The idea behind Swarm Intelligence is simple: people are smarter when they think together than on their own. But, given the negative stereotype of the “hive mind” and the celebration of the “expert,” many people are skeptical of Swarm Intelligence. As a result, reporters love to challenge the popular Swarm Intelligence named UNU to predict major events like the Oscars and the Super Bowl and the Kentucky Derby.
This is why a reporter from CNET and hockey enthusiast challenged UNU to predict the NHL playoffs. At the time, four teams were left in the competition: the Penguins, Lightning, Sharks, and Blues, with no expert consensus about which teams would advance. A swarm of 29 hockey fans was formed to make the predictions, including the CNET reporter who participated in the Swarm and wrote about the session in his article, “AI startup taps Human Swarm Intelligence to predict winners.” And, because we wanted to see how the Swarm would do against the average fan working individually, all 29 fans were asked to make the same predictions on their own.
The Swarm was asked to pick not just the winner of the Stanley Cup, but also how many games each series would go. There were 21 possible games to predict: 7 games between the Sharks and the Blues, 7 games between the Penguins and Lightning, and 7 games between the two eventual winners of the semi finals. The quick video below shows the Swarm recording its pick for the Western Conference Finals, and a full set of those picks was recorded on our BLOG and is summarized below.
UNU said….
WESTERN CONFERENCE FINALS:
- Sharks would win 4 games against Blues.
- Blues would win 2 games against Sharks.
- There would be no 7th game.
EASTERN CONFERENCE FINALS:
- Penguins would win 4 games against Lightning.
- Lightning would win 3 games against Penguins.
STANLEY CUP FINALS:
- Sharks would play Penguins.
- Penguins would win 4 games against Sharks.
- Sharks would win 3 games against Penguins.
So, how did UNU? And, just as importantly, how did the individual users do in predicting the same set of games? After all, the true measure of Swarm Intelligence is not its ability to predict the 541 to 1 Superfecta, but rather its ability to outperform the average person.
UNU made 20 correct predictions out of the 21 games… a stunning 95% accuracy!
UNU’s single misstep is that he expected the Sharks to win one more game against the Penguins than they did before ultimately losing the Stanley Cup. But UNU correctly forecast that the Sharks would advance in six games over the Blues, the Penguins would beat the Lightning in seven, and, as you can see below, that the Penguins would win the cup.
Even more compelling is the fact that UNU did better than every single participant in the swarm. Remember, all 29 people were asked to make their own individual predictions, and none of them were 95% accurate. In fact, the best that any individual did was 80% correct, with most people doing far worse than that. In addition, the most popular individual predictions ( otherwise known as the Wisdom of the Crowd ), picked the Sharks to win it all, as did the CNET reporter.
So, what does this experiment demonstrate about the efficacy of Swarm Intelligence? Let’s recap:
- UNU was nearly perfect, predicting the winner of 20 out of 21 possible games
- UNU outperformed every single individual who comprised the Swarm
- UNU’s predictions did NOT reflect the “Wisdom of the Crowd,” but rather the Wisdom of the Crowd working together and correctly picked the Penguins to win the Cup, where a vote would have bet on the Sharks
What’s next for Swarm Intelligence? As you may have read, UNU picked the Warriors to win the NBA finals, and has lead wildly popular Q&A sessions on Reddit’s R/IAMA and R/Futurology subreddit. Our Beta community is using UNU around the clock, and we’d love to have you join the Swarm.
Come check it out here: UNU
Pingback: Crowdsourced AI exactly predicted the most improbable Super Bowl of all time - Kingdom Done