The best ideas are born out of spite. After I taunted my friend Brett O’Connor mercilessly with tweets during the Super Bowl (he’s from Denver), he responded the only way he knew how: by creating a Twitter bot.
This is how @knguyen_ebooks came to be. To build a Twitter bot takes a little bit of programming know-how, but it’s surprisingly unsophisticated. @knguyen_ebooks is built on a framework from an Australian developer who goes by “mispy.” Similar to its namesake @horse_ebooks (before its tweets were written by a human), it uses Markov chain, a mathematical system that is “memory less.” With text, it chains together words based partly on randomness and partly on what is likely to follow a given word. The result, when applied to a large corpus of text (in this case, my last two thousand tweets), is an output that often sounds like a bizarro version of me.
Girl, you probably think this subtweet is about you.
— knguyen ebooks (@knguyen_ebooks) February 12, 2014
2014 prediction: NYT piece about a $219 webinar on Excel PivotTables.
— knguyen ebooks (@knguyen_ebooks) February 19, 2014
Just misread "sports bra" as "sports bro" and now I can't tell who is online and who isn't white/doesn't live in NY?
— knguyen ebooks (@knguyen_ebooks) February 6, 2014
It’s startling how easily I can be imitated by a Twitter bot, hosted on a Raspberry Pi, a $25 computer the size of a credit card.
According to Brett, @knguyen_ebooks has a completely random chance of doing one of the following: tweet every 30 minutes, reply to someone it follows, reply to a mention, and continue to reply in a conversation. I’ve retweeted @knguyen_ebooks a handful of times, and it’s convincing enough to make people think, at least for a second, that the bot is me.
— Aaron Lynch (@lynchmonster) February 4, 2014
— Sean A. Higgins (@luckycloud) February 5, 2014
— Garrett Miller (@heyitsgarrett) February 6, 2014
Bots that imitate humans are nothing new. @tofu_product, a bot created by Joe Toscano last October, “absorbs flavor,” meaning it replies to your tweets in a voice resembling your own. The bot attempts to be conversational by picking up key words in your reply, but the responses rarely make convincing sentences. Therein lies the humor of @tofu_product: it sounds like a broken, mangled mirror of yourself. Rob Dubbin’s bot @oliviataters tweets like a teenager by searching for phrases surrounding “is literally” and “was totally” and splicing them into new sentences, convincingly enough that on a handful of occasions it (she?) has gotten into Twitter spats with users who believe she is human.
But how important is a bot’s ability to resemble a human? Does it matter if we know it’s a bot or not? Whether a bot could pass the Turing test seems beside the point. Before the reveal that @horse_ebooks was not a bot, the charm of the account was the way it strung together poetic and cogent tweets from what appeared to be an algorithm; the same is true of @knguyen_ebooks. The fascination of these types of bots is not what they say, but how people interact with them. Since a bot’s output has no intention or motive, the meaning is what a human takes from it.
Having done web programming for so long, Brett wanted to try something completely different with his skills. So much of web development is focused on building tools or services that perform a specific useful function. With bots, they exist simply to delight and annoy. Their purpose (if we’re being liberal with the definition of “purpose”) is to evoke an emotional response. This sentiment is echoed in a Boston Globe profile of Darius Kazemi by Leon Neyfakh. He writes, “By imitating humans in ways both poignant and disorienting, Kazemi’s bots focus our attention on the power and the limits of automated technology, as well as reminding us of our own tendency to speak and act in ways that are essentially robotic.”
Kazemi is perhaps the godfather of Twitter bots (or at least the person who has probably built the most). His work includes bots that make puns out of trending Twitter topics, combines news headlines, and generates games of Fuck, Marry, Kill. Brett cites Kazemi as an inspiration for the four bots he has created, including one bot imitates himself as well.
“I’ve called my bot an ‘algorithmic mockery of my life’ before because every once in a while it will assemble some of my tweets into something that seems like a criticism or brings back a bad memory,” Brett says. “Of course this is all just me reacting to noise and garbage. I sometimes think it’s like ‘internet training,’ because humans make noise and garbage on the internet too.”
In following @knguyen_ebooks, I’ve recognized patterns in my own tweets. I talk obsessively about the same subjects (football, books, videogames) and too often use the same joke constructions. In a lot of ways, @knguyen_ebooks is funnier than I am because it heightens the tedious things I say by breaking the expectations of those who are familiar with my Twitter account.
When friends have asked me about my bot, I’m reminded of an interview with Larry David. Not surprisingly, David says the question he’s asked most often is how closely he resembles the character in Curb Your Enthusiasm. Every time, he says the character is not necessarily him, but the person he wishes he was. Larry David the character is funnier and weirder and more surprising than the real one.
A sentence is perhaps the simplest way to express an idea. It’s this understanding that makes Twitter’s limitations so powerful. All tweets are presented uniformly in a stream — chronologically, showing no preference (except to ads I suppose) — so the voice of an individual, a news organization, a corporation, and a robot are all on equal footing.
And yet, Twitter’s lack of context forces the user to create context. By design, Twitter gives us so little to go on that we must be conscious of our internet literacy. When an unfamiliar person follows us, we check to see if it’s a human or a spambot; when a famous person dies, we make sure it’s being reported by a credible news source. Twitter’s power comes from this tension between the lack of context and our ability to discover it. @knguyen_ebooks is only funny if you’re familiar with @knguyen. Curb Your Enthusiasm only makes sense if you know who Larry David is.
Still, the contextual relationships we have to Twitter don’t necessarily have to be true. Since my Twitter handle is my first initial followed by my last name — the 57th most common surname in the U.S. — I often get replies that are meant for a different K. Nguyen. A couple years ago, a college student from South Carolina began regularly tweeting insults at me, which I later realized were meant to be directed at a different Kevin Nguyen. When he figured it out, he decided to continue following me, occasionally retweeting me to his group of friends. I was then referred to as “Bootleg Kevin,” a sort of alternate-universe version of their Kevin Nguyen.
It was all in good fun. Apparently I resembled the complete opposite of their friend — a liberal who doesn’t regularly quote Family Guy and binge drink until he blacks out — and they would joke and retweet whenever I would say something that betrayed the reality of their Kevin Nguyen. Occasionally, I was tempted to tweet back at them, but I didn’t want to break the role they’d created for me. It didn’t matter that I was a real person; to these college kids, my account was an inside joke, a context they had invented, one that was funny and personal and delightful to them.
Illustrations courtesy of Bob May