Shulou(Shulou.com)06/02 Report--

Big data Abstracts by Liu Junhuan, Bing

Recently, John Bolton, a former national security adviser, published his new book, The Room Where It Happened, in which Bolton wrote that Trump is easily influenced by authoritarian leaders and is often despised by his own advisers.

Trump could not sit still in the face of this "accusation". In an interview with the Wall Street Journal, Trump said directly that "Bolton is a liar" and "everyone in the White House hates him."

Trump's statement has not yet been responded to by a Bolton spokesman.

Confusing remarks like Trump are common to people all over the world, but have you ever wondered what it would be like to use AI to identify Trump's "Sichuan dialect"?

Don't worry, let's go back to three years ago, when Trump was giving a speech on World War II, shortly after the advent of Trump-style English. its wild choice of words and sentences, randomly switching tenses and subjects, distorted grammar, conflicting conjunctions, flying prepositions, clauses, parentheses, parentheses, and somewhat inadaptable.

Maybe the American audience was deceived by it.

But it was then that Bill Frischling began asking his AI robot, Margaret, to transcribe a 127th-word passage from Trump's speech, and when Trump mentioned "winning" for the fourth time, Margaret crashed. "it's still trying to punctuate the passage as it does with ordinary English, rather than with 'TrumpEnglish,'" Frischling recalls. "

Frischling, director of the artificial Intelligence and Robotics Laboratory at the University of Colorado, is a self-taught programmer who worked at home in Virginia during the outbreak.

As a result, Margaret had to reset the Trump model to learn Kawagawa dialect from scratch, so Frischling hired a computer expert with an auto-punctuation doctorate to teach Margaret to "forget" normal grammar and learn to analyze Trump's speech instead.

Is Trump angry?

First of all, we need to correct a claim that Trump is not the first US president to learn from AI. Many scholars have used AI to analyze Lincoln's Gettysburg address from various angles. Although the whole speech has only 272 words, it can be said that related research books can fill the whole library.

AI is good at dealing with "orthodox" US presidents, but is Trump an ordinary person? Therefore, it is very difficult to learn "Sichuan dialect".

For example, Trump held a briefing on novel coronavirus at the White House on April 23, when Trump suggested that doctors should consider injecting patients with household disinfectants to kill the virus. Fifty-two minutes after the meeting, Washington Post reporter Philip Rucker was questioning Trump whether he admitted to irresponsibly trumpeting unconfirmed news that heat and sunlight could cure the coronavirus.

Trump hit back: "I am the president, and you are making fake news!"

Let's move on. Trump had an unforgettable stand-off with CNN's Jim Acosta at a news conference after the 2018 mid-term elections, when Republicans suffered heavy losses.

At the beginning of the conversation, Trump yelled "enough" when he put pressure on him on Acosta over immigration. Then he stepped down.

Did Trump really get angry at both conferences?

Margret's answer is that Trump was really angry at COVID-19 's briefing and Trump was really angry at the 2018 news conference, but when he stepped down, his mood changed "significantly" and he was enjoying the conflict.

What is the difficulty in learning Sichuan dialect?

To some extent, Trump is completely predictable in the eyes of Margaret.

In almost all cases, Trump speaks faster when he is unscripted, about 220 words per minute, compared with 110-150 words per minute in the United States. At the same time, most people slow down, hesitate or feel uncomfortable when they are about to tell lies, but instead, Trump starts to speed up and transition with "function words" such as "first" and "people are talking". Then gradually slow down to normal speed.

Frischling calls this speed change Trump's "salesman model".

Trump's speech slows down to 111 words per minute when reading in front of a teleprompter, and Trump sounds monotonous and numb to such prepared speeches, playing at random on key points as if he had seen the manuscript for the first time. Trump has said before that reading is illegal for people running for president.

Frischling concluded: "you can think of this process as driving, the average person is driving normally, Trump is speeding." However, he believes that the result does not explain the true or false elements of Trump's words, and in this regard, he will separately ask someone to cross-check the data after checking the facts.

Apart from the differences in language, what about physically? Here we can answer the above answer.

When Trump is really angry, his words become concise and his waving arms stand still. "when he stops making gestures, it means it's time to be careful. No matter what happens, please remain vigilant."

According to the change in Trump's stress level depicted by Margret, the curve suddenly soared when he confronted a reporter from the Washington Post at COVID-19 's briefing. But based on past experience, Trump has no fluctuations in his heart when faced with attacks like "fake news". "the voice, the speed of speech, the way of speaking... everything is the same as when you say 'what's the weather like'," Frischling said.

At the post-election press conference, when Trump was on stage, the curve was also at a very high level, but when he dramatically stepped off the podium and folded his arms, the stress curve fell to the average. it's a sign that he's enjoying the conflict.

After learning 11 million words, Margret knows himself better than Trump

So far, Margaret has learned more than 11 million words of Trump's speeches, tweets, books, rallies, videos, radio and television clips, dating back to 1976, when Trump first published an open letter in the New York Times.

Today, Margaret understands Trump's speech patterns better than many Americans, and even better understands Trump, his words, his subconscious expressions, his tendencies and habits. It is not too much to say that Margret is the person who knows Trump best in the world.

And she also has great patience and learning ability, she will not cheer, ridicule or change the stage because of Trump's speech, she will just quietly analyze Trump's every word and the way he speaks, and then use the algorithm to collect information from a database of more than 40 years of language data in an attempt to interpret his "unstable heart."

One of the important findings of Margaret is that Trump can quickly tell some obviously ridiculous lies because he doesn't care.

"most people don't feel more comfortable when they don't tell the truth," Frischling explained. "but Trump is just the opposite." Margaret can assess Trump's stress level, whether he is calm, really angry, or just acting, and can even simulate Trump's public speech.

To draw these conclusions, Margret tracks Trump's pause, gesture, speed, the types of adjectives he uses, whether he is using his common words, his tone, and so on.

"every word he says makes Margaret smarter and allows her to recognize more subtle differences." Frischling said.

MIT student: building a Trump model? It's too easy!

Trump's analysis is not recent.

Back in 2016, Bradley Hayes, a graduate student at the Massachusetts Institute of Technology, created a Twitter robot @ DeepDrumpf that can mimic Trump's remarks on social media in confusing language, such as "We have to make America. They can't. Because I have to pay for this country."

Hayes had to block a tweet after the account mentioned that violent terrorists would be released to threaten former President Obama. But he said that if there were more time and better data, the robot would become more powerful.

Hayes admitted in an interview that Trump's language is simpler than others, so Trump's model is the easiest to build in modeling. If you can communicate with everyone who has talked to Trump in one day, you are likely to build a more accurate model to better predict what he is going to talk about and where he stands.

Hayes said they only asked AI to study Trump's speech for a few hours, then randomly selected letters as the beginning of the sentence and built it letter by letter. For example, if AI chooses the letter "M" first, it may immediately follow the letter "A" and then "K" until the whole word and sentence articulates Trump's favorite slogan, "make America Great again" (Make America Great Again).

Trump unexpectedly asked AI to join hands with other experts in many fields.

AI technology has obviously improved the means and standards of research.

Margret is the source of choice for journalists, academics and politicians, and Amazon's millions of voice-activated Alexa devices also use Margaret to get Trump's schedule and the latest tweets. Weeks of nightly briefings provided the most typical material for analysis during the novel coronavirus crisis, with unscrupulous claims of success and complete disregard for questions about the government's misresponse.

So far, in order to study Trump's unorthodox speech style, AI has established partnerships with linguists, cognitive experts, theoretical psychologists and political scientists.

Computer scientists believe that artificial intelligence can detect large amounts of data that are too complex for the human brain to find patterns. Intelligence experts suspect that foreign espionage services are combining AI analysis with personality traits and traditional individual-based research methods to analyze Trump and other world leaders.

"at first, Trump confused everyone," said John Sipher, a former CIA agent who has been deployed in Russia and other countries for decades. "now people are starting to understand what he's thinking."

Aleksander Madry, a professor at the deployable Machine Learning Center at the Massachusetts Institute of Technology, warns that the technology is not foolproof and that the patterns discovered through machine learning are "not usually what we think."

Some programmers have used robots to fake Trump speeches, and these "fakes" have the ability to disrupt American politics. This application is a red flag.

Bradley Hayes says almost anyone with a large data set can use the AI robot to draw meaningful conclusions-or distort them.

If Trump's language is usually predictable, his decisions sometimes seem completely random.

Frischling, who has studied some members of Congress for his private clients, believes there will be signs before they decide to change policy, such as making more public comments on certain topics or changing the way they speak. But these relationships are rarely seen in Trump.

"for a policy proposal, he can first say,'I think this is too bad, 'and then sign an executive order to approve the policy within 10 hours," Frischling said. Margaret is still trying to understand this.

Perhaps it can be said that Trump's bewilderment is also protecting the United States.

Related reports:

Https://www.latimes.com/politics/story/2020-05-07/cant-decipher-trump-speak-meet-margaret-the-computer-bot

Https://qz.com/631497/mit-built-a-donald-trump-ai-twitter-bot-that-sounds-scarily-like-him/

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.