MIT Built a Donald Trump AI Twitter Bot that Sounds Scarily Like Him

Republican presidential candidate Donald Trump speaks during a news conference Saturday, March 5, 2016.

Republican presidential candidate Donald Trump speaks during a news conference Saturday, March 5, 2016. Brynn Anderson/AP

The AI was trained on a few hours of Trump’s speeches.

Donald Trump knows many words. He has the best words. He’s going to find the best people to help him run the country. He knows many smart people. Tons. His presidency will be classy, huge, even.

Imitating Trump has become a popular sport as the American businessman leads the race for the Republican party’s presidential nominee and one that even inanimate objects can play. A researcher at MIT’s Computer Science and Artificial Intelligence lab has trained a neural network—a learning AI system modeled after the structure of the human brain that can be taught to find patterns in information on its own—to create sentences based on Donald Trump’s victory speeches in the presidential primary elections.

Creator Bradley Hayes was inspired by a similar neural network trained to simulate Shakespeare quotes, as well as a recent report that Trump generally speaks at an elementary-school level.

“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 a release.

The Trump AI was trained on a few hours of Trump’s speeches, randomly choosing letters to start sentences and then building them out letter-by-letter. As MIT explained in a release, if the bot chooses the letter “M,” it will potentially be followed by the letter “A” and then “K,” until it spells out Trump’s favorite hat-based saying, “Make America Great Again."

Hayes applied the bot to one of Trump’s favorite outlets, Twitter. The twitter bot called @DeepDrumpf (named after a recent John Oliver segment about Trump’s heritage) builds tweets that sound remarkably similar to the staccato style of Trump’s tweets.

The tweets aren’t always perfect—it is a randomly generated string of letters made by a computer, after all. “@DeepDrumpf’s Tweets don’t always make complete sense, but are usually at least partially coherent —much like the candidate himself,” MIT said in a release. Some of them are pretty much perfect, sounding somewhere between how Trump speaks, what what his internal monologue likely sounds like:

Hayes said that he may build bots for the other candidates (including a “DeepBern” for Democrat Bernie Sanders) and have them talk to each other. Perhaps when it comes to the next election, we can just skip the debates, and have the candidates’ bots bash it out Twitter. It’ll save us all a lot of time.