Meet Polly, the AI pollster that wants to predict elections using social media

Thanks to technology, traditional pollsters don't have it easy when it comes to finding out which way an election might go, according to documentary maker Peter Gombos.

As people lead increasingly mobile lives — and shed the tethered landline home phone — getting in touch with the electorate can be tricky. 

Pollsters can reach some through internet surveys, and others do pick up their cell phones. But at least one pollster named Polly thinks there's an easier way.

Polly is no traditional pollster. Polly is an artificial intelligence system built by Ottawa-based startup Advanced Symbolics Inc. (ASI) that scrapes public data from social media networks to predict election outcomes.

"To find public opinion, you have to go where the new public sphere is and that's defined by technology," Gombos told Day 6 host Brent Bambury. 

Gombos is the director of Margin of Error, a new documentary premiering on TVO this month. The film follows Polly's predictions for the 2019 Canadian federal election.

In the days before Canadians cast their ballot, Polly predicted a Liberal Party minority government with 77 per cent confidence — and it wasn't the system's only correct guess.

It also accurately predicted that Britons would vote to leave the European Union in June 2016, as well as Donald Trump's rise in popularity and the popular vote of the 2016 U.S. election. Both predictions were made based on sentiments publicly shared by users of social media platforms.

Now Polly is once again turning its attention south of the border ahead of next month's vote.

"You have to look on social media," Gombos added. "That's where I think the predictors have to go. Whether that's going to be a problem or not is a bit of an open debate."

Contentious approach

The idea of scraping the internet for information people posted online is contentious. British consulting firm Cambridge Analytica made headlines — and faced public scorn — in 2018 after consultant Christopher Wylie revealed that they harvested information from 50 million Facebook accounts.

The firm used the data to support social media ad campaigns for Donald Trump's 2016 election campaign.

"You can't work on this kind of doc without thinking of Cambridge Analytica and the way it was piggybacking on Facebook to microtarget users, and push skewed information towards people based on psychological profiles," Gombos said.

But Gombos explains ASI's approach with Polly is different. Rather than micro target users directly, ASI creates an anonymous, representative picture of what's on voters' minds based on the information that they've shared publicly, he says. On their website, ASI states that they do not access any private communications.

"In the era of Cambridge Analytica and all that, people think it's some sort of voodoo science," says ASI CEO Erin Kelly in the documentary. "I want people to understand, this is just math and it's science, like anything else, and it's probabilistic. It's statistics just like it's always been, except it's being done by Polly instead of by phone operators."

'Maybe the traditional pollsters are afraid'

Traditional pollsters have long relied on surveying random individuals by phone to capture data on voting intentions and hot-button issues. In recent years, online surveys have become more common.

Polly's approach is to aggregate voter opinion data from content posted by users on social media platforms like Twitter. Polly aims to create a representative sample — factoring in demographics — and, according to ASI, can break down the data on a riding-by-riding basis.

From there, the system parses the data and determines the patterns within it to inform its guesses.

But some experts have pushed back on the idea that artificial intelligence can replace traditional methods for testing public opinion.

For one thing, many social media users who discuss politics online tend to skew hyper-partisan and are actively engaged in politics. Active social media users, particularly on Twitter, also make up only a fraction of the total population.

In a scene from the film, long-time pollster and CEO of polling firm Ipsos Public Affairs Darrell Bricker says that while the approach is interesting, he has his hesitations.

"Would I bet the farm tomorrow on being able to do an election in which I predict every single riding correctly based on social media? No."

But Polly's creator, ASI Chief Data Scientist Kenton White, sees it differently.

"I think that maybe the traditional pollsters are afraid of this, because this is the future," he said in Margin of Error.

Polly's future

There have been calls to regulate — and protect — how the user-submitted data on social media platforms is used by companies and other groups. That has some wondering about the future of AI that relies on such data.

But Gombos says that ASI isn't worried. "I think Polly's creators believe at this point that there is such a demand for people to be able to share information publicly — freely — on social media, that there isn't, in the short term, a likely restriction on access to data."

In the meantime, Gombos says that he's keeping an eye on Polly's predictions for next month's election.

As of Oct. 15, Polly is predicting that Democratic presidential nominee Joe Biden will sweep the U.S. election next month, in both the electoral college and popular vote.

According to ASI, the prediction is based on data from a single sample of more than 287,000 Americans. By using the same sample in each prediction, ASI says that it can "dive deeper" into each voters' history and develop greater confidence.

"I'm wondering if it's going to be a Biden landslide, as she claims currently."

Written by Jason Vermes. Produced by Annie Bender.