What does the world’s most powerful artificial intelligence say about the U.S. Presidential Candidates?
I spent the last week engaging with the world’s smartest artificial intelligence brainiac about the U.S. Presidential Election. I pressed IBM’s Watson to light up its neural networks, apply cognitive systems, crunch semantic learning, and emote sentimentalism to help me understand what the candidates are saying and meaning. The results were astonishing.
IBM’s Watson platform is a cognitive system enabling a new partnership between people and computers. It is most famous for beating the Jeopardy champions for a $1 million prize. The Watson platform has introduced groundbreaking advancements in healthcare, finance, biotechnology, and other sciences, and is entering the fields of education and finance. With that breadth of intelligence, a natural question to ask Watson would be, “What do you think of the U.S. presidential race?”
Mr. Watson. Come Here. I Want to See You
As a cloud-based computer platform, Watson should be an apolitical judge. Watson uses natural language processing to understand meaning in text messages. Coyly, one may argue there is nothing natural about this presidential election. Nevertheless, Watson is the best artificial mind we have to interpret what the candidates are saying and what they actually mean.
When engaging Watson, I didn’t explain the context of everything related to the election, not Donald’s business successes/failures, nor Hillary’s emails. Why bother? Watson is in the “cloud.” It lives in the Internet even more than I do.
To prime Watson for our conversation, I provided transcripts from acceptance speeches at each party’s national conventions and ran them through IBM’s Alchemy Language – Watson’s brain for text analysis using natural language processing. Watson’s Alchemy Language engine analyzes text to help understand concepts, things, keywords, sentiment, and emotions.
Determining context from a phrase is not an easy task neither for humans with political candidates, nor for Watson trying to dissect and understand a political speech. For example, in Hillary Clinton’s speech she said, “I will be a President for Democrats, Republicans, and Independents.” Is this neutral, positive, or negative? For Republicans and Independents, maybe negative. Unless you’re a republican who loathes the idea of a President Trump. The beauty is in the eye of the beholder.
When Donald Trump said, “The problems we face now – poverty and violence at home, war and destruction abroad – will last only as long as we continue relying on the same politicians who created them,” his words contained problems, poverty, violence, war, destruction – four highly negative words. However, anyone reading the line may interpret it as a call for change, hope, and a better future.
To best illustrate the problem of attaching sentiment and emotion is the line from Donald Trump’s speech when discussing the horrific Orlando nightclub massacre, “49 wonderful Americans were savagely murdered by an Islamic terrorist.” The word wonderful is a positive emotion attached to Americans. Savagely, murdered, and terrorist are all negative words also attached to the Americans, Islam, and terrorists. Donald Trump was not being negative towards Americans, but the semantics of the words associated with their objects was negative. These are the complex problems cognitive systems must understand to be most beneficial to humans.
Donald Trump’s Acceptance Speech Word Cloud
That’s Good, But How Do You “Feel?”
When Watson extracted and examined the top 50 things (entities) that each candidate mentioned in their speeches, it found that Donald Trump was positive only 22% of the time, compared to 38% for Hillary Clinton. Both candidates were negative about 35% of the time. Donald Trump’s primary sentiment towards things was neutral, while Hillary Clinton’s was positive.
When Watson extracted and examined the top 50 keyword concepts from each candidate’s speech, Watson found that Donald Trump was positive only 22% of the time versus 58% for Hillary Clinton. Inversely, Trump was negative 66% of the time compared to 30% for Hillary.
For targeted sentiments, Donald Trump was negative 65% of the time, while Hillary Clinton was positive 65% of the time. Donald was positive only 25% of the time, while Hillary was negative only 30% of the time.
Eliminate All Other Factors, and the One Which Remains Must Be the Truth
There were many areas where Watson hit the bullseye. Below are some top examples:
- Each candidate’s number one topic was America
- Each candidate’s number two topic was their opponent, and they were negative about their opponent.
- Donald Trump speech spoke often about countries in the Middle East and was negative about these countries.
- Both candidates spoke positively about their family members,
- Hillary Clinton was positive about “young people,” “small businesses,” “good paying jobs,” and “better lives.”
- Donald Trump was positive about the “American People,” “United States,” “new wealth,” and a “truly great mother.”
The Empires of the Future are the Empires of the Mind
What were the core topics covered in each speech? According to Watson, Trumps speech mostly covered societal unrest and war, and children. Watson wasn’t confident that Trump’s speech was about government. Perhaps Watson has an ironic point. Clinton’s speech, by contrast was about foreign policy, war and unrest, and education. Watson was confident about all three topics for Hillary’s speech.
Hillary Clinton’s Acceptance Speech Word Cloud
We’re All Quite Mad Here! You’ll Fit Right In
When Watson examined the top 5 concepts relating to each speech, here’s what it found:
Donald Trump’s top 5 concepts in his Acceptance Speech:
- Hillary Clinton
- Bill Clinton
- Barack Obama
- Illegal Immigration
- United States
Hillary’s Clinton’s Top 5 concepts in her Acceptance Speech:
- Joe Biden
- United States
- Barack Obama
- President of the United States
- Donald Trump
Wow! I have watched, listened, and read Hillary’s speech several times now, and I cannot understand how the top concept Watson identified was Joe Biden. Above the grand old USA. That’s a big miss. Also surprising is that Bill Clinton was #2 on Donald Trump’s list.
Come On, Group Hug! You Too, Anger.
Watson examines the same 5 emotions as the Pixar animated film, Inside Out: Fear, Joy, Disgust, Anger, and Sadness. While humans have many more emotions, both Pixar and Watson limit their evaluation to these five. Watson found the emotions presented in the acceptance speeches both conveyed Anger and Fear. The other three emotions (Sadness, Disgust, and Joy) didn’t rise high enough to be considered significant. Sadly, Watson disgustingly didn’t find either speech to be joyful.
No Great Mind has Ever Existed Without a Touch of Madness
We are at the late dawn of cognitive systems, and by using Watson to interpret the presidential candidates’ acceptance speeches we are giving it an arduous task that exposes how well it works as well as areas requiring advancements in interpretation both for the computer, for the analyst, and the reader. For the list below, I examined Watson’s interpretation, then re-read the speeches to try to understand what Watson was thinking and if it was correct or erroneous.
Watson’s biggest misses were:
- Hillary Clinton’s sentiment about America was negative.
- Hillary Clinton was negative about Bernie Sanders.
- Hillary Clinton was negative about the President and vice president.
- Donald Trump was neutral about President Obama.
- Donald Trump was negative about Americans, country, and America.
- Donald Trump was negative about law enforcement officials.
- Donald Trump was neutral about African American youth.
- Hillary Clinton was positive about Trump picture frames.
- Hillary Clinton was negative about police officers.
- Hillary Clinton’s top concept was Joe Biden.
- Donald Trump was neutral about Fred Trump
- Hillary Clinton was neutral about Bill Clinton.
- Hillary Clinton was negative about First Lady Michelle Obama.
- Hillary spoke negatively about Joe Biden
- Hillary Clinton was negative about “real change.”
- Donald Trump was positive about “Special interests.”
- Donald Trump’s 2nd most relevant concept was Bill Clinton.
When Gunpoint is a Country
Watson assigns entities to a category, and some of these categories were incorrect. Watson assigned the White House to be an Organization in Donald Trump’s speech, but a Facility in Hillary Clinton’s speech. In Donald Trump’s Speech, Watson found America to be a continent, and Gunpoint to be a country. In Hillary Clinton’s Speech, Watson misidentified New York state as a city even though the sentence where it was mentioned said, “the great state of New York.” Watson also identified the Secretary of State as a “Field Terminology” rather than a JobTitle. (It understood President to be a JobTitle.)
Note: FieldTerminology appears to be the category when Watson knows an entity is an unnamed thing.
Watson, the Case is a Foot
Understanding meaning isn’t as easy as we humans give it credit for. Computer scientists will tell you that the most valid results come from providing enormous data sets for a cognitive system to learn from (big data analytics), and then allow it to identify patterns that develop over time. Giving a single speech is likely to provide atypical results. Since this is an atypical U.S. presidential election, it seemed appropriate to ask Watson to weigh in.
Cognitive systems are an emerging technology. I’ve worked with semantic matching engines for over a decade and know much about their horrific abilities to misinterpret and misunderstand. It’s easy to poke fun at them and find their deficiencies relative to human understanding. However, it’s even more amazing to see what they can do and what they do get right. IBM is aligning their business to the rise of cognitive systems, and it’s making a difference. IBM Watson Healthcare is helping improve healthcare performance at a local and global level. IBM Watson is beginning to focus on helping improve education, and is making inroads into banking.
Elementary, Dear Watson
After all Watson’s analysis of the candidate’s acceptance speeches, where does it leave me? Where does it leave you? Us?
Watson conclusively found Donald Trump’s speech to be significantly more negative than Hillary Clinton’s. It did an amazing job of correctly extracting and interpreting much of the speeches text with the whimsical, endearing, and head-scratching exceptions noted above. Overall, Watson leads us to the two questions we really want the answer to: “Who will win?” and “What will that mean?”
The tables below are directly taken from Watson’s search results. Notes are indicated to help address any confusion and anomalies.
- You can test Watson’s Alchemy Engine out for free at https://alchemy-language-demo.mybluemix.net. Here you can enter a body of text or a URL, and Watson will perform the same analysis used in this report.
- Speech transcripts were obtained from Politico.com, and the URLs for the speeches on Politico.comPolitico.com were used, not a copy of the body of text from the site.