requests (Mowbray). Typically, bots are programmed and designed to be as humanlike as possible, a goal that has been pursued ever since Alan Turing proposed what has now become known as the Turing test (Gehl; Weiss and Tschengeli). The Turing test provides the classic challenge for artificial intelligence, namely, whether a machine can impersonate a human so convincingly that it becomes indistinguishable from an actual human. This challenge is particularly pertinent to spambots as they need to dodge the radar of increasingly complex spam filters and detection algorithms. To avoid detection, bots masquerade as “real” accounts, trying to seem as human as possible (Orlean “Man”).
Not all bots, however, pretend to be humans. Bots are created for all kinds of purposes. As Mowbray points out, “many bots are designed to be informative or otherwise useful” (184). For example, bots are designed to tweet news headlines, stock market quotes, traffic information, weather forecasts, or even the hourly bell chimes from Big Ben. Others are made for more artistic purposes or simply for fun by hackers and other Internet pundits. These bots tell jokes, automatically respond to certain keywords typed by other users, or write poems (i.e. @pentametron, @ProfJocular).
Amidst the growing bot population on Twitter, @Horse_ebooks is perhaps one of the best known and most prominent. The account was originally created by Russian web developer Alexey Kouznetsov and launched on 5 August 2010. In the beginning, @Horse_ebooks periodically tweeted links to an online store selling e-books, some of which were themed around horses. What most people did not know, until it was revealed to the public on 24 September 2013 (Orlean “Horse”), was that the @Horse_ebooks account had been taken over by artist and Buzzfeed employee Jacob Bakkila in September 2011. Only a year after its inception, @Horse_ebooks went from being a bot to being a human impersonating a bot impersonating a human.
After making a deal with Kouznetsov, Bakkila disabled the spambot and started generating tweets on behalf of @Horse_ebooks, using found material and text strings from various obscure Internet sites. The first tweet in Bakkila’s disguise was published on 14 September 2011, saying: “You will undoubtedly look on this moment with shock and”. For the next two years, streams of similar, “strangely poetic” (Chen) tweets were published, slowly giving rise to a devoted and growing fan base. Over the years, @Horse_ebooks became somewhat of a cultural phenomenon—an Internet celebrity of sorts. By 2012, @Horse_ebooks had risen to Internet fame; becoming one of the most mentioned “spambots” in news reports and blogs (Chen).
Responses to the @Horse_ebooks “Revelation”
On 24 September 2013, journalist Susan Orlean published a piece in The New Yorker revealing that @Horse_ebooks was in fact “human after all” (Orlean “@Horse_ebooks”). The revelation rapidly spurred a plethora of different reactions by its followers and fans, ranging from indifference, admiration and disappointment. Some of the sadness and disappointment felt can be seen clearly in the many of media reports, blog posts and tweets that emerged after the New Yorker story was published. Meyer of The Atlantic expressed his disbelief as follows:
@Horse_ebooks, reporters told us, was powered by an algorithm. [...] We loved the horse because it was the network talking to itself about us, while trying to speak to us. Our inventions, speaking—somehow sublimely—of ourselves. Our joy was even a little voyeuristic. An algorithm does not need an audience.
To me, though, that disappointment is only a mark of the horse’s success. We loved @Horse_ebooks because it was seerlike, childlike. But no: There were people behind it all along. We thought we were obliging a program, a thing which needs no obliging, whereas in fact we were falling for a plan. (Original italics)
People felt betrayed, indeed fooled by @Horse_ebooks. As Watson sees it, “The internet got up in arms about the revelation, mostly because it disrupted our desire to believe that there was beauty in algorithms and randomness.” Several prominent Internet pundits, developers and otherwise computationally skilled people, quickly shared their disappointment and even anger on Twitter. As Jacob Harris, a self-proclaimed @Horse_ebooks fan and news hacker at the New York Times expressed it:
Harris’ comparisons to the winning chess-playing computer Deep Blue speaks to the kind of disappointment felt. It speaks to the deep fascination that people feel towards the mysteries of the machine. It speaks to the fundamental belief in the potentials of machine intelligence and to the kind of techno-optimism felt amongst many hackers and “webbies.” As technologist and academic Dan Sinker said, “If I can’t rely on a Twitter bot to actually be a bot, what can I rely on?” (Sinker “If”). Perhaps most poignantly, Sinker noted his obvious disbelief in a blog post tellingly titled “Eulogy for a horse”:
It’s been said that, given enough time, a million monkeys at typewriters would eventually, randomly, type the works of Shakespeare. It’s just a way of saying that mathematically, given infinite possibilities, eventually everything will happen. But I’ve always wanted it literally to be true. I’ve wanted those little monkeys to produce something beautiful, something meaningful, and yet something wholly unexpected.
@Horse_ebooks was my monkey Shakespeare. I think it was a lot of people’s…[I]t really feels hard, like a punch through everything I thought I knew. (Sinker “Eulogy”)
It is one thing is to be fooled by a human and quite another to be fooled by a “Buzzfeed employee.” More than anything perhaps, the question of authenticity and trustworthiness seems to be at stake. In this sense, “It wasn’t the identities of the feed’s writers that shocked everyone (though one of the two writers works for BuzzFeed, which really pissed people off). Rather, it was the fact that they were human in the first place” (Farago). As Sinker put it at the end of the “Eulogy”:
I want to believe this wasn’t just yet another internet buzz-marketing prank.
I want to believe that @Horse was as beautiful and wonderful today as it was yesterday.
I want to believe that beauty can be assembled from the randomness of life all around us.
I want to believe that a million monkeys can make something amazing
I really, really do want to believe.
But I don’t think I do.
And that feels even worse.
Bots as Personae: Revisiting Horton and Wohl’s Concept of Para-Social Relations
How then are we to understand and interpret @Horse_ebooks and peoples’ responses to the revelations? Existing research on human-robot relations suggest that machines are routinely treated as having personalities (Turkle “Life”). There is even evidence to suggest that people often imagine relationships with (sufficiently responsive) robots as being better than relationships with humans. As Turkle explains, this is because relationships with machines, unlike humans, do not demand any ethical commitments (Turkle “Alone”). In other words, bots are oftentimes read and perceived as personas, with which people forge affective relationships.
The term “persona” can be understood as a performance of personhood. In a Goffmanian sense, this performance describes how human beings enact roles and present themselves in public (Goffman). As Moore puts it, “the persona is a projection, a puppet show, usually constructed by an author and enlivened by the performance, interpretation, or adaptation”. From Marcel Mauss’ classic analysis of gifts as objects thoroughly personified (Scott), through to the study of drag queens (Stru¨bel-Scheiner), the concept of persona signifies a masquerade, a performance. As a useful concept to theorise the performance and doing of personhood, persona has been used to study everything from celebrity culture (Marshall), fiction, and social networking sites (Zhao et al.). The concept also figures prominently in Human Computer Interaction and Usability Studies where the creation of personas constitutes an important design methodology (Dong et al.). Furthermore, as Marshall points out, persona figures prominently in Jungian psychoanalysis where it exemplifies the idea of “what a man should appear to be” (166).
While all of these perspectives allow for interesting analysis of personas, here I want to draw on an understanding of persona as a medium specific condition. Specifically, I want to revisit Horton and Wohl’s classic text about para-social interaction. Despite the fact that it was written almost 60 years ago and in the context of the then emerging mass media – radio, television and movies – their observations are still relevant and useful to theorise the kinds of relations people forge with bots today.
According to Horton and Wohl, the “persona offers, above all, a continuing relationship. His appearance is a regular and dependable event, to be counted on, planned for, and integrated into the routines of daily life” (216). The para-social relations between audiences and TV personas are developed over time and become more meaningful to the audience as it acquires a history. Not only are devoted TV audiences characterized by a strong belief in the character of the persona, they are also expected to “assume a sense of personal obligation to the performer” (Horton and Wohl 220). As Horton and Wohl note, “the “fan” - comes to believe that he “knows” the persona more intimately and profoundly than others do; that he “understands” his character and appreciates his values and motives (216). In a similar vein, fans of @Horse_ebooks expressed their emotional attachments in blog posts and tweets. For Sinker, @Horse_ebooks seemed to represent the kind of dependable and regular event that Horton and Wohl described: “Even today, I love @Horse_ebooks. A lot. Every day it was a gift. There were some days—thankfully not all that many—where it was the only thing I looked forward to. I know that that was true for others as well” (Sinker “Eulogy”).
Judging from searching Twitter retroactively for @Horse_ebooks, the bot meant something, if not much, to other people as well. Still, almost a year after the revelations, people regularly tweet that they miss @Horse_ebooks. For example, Harris tweets messages saying things like: “I’m still bitter about @Horse_ebooks” (12 November 2013) or “Many of us are still angry and hurt about @Horse_ebooks” (27 December 2013). Twitter user @third_dystopia says he feels something is missing from his life, realizing “horse eBooks hasn’t tweeted since September.” Another of the many messages posted in retrospect similarly says: “I want @Horse_ebooks back. Ever since he went silent, Twitter hasn’t been the same for me” (Lockwood).
Indeed, Marshall suggests that affect is at “the heart of a wider persona culture” (162). In a Deleuzian understanding of the term, affect refers to the “capacity to affect and be affected” (Steward 2). Borrowing from Marshall, what the @Horse_ebooks case shows is “that there are connections in our culture that are not necessarily coordinated with purposive and rational alignments. They are organised around clusters of sentiment that help situate people on various spectra of activity and engagement” (162). The concept of persona helps to understand how the performance of @Horse_ebooks depends on the audience to “contribute to the illusion by believing in it” (Horton and Wohl 220). “@Horse_ebooks was my monkey” as Sinker says, suggests a fundamental loss. In this case the para-social relation could no longer be sustained, as the illusion of being engaged in a relation with a machine was taken away.
The concept of para-social relations helps not only to illuminate the similarities between how people reacted to @Horse_ebooks and the way in which Horton and Wohl described peoples’ reactions to TV personas. It also allows us to see some crucial differences between the ways in which people relate to bots compared to how they relate to a human. For example, rather than an expression of grief at the loss of a social relationship, it could be argued that the responses triggered by the @Horse_ebooks revelations was of a more general loss of belief in the promises of artificial intelligence. To a certain extent, the appeal of @Horse_ebooks was precisely the fact that it was widely believed not to be a person. Whereas TV personas demand an ethical and social commitment on the part of the audience to keep the masquerade of the performer alive, a bot “needs no obliging” (Meyer). Unlike TV personas that depend on an illusory sense of intimacy, bots do “not need an audience” (Meyer).
Whether or not people treat bots in the same way as they treat TV personas, Horton and Wohl’s concept of para-social relations ultimately points towards an understanding of the bot persona as “a function of the media themselves” (Horton and Wohl 216). If quizmasters were seen as the “typical and indigenous figures” of mass media in 1956 (Horton and Wohl 216), the bot, I would suggest, constitutes such an “indigenous figure” today. The bot, if not exactly a “new type of performer” (Horton and Wohl 216), is certainly a pervasive “performer”—indeed a persona—on Twitter today.
While @Horse_ebooks was somewhat paradoxically revealed as a “performance art” piece (Orlean “Man”), the concept of persona allows us to see the “real” performance of @Horse_ebooks as constituted in the doing of botness. As the responses to @Horse_ebooks show, the concept of persona is not merely tied to beliefs about “what man should appear to be” (Jung 158), but also to ideas about what a bot should appear to be. Moreover, what the curious case of @Horse_ebooks shows, is how bots are not necessarily interpreted and judged by the standards of the original Turing test, that is, how humanlike they are, but according to how well they perform as bots. Indeed, we might ultimately understand the present case as a successful reverse Turing test, highlighting how humans can impersonate a bot so convincingly that it becomes indistinguishable from an actual bot.
Chen, Adrian. “How I Found the Human Being Behind @Horse_ebooks, The Internet's Favorite Spambot.” Gawker 23 Feb. 2012. 20 Apr. 2014 ‹http://gawker.com/5887697/how-i-found-the-human-being-behind-horseebooks-the-internets-favorite-spambot›.
Dong, Jianming, Kuldeep Kelkar, and Kelly Braun. “Getting the Most Out of Personas for Product Usability Enhancements.” Usability and Internationalization. HCI and Culture Lecture Notes in Computer Science 4559 (2007): 291-96.
Farago, Jason. “Give Me a Break. @Horse_ebooks Isn’t Art.” New Republic 24 Sep. 2013. 2 Apr. 2014 ‹http://www.newrepublic.com/article/114843/horse-ebooks-twitter-hoax-isnt-art›.
Gehl, Robert. Reverse Engineering Social Media: Software, Culture, and Political Economy in New Media Capitalism. Temple University Press, 2014.
Goffman, Erwin. The Presentation of Self in Everyday Life. New York: Anchor Books, 1959.
Harris, Jacob (harrisj). “For a programmer like me who loves whimsical code, it’s a bit like being into computer chess and finding out Deep Blue has a guy inside.” 24 Sep. 2013, 5:03. Tweet.
Harris, Jacob (harrisj). “I’m still bitter about ?@Horse_ebooks.” 12 Nov. 2013, 00:15. Tweet.
Harris, Jacob (harrisj). “Many of us are still angry and hurt about ?@horse_ebooks.” 27 Dec. 2013, 6:24. Tweet.
Hill, Kashmir. “The Invasion of the Twitter Bots.” Forbes 9 Aug. 2012. 13 Mar. 2014 ‹http://www.forbes.com/sites/kashmirhill/2012/08/09/the-invasion-of-the-twitter-bots›.
Horton, Donald, and Richard Wohl. “Mass Communication and Para-Social Interaction: Observations on Intimacy at a Distance.” Psychiatry 19 (1956): 215-29.
Isaacson, Andy. “Are You Following a Bot? How to Manipulate Social Movements by Hacking Twitter.” The Atlantic 2 Apr. 2011. 13 Mar. 2014 ‹http://www.theatlantic.com/magazine/archive/2011/05/are-you-following-a-bot/308448/›.
Jung, Carl. Two Essays on Analytical Psychology, 2nd ed. London: Routledge, 1992.
Laboreiro, Gustavo, Luís Sarmento, and Eugénio Oliveira. “Identifying Automatic Posting Systems in Microblogs.” Progress in Artificial Intelligence. Ed. Luis Antunes and H. Sofia Pinto. Berlin: Springer Verlag, 2011.
Lee, Kyumin, B. David Eoff, and James Caverlee. “Seven Months with the Devils: A Long-Term Study of Content Polluters on Twitter.” Proceedings of the Fifth International AAAI Conference on Weblogs and Social Media, 2011.
Lockwood, Alex (heislockwood). “I want @Horse_ebooks back. Ever since he went silent, Twitter hasn’t been the same for me.” 7 Jan. 2014, 15:49. Tweet.
Love, Dylan. “More than One Third of Web Traffic Is Fake.” Slate 24 Mar. 2014. 20 Apr. 2014 ‹http://www.slate.com/blogs/business_insider/2014/03/24/fake_online_traffic_36_percent_of_all_web_traffic_is_fraudulent.html›.
Marshall, P. David. “Persona Studies: Mapping the Proliferation of the Public Self”. Journalism 15.2 (2014): 153–70.
Meyer, Robinson. “@Horse_ebooks Is the Most Successful Piece of Cyber Fiction, Ever.” The Atlantic 24 Sep. 2013. 2 Apr. 2014 ‹http://www.theatlantic.com/technology/archive/2013/10/an-amazing-new-twitter-account-that-sort-of-mimics-your-tweets/280400›.
Moore, Chris. “Personae or Personas: the Social in Social Media.” Persona Studies 13 Oct. 2011. 20 Apr. 2014 ‹http://www.personastudies.com/2011/10/personae-or-personas-social-in-social.html›.
Mowbray, Miranda. “Automated Twitter Accounts.” Twitter and Society. Eds. Katrin Weller, Axel Bruns, Jean Burgess, Merja Mahrt and Cornelius Puschmann. New York: Peter Lang, 2014. 183-94.
Orlean, Susan. “Man and Machine: Playing Games on the Internet.” The New Yorker 10 Feb. 2014. 13 Mar. 2014 ‹http://www.newyorker.com/reporting/2014/02/10/140210fa_fact_orlean›.
Orlean, Susan. “@Horse_ebooks Is Human after All.” The New Yorker 24 Sep. 2013. 15 Feb. 2013 ‹http://www.newyorker.com/online/blogs/elements/2013/09/horse-ebooks-and-pronunciation-book-revealed.html›.
Pearce, Ian, Max Nanis, and Tim Hwang. “PacSocial: Field Test Report.” 15 Nov. 2011. 2 Apr. 2014 ‹http://pacsocial.com/files/pacsocial_field_test_report_2011-11-15.pdf›.
Perlroth, Nicole. “Fake Twitter Followers Become Multimillion-Dollar Business.” The New York Times 5 Apr. 2013. 13 Mar. 2014 ‹http://bits.blogs.nytimes.com/2013/04/05/fake-twitter-followers-becomes-multimillion-dollar-business/?_php=true&_type=blogs&_php=true&_type=blogs&_r=1›.
Scott, Linda. “The Troupe: Celebrities as Dramatis Personae in Advertisements.” NA: Advances in Consumer Research. Vol. 18. Eds. Rebecca H. Holman and Michael R. Solomon. Provo, UT: Association for Consumer Research, 1991. 355-63.
Sinker, Dan. “Eulogy for a Horse.“ dansinker.com 24 Sep. 2013. 22 Apr. 2014 ‹http://web.archive.org/web/20140213003406/http://dansinker.com/post/62183207705/eulogy-for-a-horse›.
Sinker, Dan (dansinker). “If I can’t rely on a Twitter bot to actually be a bot. What can I rely on?” 24 Sep. 2013, 4:36. Tweet.
Slotkin, Jason. “Twitter ‘Bots’ Steal Tweeters’ Identities.” Marketplace 27 May 2013. 20 Apr. 2014 ‹http://www.marketplace.org/topics/tech/twitter-bots-steal-tweeters-identities›.
Stetten, Melissa (MelissaStetten). “Finding out @Horse_ebooks is a Buzzfeed employee’s “performance art” is like Banksy revealing that he’s Jared Leto.” 25 Sep. 2013, 4:39. Tweet.
Stewart, Kathleen. Ordinary Affects. Durham: Duke University Press, 2007.
Strübel-Scheiner, Jessica. “Gender Performativity and Self-Perception: Drag as Masquerade.” International Journal of Humanities and Social Science 1.13 (2011): 12-19.
Turkle, Sherry. Alone Together: Why We Expect More from Technology and Less from Each Other. New York: Basic Books, 2011.
Tea Cake (third_dystopia). “I felt like something was missing from my life, and then I realized horse eBooks hasn't tweeted since September.” 9 Jan. 2014, 18:40. Tweet.
Turkle, Sherry. Life on the Screen: Identity in the Age of the Internet. New York: Touchstone, 1995.
Watson, Sara. “Else 9:30: The “Monkeys with Typewriter” Algorithm.” John Battelle’s searchblog 30 Sep. 2013. 23 Mar. 2014 ‹http://battellemedia.com/archives/2013/09/else-9-30-believing-in-monkeys-with-typewriters-algorithms.php›.
Weiss, Astrid, and Manfred Tscheligi.”Rethinking the Human–Agent Relationship: Which Social Cues Do Interactive Agents Really Need to Have?” Believable Bots: Can Computers Play Like People? Ed. Philip Hingston. Berlin: Springer Verlag, 2012. 1-28.
Zhao, Shanyang, Sherri Grasmuck, and Jason Martin. “Identity Construction on Facebook: Digital Empowerment in Anchored Relationships.” Computers in Human Behavior 24.5 (2008): 1816-36.
Zangerle, Eva, and Günther Specht. “‘Sorry, I Was Hacked’: A Classification of Compromised Twitter Accounts.” Proceedings of ACM Symposium on Applied Computing, Gyeongju, Republic of Korea, 2014.