U.S. election: CEO of AI pollster believes complex ballot issues led to incorrect predictions
Print/DownloadTORONTO —
The CEO of a Canadian artificial intelligence pollster that correctly predicted the 2019 Canadian election and the Brexit vote believes the sheer volume of election issues can be at least partially to blame for its predictions being so wrong this time around.
Polly Pollster, which uses social media data to predict real-world outcomes, had suggested that Democratic nominee Joe Biden had a 98 per cent chance of winning the U.S. election and predicted Biden would win a whopping 372 electoral college votes with 55 per cent of the popular vote.
As of Wednesday afternoon, Biden and U.S. President Donald Trump are in a dead heat to win the electoral college.
Erin Kelly, CEO of Advanced Symbolics Inc., told CTV News Channel on Wednesday that the volume of issues in this election — COVID-19, racism in the U.S., and unemployment among them — led Polly to mix up which issues would resonate most with voters.
“We saw Polly struggling to decide which of these things was most important,” she said. “That’s why we’re seeing pollsters having a more difficult time calling it, because the population is getting more dimensional in how it looks at things.”
Polly also suggested the Democrats would win each of the swing states, including Florida, Pennsylvania, Ohio and Michigan, and that they would make historic flips in both Texas and Georgia.
“Obviously some of those are still up in the air, we don’t know who won them,” Kelly said. “We do know that Polly got Florida wrong and she got Texas wrong,”
When it comes to Florida and Texas, Kelly said that Polly underestimated the complexity of the Latino and Hispanic votes.
“She saw them as one group and we now realize…she should have seen each of those ethnic groups — like the Cubans versus Venezuelans — as separate groups because they really did not behave similarly on Election Day,” she said.
Kelly said Polly will be tweaked before the pollster will be implemented in another election.
“Like any intelligence with Polly, we have to take her back, fine tune with every election to see how she can get better,” she said.
In the meantime, Polly will be used in health care to forecast COVID-19 infections, for market research and to recognize depression in populations to minimize suicide rates.
