Since the 1950’s, when researchers began searching for ways to create artificial intelligence, much focus has been on developing artificial neural networks–building intelligence from scratch. Turning this concept on its head, an approach dubbed “artificial swarm intelligence” uses the power of nature in a different way: by harnessing groups of human minds to come up with predictions for real world events.
Dr. Louis Rosenberg, CEO of Unanimous AI, is building a software platform, UNU, that assembles groups to make collective decisions. “What’s different about this is that it fundamentally keeps people in there,” he said. “We’re focused on using software to amplify human intelligence.”
It’s not technically AI–“if a program can’t be intelligent without people, it is not artificially intelligent,” said Roman Yampolskiy, director of the Cybersecurity Lab at the University of Louisville.” And the “wisdom of the crowd” concept, derived from human-based computation models, goes back at least a dozen years. Yampolskiy himself developed a similar algorithm, dubbed “Wisdom of Artificial Crowds,” using a group of “simulated intelligent agents” to make predictions.
Still, the decision-making capability that Unanimous AI has tapped into is impressive. The UNU platform has been remarkably accurate so far in predicting the outcome of things like last year’s Oscars, the Superbowl and even political races.
The idea behind swarm intelligence is to learn from what happens in nature. “It’s why fish form schools and birds form flocks and bees form swarms,” said Rosenberg. “In a nutshell, it’s allowing the group to make better decisions than individuals could make alone.” This type of swarm intelligence, it should be noted, is not quite the same as another kind called “flocking,” which applies to robotics. “Flocking” is a way that groups can efficiently navigate through environments, relying on each other for help.
“There’s a vast amount of knowledge and wisdom and intuition in groups that we work to tap into,” said Rosenberg, “to create intelligence that amplifies their natural abilities.”
Applying rules from other natural environments to humans
The honey bee, said Rosenberg, which is “remarkable” when it comes to decision-making, is the inspiration for the types of systems he creates. “Most people don’t appreciate how amazing they are,” Rosenberg said.
Picture this: each year, honey bees must split off of their colony and send out scout bees to find a new location for their new home. Hundreds of scouts are sent out, some traveling up to thirty square miles to find the best new home, and returning to the hive to form a swarm and make an assessment of the new environments. Four hundred to 600 bees engage in a “waggle dance” to convey information to the group. “The swarm is how they make this decision,” said Rosenberg. “It turns out it’s a very complex decision. They’re factoring a whole bunch of conflicting variables.”
Professor Healy at Cornell found that the bees make the optimal site over 80 percent of the time, said Rosenberg. “What’s fascinating is that no single bee could possibly make that decision.”
How humans get it wrong
While the bees are engaging in “real-time negotiation,” humans often have a different, less accurate approach to making predictions. Typically, humans use polls and votes, which Rosenberg calls “primitive.” They’re often wrong, he said, because they’re polarizing. “Instead of finding common ground, they force us to entrench in predictions and make it harder for us to find the best answer for the group.”
So Unanimous AI aims to avoid entrenchment and stagnation by imitating the bees, which, through “millions of years of evolution will come to the decision that is the best for the whole group.”
How does swarm intelligence work?
UNU is an online platform where anyone can log in and answer questions as a swarm. “People benefit when they form a real-time system the way bees, fish, and birds do, as opposed to people just casting a single vote the way people do,” said Rosenberg.
The swarm in action, said Rosenberg, is a little like “a massive online Ouija board.” (It’s worth checking out their video to see how it works.) Participants are given a glass puck on a string, and everyone moves that puck together. “Everyone is holding a little magnet,” Rosenberg explained. “And hundreds of people are all pulling on the same puck at the same time.”
Since humans can’t waggle dance, they perform the swarm action by collectively moving the same puck on the screen at the same time. For example, everyone will see a question like, “Who’s going to win best actor?” with a list of choices. Each member of the group starts moving the puck to one of the choices and as it starts moving, everyone changes their opinions in real-time. “Let’s say it’s moving towards an opinion that you agree with–if you see others pulling that way, you start pulling with even more conviction,” said Rosenberg. “Everyone is changing and switching their pull in real-time. If the group can’t find common ground, they’re not making any progress.”
The process requires negotiation, finding a mutually-beneficial solution.
Results of swarm predictions
Fifty people (with no prior experience) worked together on UNU to predict the Academy Awards. As individuals, they were polled to make predictions. The average was six correct out of 15. The experts got nine right. The swarm? Eleven. “They went from being 40 percent accurate as individuals, said Rosenberg, “to being over 70 percent accurate when they worked together as a swarm. It happens because people can fill in the gaps in each other’s knowledge.”
The implications of this could be profound. While betting on the Oscars or sporting events may not have a substantial impact, software like UNU could have plenty of business applications, like deciding if a particular business strategy will be successful. It could also change the game when applied to something like political races. Six months ago, UNU asked people about predictions for political candidates. “The swarm reflected strong preferences for Bernie Sanders and Donald Trump,” said Rosenberg.
At the time, he thought the algorithm must have been broken.