Ottawa company banking on real-time data over history to predict US election
An Ottawa scientist will go up against a respected American scholar in Washington, D.C., on Thursday to debate the best, most accurate methods to forecast the U.S. presidential election.
But the debate between University of Ottawa computer science professor Kenton White and American University history professor Allan J. Lichtman isn't just about pitting man versus machine — it's about whether real-time data can predict an election better than history.
Lichtman says he plans to remind the audience at American University, where the debate will take place, that he has accurately predicted elections since 1984 — without the help of computer analytics.
"My forecasting is based upon an historical analysis of Americian presidential elections from 1860 to 1980," said Lichtman in an interview from Washington. "This system has correctly predicted the outcomes of all eight American presidential elections from 1984 to 2012."
Lichtman said according to his criteria and analysis, the Republican candidate should win the White House, but he said this time, Donald Trump's candidacy is forcing him to hedge his bets.
Trump a 'history shattering' candidate
"In Donald Trump we have a history shattering, dangerous candidate," said Lichtman. "That's what makes this election so puzzling and so fascinating. I have a hedged prediction for the first time in 30 years. I'm not a hedger, I've never done this before."
But White, chief scientist at Advanced Symbolics, an Ottawa-based artificial intelligence company, isn't hedging at all. In fact his Ottawa team is confident the company predictions — based on artificial intelligence — will prove solid on election day.
"We're taking advantage of all this amazing technological innovation," said White "It's not like this crazy Canadian researcher is challenging a 100-year-old American tradition, rather we're building on that tradition with decades of computer science."
White said in a world with a 24-hour-a-day news cycle, Twitter and a very fast feedback loop, predictors need technology in order to keep up.
Advanced Symbolics' technology follows the same 200,000 U.S. voters over a year, by monitoring their opinions, news and ideas shared on social media, blogs and bulletin boards, according to company CEO, Erin Kelly.
Looking at now, not yesterday
"The artificial intelligence really looks at many things and it makes sure the people it's following is representative of the population," said Kelly.
She said Lichtman's prediction is "based on historical outcomes and projecting them to future elections which they've admitted this time will be difficult."
Kelly said her company's analytics are showing a Clinton win, but a closer election than is currently being predicted by most of the polls south of the border.
"The spread is only about three percentage points and not the six percentage point spread we're seeing from other pollsters," said Kelly.
Lichtman admits he doesn't know much about the Canadian scientist he'll be on stage with on Thursday, but he said he's willing to debate anyone.
"I have a lot of respect for artificial intelligence. I'm not saying it can't be promising for predicting presidential elections, I'm just saying until we have a significant track record, we don't know," said Lichtman.
Professor White isn't intimidated. He says is technology accurately predicted the UK Brexit vote and last year's federal election.
"I get feedback from community science community, but I'm looking forward to getting feedback from non-computer scientists and American University has an amazing history in public service."
