Smarter as a group: How swarm intelligence picked Derby winners

A platform using input from 20 people correctly predicted the top four horses at the Kentucky Derby, a sign of progress for the advanced group feedback technology. 

Garry Jones/AP
Mario Guitierrez celebrates after riding Nyquist to victory during the 142nd running of the Kentucky Derby horse race at Churchill Downs on Saturday. A panel of UNU swarm intelligence users were able to correctly predict Nyquist's win, as well as the next three runners-up.

Taking cues from the animal kingdom, a company behind the UNU swarm intelligence platform was able to use the collective input from a panel of 20 people to successfully predict the superfecta, or top four finishers, in the 2016 Kentucky Derby.

In comparison with the Derby's lineup of expert picks, whose superfecta guesses were all incorrect, UNU participants correctly found not only the top four horses in the race, but their correct finish order, by using the swarm platform. The human-backed group also outperformed the fully artificial Bing Predicts: Bing's AI guessed the winner, but picked incorrectly for second, third, and fourth place.

"Nature show [sic] us that social creatures, when working together as a unified system, can outperform the vast majority of individual members when solving problems and making decisions," Unanimous A.I., the developer of UNU, notes on its website. 

Unanimous A.I. chief executive officer and UNU inventor Louis Rosenberg.suggests that platforms like UNU could also be applied to fields from medicine to politics. For example, he notes doctors could pool knowledge to decide on a diagnosis or the best course a procedure could take, while swarm AI could give a better read on voters' choices or policy preferences than current polling methods. 

Scientific World Journal notes that there is already research into a variety of applications of swarm intelligence ranging from producing better heating systems and elevator controls to forecasting the spread of viruses.   

"Forcing polarized groups into a swarm allows them to find the answer that most people are satisfied with. Our vision is to enable the power of group intelligence for everybody," Mr. Rosenberg said. The concept of swarm intelligence could also be applied to fully-AI systems, leaving some decision-making and problem-solving up to learning machines in the future.

Many animals evolved with some facet of swarm intelligence; Unanimous notes ants' chemical connections, birds' and bees' gesture detection, and more. Humans, on the other hand, have not developed a biological method for groups to behave as a whole. But with social swarming technology, humans have the potential to think together and "achieve the same types of intelligence amplification that other species have attained," Unanimous says. 

The company's UNU system allows users to log onto an online question-answering and decision-making forum which aggregates a "swarm" of users' knowledge and opinions. UNU queries are posed to a group with multiple answers available to choose onscreen, where individual users use magnet-shaped cursors to drag a digital puck in the direction of what they believe to be the right answer. Each swarm can only end up with one response, which can lead to push-and-pull within the group until a collaborative decision is made.

The UNU Derby panel did consist of people "knowledgeable" about horse racing, yet was still able to accomplish something no other experts could. And this was far from the first time UNU has outperformed authorities: the technology has previously succeeded in predicting Super Bowls, NCAA bowl games, Academy Awards, and even presidential primary winners better than panels or polls.

UNU's Derby superfecta win, at 540-to-1 odds, netted Unanimous A.I. a whopping $10,822 profit from a $20 bet. Hope Reese, a TechRepublic writer who challenged Unanimous to predict the Derby top four, bet $1 and left Churchill Downs with $542.10.

A post-Derby poll of the 20 UNU participants found that if they had aggregated their decisions by voting, instead of using swam intelligence, they would have only guessed one horse correctly: the 3-to-1 morning line favorite and eventual winner, Nyquist. Not one of them predicted the final lineup on their own.

Systems like UNU have yet to be widely tested or implemented, and still have doubters. University of Louisville Cybersecurity Lab director Roman Yampolskiy told TechRepublic that "if swarm AI worked as advertised, it would be easy to monetize it via prediction markets," and that "as we don't see such betting happening it is a strong testament to the system's questionable performance." 

But UNU's strong Derby performance and history of on-point prognoses are encouraging others, including Rosenberg.

"When I saw the horses cross the line, I knew I was witnessing a milestone in the predictive abilities of AI, as well as a harbinger to future changes in how the world views sports gambling," Rosenberg told Newsweek. A post on Unanimous' website even hinted at the possibility of an "elevated 'super-mind'" that could result from utilizing hundreds or thousands of people, or more, in a swarm.

of stories this month > Get unlimited stories
You've read  of  free articles. Subscribe to continue.

Unlimited digital access $11/month.

Get unlimited Monitor journalism.