By watching Donald Trump, @DeepDrumpf learns to tweet like him
Follow @DeepDrumpf: An artificial intelligence Twitterbot is learning how to tweet by studying Donald Trump's speech patterns.
Can Donald Trump make artificial intelligence “great again?”
A new Twitterbot is analyzing the way Mr. Trump speaks and using the data to generate tweets. The end result is intended to be an autonomous program capable of crafting tweets that sound like Mr. Trump and it has already garnered thousands of fans.
“Much of my actual robotics research deals with these types of modeling techniques,” Mr. Hayes said in a MIT article about the project. “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.”
Although the project may be entertaining, it could also provide serious results for artificial intelligence.
Behind the Twitterbot is a powerful AI system based on deep-learning neural networks that help computer programs find patterns and learn from them. In this case, the program is finding patterns in the way Trump speaks and mimicking them in tweets. According to MIT, the process starts one letter at a time.
...if the bot randomly begins its Tweet with the letter “M,” it is somewhat likely to be followed by an “A,” and then a “K,” and so on until the bot types out Trump’s campaign slogan, “Make America Great Again.” It then starts over for the next sentence and repeats the process until it reaches the 140-character limit.
Hayes' program is the first to launch a language learning AI of a presidential candidate in a public forum, but similar programs exist elsewhere. At the University of Massachusetts, Amhearst, Valentin Kassarnig developed an AI that could write convincing political speeches. On the West Coast, Stanford University researchers developed an AI that has been gathering patterns and data from the works of Shakespeare and can now mimic his writing.
In some ways, Hayes' chose his program to be less ambitious. Trump was selected after The Boston Globe studied the speaking habits of 19 presidential candidates and concluded that Trump spoke in the simplest patterns.
“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.
But Trump's simpler speech patterns haven't resulted in a perfect mimic.
The program was initially "taught" by analyzing a few hours worth of speeches from Trump and then linked to the official @realDonaldTrump Twitter account, so it will receive more data as Trump continues to tweet. With time, the program should improve.
Hayes said in the MIT news release that he envisioned creating Twitterbots for other candidates in the future. How well AI programs will be able to mimic speaking patterns and how that technology can be applied for other uses than Twitterbots or new Shakespeare sonnets has yet to be seen.
Mr. Kassarnig told MIT Technology Reviews that politicians were unlikely to adopt his AI program for speechwriting, but one day the technology could be applied to writing articles or blog posts.