Trompe l’Tweet: The Twitter-Bot World @Horse_ebooks Left Behind

Zeuxis and Parrhasius. Image via the NYPL Digital Gallery.
Zeuxis and Parrhasius (image via the NYPL Digital Gallery)

There’s an Ancient Greek story that many art lovers know, from the 5th century BCE. Zeuxis and Parrhasius were known as the best painters of the time, so the citizens held a contest to determine who indeed was the best. At that time, the value of painting was in its re-creation of reality. A painting that could complete a trompe l’oeil, a fooling of the eye, was deemed the most successful.

As the story goes, Zeuxis painted a still life with grapes so successful that birds came down and pecked at them. Parrhasius had hid his painting behind curtains and asked Zeuxis to pull them back and see what he had made. But when Zeuxis reached out, it turned out the curtains were the painting. He immediately had to concede. Zeuxis had fooled the birds, but Parrhasius had fooled the artist himself.

Whether or not the story is true, it suggests a lot about how we value human skill in Western society. Although contemporary art has deconstructed painting to the point of abstraction, the form’s roots remain in this basic idea of faithful re-creation of reality. I thought of that last month, when the American Twitter community reacted to the news that @Horse_ebooks, one of the most popular bots on Twitter, was actually reported by the New Yorker to be a performance art piece by Jacob Bakkila, a creative director at BuzzFeed, and Thomas Bender, a former vice president at Howcast. Unlike Parrhasius, they had fooled not just the artist, but hundreds of thousands of devoted followers:

What’s more interesting to me here is the belief more than the betrayal. When Alan Turing devised his famous test to determine if a robot was acting like a human, he never developed another test — to see whether a bot really was a human. Certainly it seems like a less interesting puzzle; any actor could accomplish that pretty easily, and two actors obviously have. But I wonder if there’s also a basic assumption there. In the 1950s, it was probably impossible to imagine that we would ever care if a bot was being played by a human. And yet here we are:

The World of Twitter Bots

To understand the collective response to Horse, it helps to understand the wider world of popular Twitter bots from which it emerges. It’s a small one, but still much larger than many think, and the Twitter bots themselves emerge from an even wider world of chat bots on sites like IRC, an early chat system, and instant messenger; SmarterChild is probably the most famous of these. What makes Twitter interesting, though, is the simple follower metric. A chat bot has a few interactions, but there’s no way to know how popular it is, per se; the interaction is largely one-to-one or one-to-few. But on Twitter, it’s easy to see that a number of bots have more followers than most human beings, and with that knowledge comes bot communities.

Many of them have a basic shtick, like @yourmombot (~26k followers), which takes tweets at random and adds “Your mom” to the beginning, with surprisingly comic effect. There’s @capscop (~22k followers), which polices tweets composed entirely in caps. More algorithmically sophisticated ones include Ranjit Bhatnagar’s @pentametron (~7k followers), which retweets tweets unwittingly composed in iambic pentameter and turns them into sonnets, rhymes, and all, while @anagramatron (~2k followers), developed by “Colin R,” hilariously hunts down and retweets accidental anagrams on Twitter. Here are two recent retweets from @pentametron:

With a short bit of data analysis, I also found a few surprises. Whereas many of us might be accustomed to swatting away spam bots from our Twitter accounts, some of them have drawn a particular sort of affection from a subset of Twitter. @_spell (~15k followers) is frequently called on by folks who want to double check their writing (type “sp?” in a tweet, and it will help you figure out if the word just before is correct or not). @fart_robot (~10k followers) marks its approval when you type the word “fart,” which almost invariably brings a giggle to those who have been retweeted. Even @stealthmountain (~26k followers), which corrects people when they misspell “sneak peek” as “sneak peak,” receives replies of gratitude (“thanks” and “haha” came up most in my casual data analysis) rather than anger.

Not surprisingly, popular Twitter bots seem to be largely a Western and Japanese phenomenon, and they resonate with certain groups of people within those cultures. They tap into a peculiar sensibility that’s difficult to explain to those on the outside. Like cat videos and Onion articles, these bots bring a spot of shared joy to our lives, a sort of insider humor for tens of thousands of people and an impenetrable mystery to everyone else. But unlike the former two, Twitter bots bring joy because they are bots, and we know (or think) them to be such. Their scripts were created by humans, but their interactions are outside the makers’ control.

Horse: The Bot of Bots

Horse eBooks, on the other hand, has always been a mystery. It is, perhaps, the first internet bot companion that reached something like mainstream appeal. And by this I don’t mean sheer follower count, though it’s garnered a stunning more than 200,000. There are bots with more or nearly the same number of followers: @big_ben_clock, which tweets “BONG BONG” on the hour, has nearly double the amount, with over 400,000 followers, and @urbandictionary, which responds to questions about slang, is a pretty close contender, with over 175,000 followers.

But no English-tweeting bot that I’m aware of has inspired as wide a variety of fan fiction, a book of poetry, T-shirts, articles in mainstream press, entire comic communities on Tumblr, and spin-off x_ebooks (eg: @nytimes_ebooks, @bogost_ebooks) accounts, which were created by supposedly reverse engineering the @Horse_ebooks algorithm. In this regard, Horse’s followers elevated it to the status of a micro celebrity — not quite a Kim Kardashian (nearly 20 million followers) or even a George Takei (close to a million), but a lot more than most public intellectuals, writers, artists, and wannabe stars. Horse became a constant companion in our lives, a friendly bot amongst our human friends and followees, spitting out a nonsensical spammy phrase here and there with an ambient presence.

One of Burton Durand's Horse_eComics (via
One of Burton Durand‘s Horse_eComics (via

That @Horse_ebooks’ true human nature broke so many hearts is obvious by a quick scan of tweets and eulogistic blog posts. This act of digital trompe l’oeil touched even more people than another famous bot, Greg Marra’s @track_girl. Marra’s bot, which cleverly repackaged tweets from runners to create a human-like stream of one determined runner, triggered responses from a few concerned followers when she apparently broke her leg. But she was never quite as popular.

Horse’s demise, if we can call it that, elicited more anger and howls of frustration than Marra’s ever could. The bot we loved most wasn’t a human being played by a bot, embodied in a young runner, but a spambot played by two human beings, embodied in a horse. The Turing Test still has applications in the world of artificial intelligence, but it seems inevitable that a reverse test — a Horse Test, if you will — will forever be applied to lovable bots from here on out. The Horse Test will be the language of the quirky corner of the internet that loves bots for what they are.

A salvo here is that Horse did, in fact, start as a spambot. Bakkila and Bender, perhaps seeing an opportunity, purchased the account from the original creator, Alex Kouznetsov. As they noted in a recent interview with the New York Times, they began creating tweets in the voice of a spambot, researching texts and constructing them like a bot would. Horse’s enigmatic origins remain in the hands of a (computer) script, not a scriptwriter.

Curtains and Grapes

Bakkila and Bender tapped into what so many of us wanted: a silly, accessible robot friend. Not an intellectual equal to have conversations with or an emotional equal we could fall in love with, not a bot that could trick us into thinking it was human. We wanted something more fun and silly, something more akin to a pet rather than a peer. Horse became an accidental poet-philosopher with no purpose but to make us giggle, smile, and think.

A quick search on Topsy of Horse’s most popular tweets helps us understand the appeal of the two artists’ work. Their most beloved tweet — “everything happens so much” — got over 8,000 retweets and 5,600 favorites. “Dear Reader, You are reading” got 2,000 retweets, and “Unfortunately, as you probably already know, people” got 3,955. They’re the kinds of tweets that are too good to be true, as Mashable suspected in 2012, but they clearly resonated with thousands of people. Like the work of jesters and street preachers, they’re more powerful because we take their origins less seriously.

But let’s step back a few more years, to the period before September 2011, when the duo say they took over the account. Kouznetsov’s script was already ascendant, and this is what our human friends tapped into. There are simple, mysterious tweets like “Famous Crab” (405 retweets) and “Frog” (612). And we have amazing ones about horses, like “How to throw a horse” and “Teach a Horse How to Bow, Lie Down and Shake Hands.” The charm of the imagery and language is contagious.

We can read Bakkila and Bender’s performance as a hoax, but I prefer to read it as an homage. Not to nature, as Zeuxis and Parrhasius once did, but to the beauty of the machinery and computer scripts that surround us. Two tweets from the original @Horse_ebooks bot stand out in particular. They seem like prophecies for this three-year whirlwind relationship with our robot friend; they are the grapes and curtains from which Bakkila and Bender took their cue.

The artist can recreate reality, but reality, in the end, is infinitely more interesting.

comments (0)