Election Tech 2016 Innovation

TechRepublic's 'swarm AI' predicts tight election, gives edge to Clinton

TechRepublic and Unanimous A.I. conducted a final real-time 'swarm' to determine which candidate will be stronger on economic and technology issues and who will win the US presidency. See the results.

Image: Getty Images


On Tuesday, November 8, voters will cast their ballots for US President, marking the end of one of the most bizarre and unprecedented campaign seasons in modern history. While many are looking to polls or pundits for predictions about the results, there is a third option: Artificial intelligence.

To gauge opinions about the candidates, TechRepublic has teamed up with Unanimous A.I. to conduct three "swarm AI" sessions--in which a group of voters meet up online to make a real-time prediction. The swarm works via an online platform called UNU, which harnesses artificial intelligence to help a group come to a decision together. They gather on individual computers to "pull" a virtual magnet toward an answer--for instance, who will win the male lead at the Oscars?

The swarm AI method has proven remarkably accurate, frequently outperforming experts and polls. When TechRepublic asked a swarm which horses would place first, second, third, and fourth place at the 2016 Kentucky Derby, the swarm nailed it--beating 540-1 odds.

The swarms, which have focused on how each candidate will fare on issues related to the economy and technology, were conducted pre-conventions (full results here), pre-debates, and now, for the final swarm: Pre-election.

For each swarm, TechRepublic asked a similar set of questions, such as "How will unemployment change in a Clinton presidency?" The swarm could choose between "increase a lot," "increase a little," "decrease a lot," "decrease a little," and "no change."

Election Tech

Visualizing the Russian cyberattack

Cybersecurity experts explain how the data breach happened, and Twitter chatter reveals what competing social media factions are saying about the election hack.

The swarm was faced with a series of questions and could answer in the following ways: "Clinton, a lot," "Clinton, a little," "Trump, a lot," "Trump, a little," or "Even." The format allows them to not only address a preference, but also the strength of the preference, which has proven valuable in previous swarms.

Here are our questions, and answers from the swarm (with the answers from our pre-debate swarm in parentheses, when applicable):

  • Who will be more effective at creating jobs? Clinton, by a lot (Clinton, by a lot)
  • Who will be more effective at managing the economy? Clinton, by a lot
  • Who will be more effective at reducing unemployment? Clinton, by a lot; (Clinton, by a lot)
  • Who will be stronger about helping small businesses? Clinton, by a lot; (Even)
  • Who will be more innovative in using technology in government? Clinton, by a lot; (Clinton, by a lot)
  • Who will be stronger about national cybersecurity? Clinton, by a lot
  • Who will work better with Silicon Valley and tech companies? Clinton, by a lot; (Clinton, by a lot)
  • Who will win the election and by how much? Clinton by a little; (Clinton, by a little)

The swarm also said that Clinton benefited "a lot" more from the debates than Trump. The answer could have been foreseen in this pre-debate question: "Who will win the debate?" The swarm responded, "Clinton, by a lot."

SEE: TechRepublic will use 'swarm AI' to make predictions about the 2016 US presidential election (TechRepublic)

The swarm consisted of 45 people of voting age. Thirty percent identified as independent or other, 46% as Democrats, and 24% as Republicans. This swarm, compared to the previous, was slightly more Republican (our last was 18%) and slightly less Independent (last was 35%). According to Pew Research, 23% of voters identify as Republican. They also rated themselves on a "Unanimous' Conservatism rating scale" and were slightly left-of-center at 4.22 (with 4 being neutral). The swarm was, however, male-leaning--63% male, 37% female. It also tilted younger. Twenty-two percent were in the 18-25 range; 52% were 26-35; 20% were 36-45; 4% were 46-55 and 2% of the swarm members were over the age 55.

Our major takeaways? As in previous swarms, the group was supportive of Clinton, predicting higher performance on the economy and innovation under a Clinton presidency. Our first swarm did not think Clinton would have an impact on unemployment, but our pre-debate and pre-election swarm both say that she will reduce unemployment by "a lot." Our first swarm did not see a difference between Clinton and Trump in cybersecurity or innovation, but the last two swarms predict stronger performance on cybersecurity by "a lot" from Clinton, and see Clinton as "a lot" more innovative. Also, our pre-debates swarm predicted that Clinton and Trump would perform evenly when it comes to supporting small businesses; our current swarm believes that Clinton will be stronger by "a lot."

Image: Unanimous A.I.

But although the swarm predicted strong performance by Clinton before the debates and judge her as more likely to perform well in this pre-election swarm, the group expects it to be a close finish (see above). The swarm was conservative in predicting a win by Clinton, saying it she will win by "a little"--the same answer as the pre-debate swarm--whereas (as of Monday, October 31) Nate Silver's 538 blog gave her a 77.8% chance of winning.

Also see

Visit TechRepublic