On May 7, 2016, a winning $20 Superfecta ticket in the main stakes of the Kentucky Derby paid $11,700. If you had listened to a small predictive startup in San Francisco, you too could have earned a 585,000% return in just two minutes.
This upstart player in prediction theory is Unanimous A.I., a company founded by Dr. Louis Rosenberg in 2014. The company’s secret? The theory of Artificial Swarm Intelligence, or Human Swarming, which has proven new advancements in human behavior and forecasting success.
The company’s platform combines traditional, real-time human predictions with software algorithms that allow tens of thousands of human participants to join groups and think together as a collective swarm. But this is far different than traditional polling or surveying to predict future outcomes.
Swarm intelligence is founded on the idea that many minds and forecasts are better than a single thought or prediction. Animals in nature bolster their collective intelligence by collaborating in flocks, herds, swarms and schools. In 2014, Unanimous’ research team found that humans can engage in online swarms to bolster intelligence of one group. Since then, Unanimous has made headlines. The swarm correctly predicted the Chicago Cubs would beat the Cleveland Indians in the World Series and nine of the 10 playoff teams, four months before the October contest started. It was also challenged to predict the Stanley Cup, and correctly picked the two finalists, the winner, and the number of games played. But the results aren’t limited to sports. Newsweek challenged Unanimous to predict the winners of 2016 Academy Awards. It was right on 76% of predictions -- a figure that topped the majority of Hollywood experts.
Validating the Swarm
Rosenberg, who has several degrees from Stanford University, argues that the performance has validated what many people inherently know about human systems.
He claims that a massive untapped source of knowledge, wisdom and intuition is waiting to be cracked across many different industries. That includes the so-called experts and those who are enthusiasts and have a level of confidence about their knowledge.
“When we build a swarm, we want participants who are experienced. They don’t have to be professionals, but we want participants who are knowledgeable about the topic,” Rosenberg says. “They also know what they don’t know, right? They are realistic in their abilities to assess a topic. And so we look for people who express that they’re enthusiasts. If somebody is so confident that they’re an expert, they might not know what they don’t know.”
That is key because each swarm allows self-selecting individuals to engage based on their enthusiasm and conviction, and not their self-selection as an expert on an issue. “When you take a swarm of regular people and you have them work together as a system, as a swarm intelligence, they consistently outperform people who call themselves experts,” he says.
More interesting, the collective swarm dramatically outperforms individual participants. Even though it correctly picked all four horses in the Kentucky Derby Superfecta, not one member of the swarm picked the horses to come in that order.
In 2016, Rosenberg and Unanimous CIO John Baltaxe coauthored and released a paper with and Niccolo Pescetelli at Oxford University. The paper compared the superior results that Swarm Intelligence had compared to traditional crowdsourcing on knowledge of the NFL. According to the team’s study, the academics challenged “a crowd of 469 football fans and a swarm of 29 football fans” to make predictions of 20 proposition bets during last year’s Super Bowl between the Denver Broncos and Carolina Panthers. Even though the crowd was 16 times larger than the swarm, they were only correct 47% of the time. However, the swarm was right 68% of the time. What is even more intriguing is that the swarm outperformed all of the study’s participants by 98%. “These results suggest that swarming, with closed-loop feedback, is potentially a more effective method for tapping the insights of groups than traditional polling,” the authors write.
The reason that it outperforms traditional polling is the lack of understanding of personal conviction in modern survey studies. “There is a big flaw when people collect opinions through polls and surveys. Very often polls and surveys treat all the answers as equal even though one person might respond to a poll with very strong conviction and another person might be torn between a few different options.”
Converging on an Answer
Swarm participants aren’t engaged in opinion or blind forecasting. The members engage by displaying their conviction at different moments in time. In many swarms that could total more than 100 participants, each member of the swarm provides a different level of conviction and confidence until the collective group accomplishes its goal.
“What it’s really doing is it’s finding the solution that optimizes across, not only their collective knowledge and opinions but also optimizes across their varying levels of conviction. That makes a swarm very powerful.” Rosenberg says. “The Swarm doesn’t just reach an answer; it converges on an answer.”
Looking forward, Unanimous will begin to build swarms in different industries that want to explore the knowledge of its experts and enthusiasts. This insight comes at a time that finance still needs improvement in how expert analysis is disseminated and the continued reliance on a single person’s views of the markets. Rosenberg explains two reasons why the single-person forecast of any industry or future event is flawed from the onset. “[First,] single-person views are muddled by their own personal biases and their own personal perspectives, which are very hard to get away from, even if you’re trying to do that,” he says. “And two, expert predictions are very often colored by the goal of an expert to stand out. We see this in sports, but you also see it in finance; no one is going to get a lot of coverage by making the most logical prediction. They’re going to get a lot of coverage for making the most outlandish prediction. Even if it’s not conscious, both sports forecasters and financial forecasters, they have a bias to take the unlikely view or to profess the view that is unique. Not because it’s more likely to be accurate, but because if it is accurate, it’s more likely to generate attention for them.”
Rosenberg explains that the swarm removes these problems. “What’s nice about a swarm is it doesn’t do that. It never does that. It can’t do that because you’re taking input from 100 people scattered all over the country at once. They’re combining their knowledge, wisdom and intuition into a system where it converges on the best possible answers for the set of information that this group has. And it cancels out all of these strange biases that each participant might have.”
Are You Unanimous?
Modern Trader is partnering with Unanimous to evaluate and develop market forecasting questions that exhibit statistical predictive capabilities via swarms. In fact, we asked early participants to help swarm to determine answers to the following questions, which we completed in mid-January (see “How it works,” below.)
Moving forward, Modern Trader readers can join independent swarm intelligence teams and gain inside insight from this breakthrough in sports, cultural and financial intelligence.
Sign up at: http://unanimous.ai/trader/