In the months prior to election day, voters at polling stations were scrutinised, with data from telephone surveys and past election developments broken down to determine the likely results. Tuesday evening the pollinators were under the microscope according to the results based on the prognosis.
However, there are other ways to judge feelings. Some technology experts say that artificial intelligence, increasingly used by companies to gauge customer sentiment, can promise a better understanding of the electorate.
Subscribe to the newsletter
The morning download provides daily news and information about the CIO Journal team’s business technologies.
I wouldn’t fire sociologists, but I would ask them to focus more on machine learning, data mining and AI in their work to get better predictions, said Oren Etzioni, CEO of the Allen Institute for AI, a non-profit research center in Seattle.
The extent of this year’s voting error is still unknown, as the votes are still being counted. But the polls as a whole predicted clear democratic benefits, not an increase.
No person or algorithm can accurately predict human behavior at any given time, said Heidi Messer, president of Collective[i], a New York-based company that delivers artificial intelligence and predictive technologies to sales teams. However, the problem with traditional surveys is that the terms used by sociologists are based on classifications and historical averages.
The research must find a source of data that captures the actual behaviour of technology companies, for example
Yeah, yeah, yeah, yeah, yeah, yeah. Amazon’s algorithm doesn’t matter if I’m a human or a dog, it only knows that if I buy a leash, I’ll probably buy dry food, she says.
She went on to say: Data that represent behavior are much more difficult to collect, but are infinitely useful for dynamically determining the patterns and correlations that feed probabilistic predictions.
Several AI companies have used their models to predict elections.
Expert.ai, an Italian software company specialising in natural language processing, has applied its technology to millions of social positions around candidates. Their ATI system, partly set up for issues related to previous elections, analysed factors such as tone and emotion and projected how this might affect voting.
According to the Expert.ai system, democratic presidential candidate Joe Biden will receive 50.2 percent of the vote, while Republican president Donald Trump will receive 47.3 percent, or 2.9 percentage points. Mr. Biden received 50.5% of the votes on Friday afternoon, while Mr. Trump received 47.8% and 2.7% of the votes. (Dow Jones & Co., who publishes the Wall Street Journal, is a client of Expert.ai).
Unanimous.ai conducted a live poll in September, in which 50 participants were asked to predict who would win the presidential race in the 11 battlefield states and to what extent.
Another company, San Luis Obispo, California Unanimous.ai, sells spin-off and intelligence software. It uses AI models to aggregate the predictions and decisions of groups of people, such as investors predicting commodity prices.
Executive Director Louis Rosenberg said his team used the program to conduct a live survey among 50 people in the United States in September and asked them to predict who they think will win the presidential race in 11 states and with what benefit.
Friday morning’s opinion poll in September predicted the winner of the presidential election in eight national battlefield matches correctly, according to the Associated Press. He’s on schedule to correctly predict the other three. The research also generally predicted which breed would be the most difficult.
A real-time survey conducted by Unanimous.ai allows users to see in real time what other participants are choosing, while machine learning algorithms use factors such as changes in responses to assess who relies most on their predictions. The most confident responses have the greatest influence on the final overall response.
Our data was 50 randomly selected voters who then used the swarm to improve their intelligence; it wasn’t 100,000 polls, Dr. Rosenberg said. We constantly find that we are able to surpass really big investigations.
But not all of ATI’s polls were that accurate.
Polly, developed by AI for Advanced Symbolics Inc.’s consumer market research, has successfully predicted some 20 elections and referenda, including the UK’s withdrawal from the European Union.
The ASI predicted that Mr. Biden would win. But it allowed him to take Florida and win the election from 372 to 166. The AP has called Florida for trump, and the AP election results are much closer, but Biden is in charge.
We were wrong, said Kenton White, co-founder of the ASI and the company’s chief scientist.
To predict elections, ASI uses machine learning to monitor and analyse election-related topics discussed on major social networks, such as
The company collects public information from websites to create a sample of users representing the U.S. population.
The research also collects data on past elections and examines how people’s views in the past coincided with previous votes to make a prognosis of how people will vote in the next elections. In the last election, for example, the people who talked about getting back to work in America were Republicans, and those who talked about global warming were Democrats.
Erin Kelly, co-founder and CEO of the ASI, says the use of artificial intelligence for election forecasting is a relatively new practice. And one of AI’s strengths is that models can learn and improve over time.
For Polly, this year’s election was an instructive experience.
Copyright ©2020 Dow Jones & Company, Inc. All rights reserved. 87990cbe856818d5eddac44c7b1cdeb8