This Donald Trump Twitter Bot Is Surprisingly Convincing

There are a lot of bots out in the Twitterverse now, but only one bot has "the best words" and aims to make America great again. Meet Deep Drumpf, a Twitter bot created by a researcher at the Massachusetts Institute of Technology that pulls text from the speeches and debate transcripts of Republican presidential hopeful Donald Trump. (The name refers to the ancestral surname of the Trump clan, unearthed last week by Jon Oliver.)

We have competence. Our people don't need anybody. I have smart people.

DeepDrumpfMarch 3, 2016

Bradley Hayes, a postdoc at MIT's Computer Science and Artificial Intelligence Lab, created the bot this week using an AI technique called deep learning. The bot creates the tweets randomly one letter at a time but has learned patterns by analyzing Trump's speech. Hayes was inspired by a similar model that can spit out Shakespearean-style text, he said in a press release.

"Trump's language tends to be more simplistic, so I figured that, as a modeling problem, he would be the most manageable candidate to study," Hayes said in the release.

In fact, Trump's language is categorically simple: the Boston Globe recently ran candidate transcripts through an algorithm that determined Trump's speech could be understood by a fourth-grader. Ted Cruz, by comparison, generally speaks at just below a ninth-grade level.

"The algorithm essentially learns an underlying structure from all the data it gets, and then comes up with different combinations of the data that reflect the structure that it was taught," Hayes said.

It can even interact with Trump by pulling words from a recent tweet and replying:

DeepDrumpfMarch 3, 2016

Like many bots, a lot of the tweets wind up in a kind of uncanny zone of almost but not quite making sense. But then there are a few gems, which almost sounds pulled from the lips of the real estate mogul himself:

I'm what ISIS doesn't need.

DeepDrumpfMarch 3, 2016

Hayes said he hopes to create more bots for other candidates, with his goal being to learn through experimentation.

"Much of my actual robotics research deals with these types of modeling techniques," Hayes said. "I thought this would be a good way to learn more about some of the concepts, and have a little bit of fun in the process."