Since I don’t have time to meet with all of you in parking garages individually, I’m just going to tell you right here: You know Barack Obama? The “Twitter president?” He’s a robot.
It’s well-known that much of the socializing on social networks is done by bots. A nonzero amount of Twitter’s traffic is just one bot tweeting to all its bot buddies, some of which are sophisticated enough to reply in a semi-coherent way, using natural language algorithms.
Videos by VICE
Bot tech has gotten so sophisticated that not only can they fool each other (which, let’s be honest, don’t seem that impressive), they can fool us. They’re so slick that no one even knows how many bots there are on social media, or how much social media traffic is attributable to them.
The Terminator/Matrix robo-revolution in cyberspace has already begun. To quote a paper from the Center for Complex Systems and Networks at the The Indiana University School of Informatics and Computing, “The future of social media ecosystems might already point in the direction of environments where machine-machine interaction is the norm, and humans will navigate a world populated mostly by bots.”
The same paper admits, “Governments and other entities with sufficient resources have been alleged to use social bots to their advantage.” You know where Barack Obama works? A “government with sufficient resources.” We’re through the looking glass here, people.
But, the IU team isn’t giving up without a fight. Just like John Connor in Terminator 2, they’ve got their own counter-algorithm, designed to unmask those who would impersonate persons. It’s called Bot or Not?, and it’s how I uncovered the truth about our Commandroid-in-Chief.
Framed not as a silver bullet, but only as the next step in an “arms race,” Bot or Not? mines data from 15,000 known social bots and looks at not only their 200 most recent tweets but also the 100 most-recent tweets that mention them, and contrasts this data set with 16,000 known humans.
By examining over 1,000 features of the accounts—the number of tweets and retweets, the number of replies, mentions, and retweets, the username length and the age of the account—Bot or Not? can produce a percentage likelihood that a Twitter account is a social bot.
Image: Screengrab by author
For demonstration purposes, Bot or Not? provides some examples of both bots and humans—the bot @JusBieberPhotos draws an 81 percent. Onur Varol, a human who worked on the project, comes in at 18 percent. Robama pulls a very suspicious 72 percent. You do the math.
You can imagine my shock. I mean, this makes Watergate look like water under a gate. It makes the birthers look even dumber, since it turns out Obama wasn’t even born.
As my palms began to sweat, I wondered if Michelle knew. I ran a check on @Flotus. Her account drew an ambiguous 57 percent. Could go either way, but I’m thinking she knows.
Just to be sure, I also tested a known non-person, @bigbentheclock. It’s a Twitter account that just tweets “BONG” on the hour. It only got 24 percent, and clocks are pretty much the most mechanical thing imaginable.
Already planning my Pulitzer Prize acceptance speech (working title: “I Sang the Body Electorate”) I emailed the team behind Bot or Not?, knowing that, since they had done all the actual work, I’d have to share some credit with them.
Emilio Ferrara, a post-doc fellow at IU who worked on Bot or Not?, explained what was going on. “Highly influential profiles are more prone to be targeted by bots or fake followers (often bots seek to connect to highly popular accounts to ‘inherit’ some visibility). It’s no surprise therefore that Mr. President’s account followers (and other high profile accounts) might result as botlike from this perspective,” he said.
The algorithm looks at user features that distinguish bot from person—bot accounts are generally younger and have longer names. They don’t tweet as much, nor do they get as many replies or mentions as people, but they do retweet more. Reading this, I felt cheap.
“Real quickly, I just want to weigh in on our robot president,” he said.
“In BotOrNot we are also providing subclass scores, so users can see which features of the accounts are more bot likely,” confirmed human Onur Varol told me. “This helps to make a reasoning about an accounts. For example, in my case @onurvarol has very human-likely account, but one of the subclass score is bot-likely (content). This makes sense because most of my contents are academic news or written in non-English language.”
Obama’s content is also one of the most incriminating factors, Varol said, though he chalked it up more to the exceptional nature of the president’s account skewing the algorithm. “Bot-likely subclass scores comes from user, friends and content as a result of popularity of [Obama’s] account,” he said.
Another team member, Clayton Davis, also took some wind out of my sails. “Real quickly, I just want to weigh in on our robot president,” he said. “Given that there is probably an entire office of people dedicated to crafting and strategically releasing each individual tweet according to official style and guidelines, it’s no surprise that this account looks more ‘botlike’ than the average user who mostly shares links and off-the-cuff comments.”
It seems possible, as well, that Obama’s account is a sort of cyborg—sometimes run by a bot, sometimes used the old-fashioned way. Motherboard experimented with a bot for a while with the old @motherboard, before going back to tweeting by hand (er, click?), and it now scores the same as Michelle Obama’s account.
I guess when you get down to it, the Obama twitter account is going to (obviously) not resemble a person. If there was ever a time when presidents were allowed to look like people, it’s in the past. And, it’s important to remember that Bot or Not? is a work in progress.
“The tool is a demo developed in our lab to demonstrate our research on social bot detection. But it is not a commercial system, and it is not a production system. It has known weaknesses,” Fil Menczer said. “Our model was trained with a large dataset of bots collected a few years ago; as such, the tool may do poorly at detecting more recent bots or accounts that are wild outliers compared to our training set (as in the Obama example).”
I’ll admit a lot of the replies from the team behind Bot or Not? were variations on the theme of Barack Obama NOT actually being a robot, and how I probably should stop trying to get a hold of Bob Woodward on the phone, which bruised my ego a little bit.
But there’s another use for Bot or Not? that I discovered: narcissism! I plugged my own Twitter handle into the algorithm and drew a respectably human 39 percent. Apparently I’m usually mentioned with positive sentiment (thank you, yes, yes), but almost all of the stuff I retweet is negative. Also only 3 percent of my content is adverbs. That’s something that I’m going to change immediately.
As for Obama, he already passed the Blade Runner replicant test by talking about his mother, so maybe we can give him a provisional pass for the moment. If anything, this whole endeavor really drives home the complexity at distinguishing bots and people, as well as the simplicity of irresponsible fear-mongering. Which reminds me, Alex Jones, I’m awaiting your call.
