The Rise of Twitter Bots

Last Tuesday, Google decided that I was a spammer, and I lost access to my e-mail for twelve hours. It was my fault. One of my Twitter accounts, RealHumanPraise, was mentioned on “The Colbert Report,” where I work as a writer, at 11:46 P.M. In the course of the next hundred and twenty seconds, it acquired over two thousand followers, triggering an equal number of e-mails from Twitter to my inbox. Notifications from total strangers poured in like Skee-Ball tickets. Around twelve hours and ten thousand followers later, I noticed that I’d stopped getting new e-mails entirely. Google had cut me off, but it felt like a victory. I’d surpassed my allotted share of their massive data centers’ message-processing capabilities, and it was all thanks to my Twitter bot.

Twitter bots are, essentially, computer programs that tweet of their own accord. While people access Twitter through its Web site and other clients, bots connect directly to the Twitter mainline, parsing the information in real time and posting at will; it’s a code-to-code connection, made possible by Twitter’s wide-open application programming interface, or A.P.I. The bots, whose DNA can be written in nearly any modern programming language, live on cloud servers, which never go dark and grow cheaper by the day. This broad accessibility, magnified by Twitter’s laudably permissive stance on the creation of new accounts, has created fertile ground for such automated shenanigans, like proving our susceptibility to certain typos and making Newt Gingrich seem popular.

If you use Twitter, you’ve probably met your share of one kind of Twitter bot, spambots. They have nonsensical burner handles, and blast messages like “truly $$$ work opportunities” at anyone who so much as mentions an iPad. “Click this link,” they beckon, “so I might seize your identity and use it to hawk Panama’s finest generic Ritalin.” Hardly a compelling sales pitch, but at least it explains why they’re always awake.

Of greater interest—the signal to the spammers’ noise—is the growing population of creative bots that consume, remix, and contribute to the broader culture churn of the Internet. Many of them are remarkably productive: Adam Parrish’s everyword, for instance, has been chipping its way through the entire English language, tweeting one word at a time every thirty minutes since 2008. Ranjit Bhatnagar’s Pentametron scours Twitter every hour on the hour and retweets the first rhyming couplet that it can find. Darius Kazemi’s Professor Jocular takes a popular tweet, assumes it’s a joke, then tries to explain why people thought it was funny; the Professor’s more than twenty-three hundred attempts frequently outperform its source material.

Everyword’s specialty is juxtaposition: your friend’s tweet takes on a different flavor if it shows up in your timeline right before everyword tweets out “shithead.” Pentametron has a knack for generating rhymes about trending topics and current events, from sporting-world hyper-masculinity to the Affordable Care Act. A popular new addition is Joe Toscano’s Tofu Product (which Betsy Morais recently wrote about), a bot that talks back to its followers in a peppy bouillabaisse of their own frequently used words and turns of phrase. What drives affection for Tofu is less narcissism than reliable ersatz companionship in Twitter’s crowded, cliquey lunchroom; Tofu Product is everyone’s imaginary friend.

One of my first bots was Exosaurs, which combined Wikipedia’s list of dinosaur species and the Kepler telescope’s list of confirmed exoplanets—both freely available datasets—into an hourly feed of extrasolar mega-reptiles. The bot also credits each Exosaur “discovery” to one of its followers—“ryanpeeler, Gyposaurus of HD 290327 b”—creating a low-grade sweepstakes of speculative biology. When Exosaurs failed to recognize the programmer Ramsey Nasser after a few days, he created the bot “Fuck Exosaurs” to spew profanities at Exosaurs until it awarded him Santanaraptor of PSR B1257+12 c. Soon afterward, the novelist and coder Robin Sloan created Exoriders, which assigns each new Exosaur an intrepid galactic travel-mate, deepening the lore of an accidental universe. Exosaurs now has a community site, a leaderboard, and Exoslash—a bot I made to respond to Robin’s bot with auto-generated Exorider erotica. Richard Dunlop-Waters later made Law & Order: EXO to demonstrate that this kind of one-upmanship can only lead to brutal space murder.

Last month, when the “Add a word, ruin a movie” hashtag raced through my timeline like scabies through a pirate boat, I struggled with the grouchy feeling that, though everyone was having fun (“Friends with DISABILITY Benefits!!! ;)”), the game was too easy. So I built AddAWordBot, which took film titles from the Internet Movie Database and inserted a random word into them; it posts an entry to the hashtag every two minutes, forever. This high-concept thumb-biting failed to register during the meme’s peak popularity, when enthusiastic participants generated up to seven posts a second. But, as the wave broke, my bot’s often incoherent tweets began to crowd out the late adopters, cheerfully refusing to let the game die, blissfully unaware it had become Zombie Ruler of the Humor Desert.

My bot language of choice is Python, a medium as rich with text-manipulating possibility as it is forgiving of sloppy coding chops. There’s also help available: the TextBlob library, for example, can break down paragraphs into sentences and sentences into their component parts of speech—sparing me from having to diagram a sentence for the first time since high school. That part-of-speech data allows me to create on-the-fly Mad Libs, garbling an original sentence by swapping in new nouns, adjectives, and coördinating conjunctions from TextBlob’s limitless supply. TextBlob leverages WordNet, a sprawling database of English word relationships that’s been in development since 1985 and was first funded by the U.S. Navy. We can only be grateful the Cold War ended before they could fully weaponize the synonym.

While an argument bot that I built failed to reach more than twenty-seven followers, my next idea, RealHumanPraise, was handed a megaphone. An upcoming guest on “The Colbert Report” at the time, David Folkenflik, of NPR, had reported on a Fox News P.R. practice of inundating blogs with pro-Fox comments. This behavior felt bot-like, and I wondered if it couldn’t be approximated algorithmically. The plan was to data-mine a repository of positive sentiment (in this case, Rotten Tomatoes reviews), then substitute the names of Fox anchors for those of film actors and change words like “movie” to “news show.” The prototype’s output sold the concept more than a lengthy explanation of bot metaphysics ever could. With Leonard Richardson’s help, RealHumanPraise was set into motion, as was the notification bonanza that kneecapped my Gmail. Choosing to operate at AddAWordBot’s gruelling every-two-minute pace—anything that tweets every two minutes has made a deliberate decision to fly just under Twitter’s spam-detection radar—has meant that the account’s following tapers off by about a thousand weary dropouts a day, but we’re still hitting 564.8 theoretical eyeballs with 282.4 semi-nice things about Fox News each second.

Whether you follow them or not, Twitter bots persevere, infusing one of our era’s great time-wasters with absurdity, serendipity, and chilling reminders of omnipresent surveillance. Twitter bots represent an open-access laboratory for creative programming, where good techniques can be adapted and bad ones can form the compost for newer, better ideas. At a time when even our most glancing online activities are processed into marketing by for-profit bots in the shadows, Twitter bots foreground the influence of automation on modern life, and they demystify it somewhat in the process. Sometimes I wonder how much longer Twitter will tolerate bots’ generative graffiti, especially as the company rushes to justify its I.P.O. with generative revenue dollars.

As a service, Twitter’s greatest strength is that its users have total control over whom they follow and what they see. As a business, Twitter’s greatest opportunity is in violating that control with advertisements. This tension carries the risk of alienating the platform’s most passionate adherents and driving them to a service that resembles Twitter in its quieter, less lucrative years. Twitter is generous with mainline bot-level access today, but there’s precedent for the company getting stingier with that access as they ramp up their revenue machine. Twitter is a publicly traded company now, and it remains to be seen whether their definition of shareholder value includes Mark Sample’s My Favorite Things, which substitutes random words into Oscar Hammerstein’s lyrics while still keeping the song’s rhyme scheme. Strange midsize amateurs, loomed up with sites? These are a few of my favorite kites.

Rob Dubbin writes for “The Colbert Report.”

Illustration: Maximilian Bode