As Donald Trump leads the race for the Republican party's presidential nominee and continues to divide Americans, even those within his own party, MIT researcher Bradley Hayes created an AI Twitter bot that is starting to sound just like the real estate mogul.

Hayes created the bot at MIT's Computer Science and Artificial Intelligence (CSAIL) lab by training a neural network, which is a deep learning AI system that is designed after the unique structure of the human brain, allowing it to find and learn patterns in information on its own. The final product is an AI that can create sentences that mimic Trump's speaking style on its own.

"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," said Hayes, who was inspired by a similar neural network that was trained to simulate Shakespeare quotes.

The bot, named "DeepDrumpf" after the John Oliver segment that pokes fun at Trump's ancestral name, creates Tweets one letter at a time, computing each subsequent letter using its neural network until it creates a full sentence.

Hayes trained the AI on just a few hours worth of Trump's speeches, which it uses during the learning process, and although many of the tweets could pass for Trump's own, others are a bit rough:

Of course, this will improve with time, which is one of the great parts about AIs that utilize deep neural networks - they continue to learn and adapt with time, getting more and more effective as they acquire experience. This is clear when looking at the first few tweets from the bot that appear to be randomly generated all the way to the newer ones that have progressed into fully formed opinions that mirror Trump's own.

Hayes says that his future plans include bots for other candidates, including a "DeepBern" for Democratic presidential hopeful Bernie Sanders, and he hopes to have the different bots talk to each other.

@ 2024 HNGN, All rights reserved. Do not reproduce without permission.