To understand Trump, an AI bot had to be de-programmed from using English grammar. It uses 11 million words from Trump's remarks to tell when he's angry or lying
Donald Trump's unique speaking style — particularly his penchant to go on never ending tangents — crashed an AI robottrained to recognize patterns in speech.
- A Virginia man worked with a scientist with a PhD in machine punctuation to deprogram the bot from using correct English syntax and
grammar, instead programming it to understand Trump, according to a Los Angeles Times profile.
- Margaret, named after the meticulous West Wing character, has since evolved into a sophisticated
AIsystem that catalogues all of Trump's speeches and tweets, using more than 11 million of the president's words since 1976 to understand and predict his speech patterns.
- Amazon uses the bot to help with Trump queries on its Alexa devices, and Margaret can even run analyses on whether Trump is angry or lying based on his tone of voice, pace of speech and gestures.
- Margaret has found he's less stressful when lying, and stops gesturing when he gets mad.
- Visit Business Insider's homepage for more stories.
Whenever it seems like it's too hard to keep track of everything President Donald Trump is saying, now there's a bot for that.
Margaret, named after the meticulous West Wing character, catalogues all of Trump's spoken words, tweets and other utterances to compile in its database — more than 11 million of the president's words dating back to 1976.
A new Los Angeles Times profile of its creator, Bill Frischling, traces the history of the bot, which is used by Amazon for help with Trump-related queries on its Alexa devices.
The AI network is also available online for users to search through.
The bot can predict whether Trump is lying, if he's mad, and how stressed he is whenever speaking on camera.
In Frischling's telling, he faced unique challenges getting Margaret to figure out what, exactly, Trump was saying — much less whether he was telling the truth.
"It was still trying to punctuate it like it was English, versus trying to punctuate it like it was Trump," Frischling told the LA Times.
One passage of a Trump speech was such a tangled mess of never ending tangents and clauses that the bot crashed.
Frischling, a 48-year-old self-taught coder from Great Falls, Va., brought in a scientist with a PhD in machine punctuation to help him out.
They ended up deprogramming Margaret from relying on correct English syntax and grammar, instead focusing on how to uniquely understand Trump.
"Every word he says makes Margaret smarter and allows her to make more subtle distinctions," Frischling said.
By monitoring Trump's stress levels and gestures on camera, Margaret compares his statements to independent fact checks to determine whether he's lying.
Frischling noted that unlike most people, Trump is less stressed when lying.
Studying these kind of tells is increasingly of interest in the AI community, according to Frischling, especially when it comes to world leaders. Frischling added that he also studies members of Congress using the same technology, and shares the findings with private clients.
Trump speaks at 220 words per-minute when he's not reading off a teleprompter, which is close to double the national average for Americans speaking extemporaneously. But he slows down on the prompter to around 111 words per minute.
Unlike most members of Congress, Trump addresses so many topics at once that it's harder to tell when he may change a policy position or forecast other shifts, according to Frischling.
And when Trump gets mad, he simply stops gesturing.
"When he stops making gestures, that's the — watch out," Frischling said. "Whatever's about to happen, hang on to your tush."
- Airlines can now operate 85% of pre-Covid domestic flights, says Aviation Ministry
- Yes Bank-DHFL case: No sympathy, says court, Rana Kapoor's wife, daughters remanded to judicial custody
- Punjab Chief Minister Amarinder Singh resigns amid ongoing turmoil in Congress
- Future of Work 2021: Work from home is here now, but people demand some physical sense at work
- Deadline for PAN-Aadhaar linkage extended till March 2022