“I’m What ISIS Doesn’t Need”: The MIT Team Behind @DeepDrumpf

donald drumpf

In real life and on Twitter, the one thing you can count on Donald Trump for is saying outlandish things (or typing them out with his little sausage fingers). And after John Oliver’s recent Trump—er, sorry, Drumpf—takedown on Last Week Tonight, Bradley Hayes of MIT’s Computer Science and Artificial Intelligence Lab (CSAIL) used artificial intelligence to create a Twitter bot that would make Trump-isms on its own. @DeepDrumpf, which just started tweeting on March 3, already has more than 14,000 followers.

CSAIL’s communications manager, Adam Conner-Simons, says that Hayes created the bot in just an hour or two. In the simplest terms, Drumpf was born out of “deep learning” techniques that teach computers to identify patterns using systems called “neural networks.” “[Hayes] was inspired by an existing training model that can simulate Shakespeare,” Conner-Simons explains, “as well as a recent report that analyzed the presidential candidates’ linguistic patterns to find that Trump speaks at a fourth-grade level.” (Oof.)

By “feeding” the bot several hours worth of transcripts from Trump speeches and debates, Hayes could give it a sense of the Donald’s linguistic style and his oft-repeated slogans. That’s why even the nonsensical tweets have a nugget of Trumpishness to them, like this one, with its observation that “Our president is Obamacare.”

Of course, this was just an afternoon diversion for Hayes and his team. CSAIL is the largest research lab at MIT, and Connor-Simons says they’re at the forefront of fields including robotics, computational theory and big data. When they’re not spoofing presidential candidates, they’re developing predictive models that help doctors diagnose cancer and creating self-flying drones that can reach speeds up to 30 mph and build real-time maps of their surroundings. Y’know. Casual stuff.

But thanks to Deep Drumpf’s popularity, Connor-Simons says that someday soon, Hayes might roll out bots that mimic the other presidential candidates.

“When the bot has directly responded to Trump’s Twitter account, it’s because Brad gives the algorithm language from the real Trump’s Tweet, which primes it to give a response that is more likely to be contextually relevant,” Connor-Simons explains. “In this way, he envisions developing other accounts for candidates and feeding tweets to each other so that they can all converse in a real-time deep-learning debate.”

Yes, that’s right: We live in a world where @DeepDrumpf could insult the ear size and sweatiness of a Rubio-bot.