Hot Bots: An Interview with Leonardo Flores

by Jessica Lauer


Hi Leo! What was/is it about bots, particularly Twitter bots, that initially sparked your interest?
I discovered Twitter bots during my initial I ♥︎ E-Poetry run back in February 2013 while I was doing a series on electronic literature in social media networks. I was aware of bots, but I hadn’t given them much thought. The first bot I reviewed for that series was Bruno Latourbot, which performed a kind of cut-up operation on some of Latour’s writings, followed by Darius Kazemi’s Latour Swag and Mark Sample’s Walt FML Whitman, both of which remixed writings from the two renowned authors with content from the Twitter stream. And I was hooked!
Bots actually have quite a long history. Were you always interested in them?
They do, but I’ve never been very thrilled by earlier bots, partly because they were so limited in their data sources and goals. ELIZA, PARRY, and other bots were so focused on passing as humans via the Turing Test that their output strove for realism and wasn’t very interesting to me from a poetic perspective. But these Twitter bots were doing things that were in tune with contemporary poetry and poetics—conceptual writing, flarf, Language poetry, e-poetry, even a kind of robotic twist on Post Confessionalism (see The Way Bot)—and they were a grassroots movement. The bot makers weren’t necessarily approaching their creations from an academic or literary perspective and that was refreshing because knowing too much about it can inhibit creativity or shape it along more traditional literary concepts. More importantly, it was and continues to be a growing grassroots scene, with bots created at an accelerating rate that is hard to keep up with.
Are you able to keep up?
LOL. I try. Back when I was posting daily in I ♥︎ E-Poetry, I kept writing about more and more bots, to the extent that a few of my friends thought they were going to lose me to the bots :-). Since then, I’ve dedicated a good portion of my research to Twitter bots, joined several bot maker communities, and worked actively to invite bot makers to Electronic Literature Organization conferences by organizing panels and readings. My most current bot-related work is a monthly column in Bot Watch titled "Interview with the Bot," in which Vivian Sming and I take turns responding to questions posed by Tully Hansen’s Twitter bot Naïve Bot Questions.
How many bots do you follow on Twitter?
I’ve lost count. Must be several hundred bots. My Twitter stream moves at a rapid rate, and it’s an inspiring mélange of artistic and human tweets. This vibrant Twitter bot scene never ceases to surprise and delight me. And it is attracting mainstream attention, which is also fascinating as an indicator of increasing digital literacy. Bots are a timely artistic and literary phenomenon because it produces art designed for endless streaming social media environments.
Some of the Twitter bots out there seem like they fall into the category of e-poetry. What are some similar characteristics you see between Twitter bots and e-poetry?
If we define e-poetry as the poetic engagement of digitally mediated language, then bots created for a text-driven social media platform such as Twitter have much to offer.

For starters, Twitter’s 140 character limit already drives textual production towards poetic levels of compression, which has inspired many people to write Twitter fiction (#twitfic), six word stories (#6wordstory), micropoetry (#micropoetry), haiku (#haiku), tanka (#tanka), and others. Twitter’s hashtags frequently become social games that lead to poetic engagements of language, such as #lesserfilms, #MillenialBooks, #in4words, and many more.
So is poetry written for Twitter considered e-poetry?
Not really, but it’s a step in the right direction because it results in poetry that may not have been written if not prompted by the social network’s affordances and constraints. I think a deeper engagement with digital media is required for it to be considered (or interesting as) e-poetry, engaging computation, interactivity, or multimedia presentation.

For example, there’s a bot titled “In X Words” (@inXwords) which detects trending hashtag wordplay and contributes to the conversation with phrases from Project Gutenberg that meet the constraint. The results are surprisingly effective because the hashtag offers a frame of reference for the readers to interpret a brief set of words, transforming non-sequiturs into metaphors, metonyms, symbols, and other figures of speech. A common strategy among bot makers is to identify a formula and find ways to explore them with a bot.
Are those hashtags e-poetry?
Not necessarily, but as popular language-based games in social media they’re a step towards e-poetry, and they offer templates that can be engaged with computation.

Traditional poetic forms are also very suitable for algorithmic exploration. For example, one can use online dictionary services—such as Wordnik, GCIDE, CMUdict, and The Engish Lexicon Project—to discover syllable count, stress patterns, rhyme, synonyms, antonyms, and other information and produce poetic forms, including meter and stanzas structures. Two bots that successfully use such services to discover and produce poetry from the Twitter stream are “Pentametron” (@pentametron) and “HaikuD2” (@HaikuD2). The links above lead to more detailed explanations of their mechanics and poetics, but in a nutshell, “Pentametron” detects and retweets rhyming tweets in iambic pentameter, while “HaikuD2” identifies, formats, and re-publishes accidental haiku from the Twitter stream.
How else do bots create e-poetry?
Bots also produce e-poetry by generating and reframing language in ways that draws attention to the language itself (what Jakobson (1960) theorized as the poetic function of language), inviting readers to re-examine it with fresh eyes. Think of the examples just provided. What a wonderful discovery to see that one’s tweet just happened to be in iambic pentameter, or that it could be reformatted and read as a haiku! Pentametron goes a step further by pairing the tweet with one that rhymes with it—effectively generating a couplet, and inviting readers to read them as such. This juxtaposition is an important poetic strategy for bots such as “Dreams, Juxtaposed,” “And Now Imagine,” and “Two Headlines” (all three of which I discuss in this I ♥︎ E-Poetry entry), because they create cognitive dissonances, surprising convergences, and conceptual metaphors in the language they bring together.
What do you mean by this?
Let me explain with an example of a bot I’m kind of obsessed with lately: Regrets to Egrets (@regrettoegret) by Tobi Hahn. This bot performs a simple set of operations: it detects tweets with the words “regret” or “regrets,” copies its text into a new tweet, and removes the letter “r” from “regret(s).” The result is language written for one frame of reference—regrets, guilt, remorse, and the past—used for a completely different one—egrets, waterfowl, rivers, nature. A snapshot of its profile has a few examples, including its description.
Screenshot of Regrets to Egrets twitter account page; picture of flying egret with profile description 'no egrets, just lessons learned.' Tweet text 1: 'live life with non egrets', Tweet text 2: 'No egrets today'.
Regrets to Egrets Twitter account page
Notice how the statements make sense, but in a completely different way than intended, because they were conceptualized as discussing regrets, not egrets. One can find both poetry and humor in the mental gap between the original and generated statements. The bot will detect many statements that don’t make much sense—particularly when regret is used as a verb—but its best moments are when it generates surprisingly coherent statements, such as this one:
Screenshot of Regrets to Egrets twitter post with text: 'uni is ordering pizza at 11:30pm with no egrets'
Regrets to Egrets Twitter post
This tweet cracks me up! :-) The regret of ordering (and eating) a pizza late at night is transformed into egrets as an ingredient that’s not included in the pizza (but available on the menu). The operative poetic device here is a conceptual metaphor, defined by Lakoff and Turner (2009) as “metaphors allow us to understand one domain of experience in terms of another” (p. 135). The cognitive mapping from one domain to another happens at the level of the bot’s name and profile, and the tweets offer mappings, allowing readers to work out their logical dissonances and convergences. The bot is basically exploring a metaphor over time, inviting its followers to discover its poetic range.
I see. Are there any other e-poetic strategies?
Certainly, there are many, but I’m still gathering information about bot e-poetry and don’t feel quite ready to offer a complete framework of practices. Part of what I love about this genre is that it continues to surprise me with its versatility. My Genre: Bot resource in I ♥︎ E-Poetry offers an initial categorization, connecting to entries on over 50 bots which analyze them from an e-poetic perspective. And even this pooling of bots caught in the Twitter stream has attracted some regrets.
I saw what you did there.

In the meantime, I recommend reading Harry Giles’ (2016) recent article "Some Strategies of Bot Poetics," which does an excellent job of identifying strategies, discussing text sources for bots, and differentiating them from poetry in print. The article is grounded in poetic, artistic, and bot making practices and offers concrete examples of bots.

And when reading a bot, think about how it invites you to examine language with fresh eyes.
The Hostos Bot 1.0 your digital humanities internship created was really cool, and I have a couple of questions about it:
First, what made you decide to go with the works of Eugenio María de Hostos?
Several reasons informed this decision. Eugenio María de Hostos is a key figure in Puerto Rican letters, whose writings and work have had such a positive impact in Puerto Rico and Latin America that he’s known as “the great citizen of America.” As a native of my hometown, Mayagüez (his birthplace, now a museum and library, are 10 minutes away from my house), I was drawn to this prócer (benefactor) whose work had a global impact. When we started investigating his work, we found that there were no electronic editions of his books (now there are, fortunately). The only material available online was scanned PDFs of old editions of his books at the Biblioteca Virtual Miguel de Cervantes in Spain. So, in order to draw attention to his work in a creative way, we created this odd little poetic bot.
Any metaphorical meaning(s) or symbolism behind the concept of remediating the works of Hostos, a pioneer in Puerto Rican education as well as an advocate for Puerto Rican independence?
I like to think of this bot as Hostos’ digital ghost, who would’ve been simultaneously delighted to see how far Puerto Rico has come along in providing access to higher education, and horrified by how our colonial relationship to the United States is currently playing out. The bot cannot express these opinions: but readers can read its output and interpret it from the frame of reference of Hostos’ thought. I think it’s fitting for Hostos to haunt our digital spaces and remind us of what he fought for.

As a project undertaken in the context of a course titled Digital Humanities Internship at the University of Puerto Rico: Mayagüez, this project is well attuned with Hostos’ philosophy. He advocated for “the integrated study of different disciplines and crafts, elevating authentic learning experiences over memorization, regulating the number of students in the classroom, and basing educational processes on the scientific method’s principles of observation, evaluation, experiment, and subsequent modification of propositions/hypotheses” (Cooper, 2016). Hostos would approve of this project, produced in an interdisciplinary course taught at a highly subsidized public University of Puerto Rico, in a class with six students working collaboratively in a project that required them to combine their humanistic training with computer programming to produce all the digital objects that constitute a bot.
Going into 1.0, what were some learning goals you had for the students participating? Anything you are going to change for 2.0?
This assignment served several learning goals for my students. They got to research Eugenio María de Hostos’ work online, finding his work in the digital library in Spain. The learned about issues concerning digitization of old books, which results in images of text, but not text itself. They got to use Google’s Tesseract OCR Engine, to convert it to text, and then used algorithmic and manual methods to clean up the text of each book to produce documents that only contained Hostos’ writing, the textual corpus for the Hostos Bot. As we prepared this corpus, we had discussions about signal to noise ratios and glitch aesthetics. We also discussed and fine tuned the Markov chain generator that would probabilistically generate new texts from the corpus. We produced the Hostos Bot Tumblr, Twitter, and Facebook accounts and profiles, each of which requires consistent art design with different parameters. Along the way, the students got to familiarize themselves with Eugenio María de Hostos and his work, and how the bot revealed his different voices as a writer of sociology, educational treatises, fiction, and intimate letters. It was a wonderfully challenging experience that resulted in a work of electronic literature we are proud of, glitches and all.
Anything you are going to change for 2.0?
The 1.0 version runs locally on my computer. Every so often I need to wind it up, so to speak, and laboriously fill its Tumblr queue, where it posts on a schedule republishing its content (imperfectly) on Twitter and Facebook. So the first thing I want to do is to host it online so it can run independently, posting output directly to Twitter, Tumblr, and Facebook in a way that it’s optimized for each social network. I also want it to be interactive, particularly in Twitter, so that it can respond to questions and comment on headlines from Puerto Rican news sources. This functionality would help the Hostos Bot resonate with what is happening in Puerto Rico rather than being just an e-poetic exploration of his voice as a writer.
In a Boston Globe article about bots, TV writer and bot-maker Brett O’Connor is paraphrased as saying about computer programmer and fellow bot-maker Darius Kazemi that Kazemi’s "bots are not speaking from emotion, but they can say things that trigger emotional reactions” (quoted in Neyfakh, 2014). What reactions do you hope some of the Twitter bots you've created (or your students have created) have sparked?
The main reactions I seek to evoke with my Twitter bots are surprise and delight, though I know of bots that evoke shock, anger, laughter, sadness, thought, and other powerful emotions in their audiences. The fact that they’re able to evoke reactions attests to their power as art forms. Also, while Kazemi’s bots may not be emotionally driven, that is more of a reflection of his poetics. Other bot makers use more sensitive topics or materials, such a Mark Sample’s protest bots (see Sample, 2014). His NRA Tally bot uses real data to generate hypothetical mass shootings and NRA responses, seeking to provoke outrage at the gun lobbying organization. Sui sea’s bot ☆ you are strong ☆ promotes wellness and self esteem, as does @chevalier_cygne’s validation flowerbot. One of my favorite bots created by two Digital Creative Writing students of mine is A Tiny Zoo, which uses Unicode characters and emoji to generate friendly little zoos filled with animals and visitors. I mean, look at this adorable little zoo.
Screenshot of multiple human and animal emoji arranged in a zoo
A Tiny Zoo emoji zoo
In the Bot Collection you created, one of the bots highlighted is @everyword—a bot that tweeted every word in the English dictionary, every day between 2007 until 2014, when it ran out. In your editorial statement, you mentioned the interactive opportunities for followers of @everyword, such as the words becoming recontextualized as it appears in each user's individual timeline, as well as words spurring Twitter discussions. @everyword was also published into a PDF book. What unique interactive opportunities does a PDF provide that are different from the digital, Twitter bot version? In short, what is the point of creating a PDF?
The print and ebook version of @everyword is the culmination of the bot driven project because it documents the Twitter publication and reception (in Favorites and Retweets) for every word. It also lists the most favorited and retweeted words (the top three are “weed,” “vagina,” and “sex”—make what you will of that). The book has a great introduction by its author Allison Parrish (2015) that discusses, among many things, the computational plot twist at the end of the bot’s mission. Most importantly, it comes full circle back to what informed the original word listing: a book.
In your interview with Litteraturtidsskriftet LASSO, you mention that the ♥︎ symbol “is a graphical element that has entered language at a foundational level." Any other symbols come to mind that have made a similar entry in the last decade or so into all (or most) languages?
If you don’t mind, I will respond to this question with emoji. ℹ︎ ❤️ 💻 ☞ 🤖 👍 👀💡💾 🇵🇷
If you could use the archives of any celebrity Twitter feed to create a bot with, whose would you use and what would you want the bot to look like or do? (Ex: KimKierkegaardashian)
Hmmm. The tricky thing about Twitter feeds is that they tend to be as varied and inconsistent in their writing tone and topics. To have a good corpus to work with, you need a consistent performance of a celebrity: something that can be elevated to an idea that you can amplify, modify, or push against. So I think I’d go for someone like Donald Trump’s Twitter feed, which is so absurd in its ideas and slogans that it’s just begging to be mocked or somehow subverted. How to do so effectively also has its challenges, but a first stab could be to run it through a Markov chain generator, or invert its syntax to produce Yoda-like language, or run an Oulipian N+7 operation, translate it into another language and back to English to create weird synonyms, or cut it into lines to transform it into different poetic forms, or really any operation that can stress test his language and draw attention to its ideology and absurdities.

Bots are quite good for this kind of activism, as can be seen with Erowid Sarah Palin which uses a Markov chain generator to mash up Sarah Palin’s speeches and erowid trip narratives, satirizing her bizzarely, incoherent speeches.
Screenshot of Erowid Sarah Palin Twitter profile page
Erowid Sarah Palin Twitter page
Thanks, Leo, for a very fun and insightful interview!
My pleasure, Jessica! And don’t forget to check out the Bot Collection in the Electronic Literature Collection, Volume 3. Shall we continue this conversation in Twitter?