In 2019, the Miami Heat NBA team wanted to know which of its services most satisfied its season ticket holders. The franchise believed that if it knew what drove fans to renew their memberships it could boost renewals and increase its income. It had options. The team could have simply handed out a survey and asked its fans to score from one to five a list of the benefits they enjoyed as season ticket holders. It would then have spent time crunching the data to understand sentiment. That would have been the usual way to gauge customer feeling.
But it took a different approach. The franchise turned to a “Swarm AI” platform created by Unanimous, a new kind of artificial intelligence company. It brought together a group of season ticket holders and asked them typical survey questions like “How likely are members to recommend a membership to friends/family.” But instead of each participant then giving a score to show their satisfaction then turning in their papers, they worked together. They simultaneously moved a kind of virtual magnet across a hexagon whose corners marked different degrees of likelihood. As the magnet drifted towards “a little likely” for example, participants could try to nudge it towards “moderately likely” or “very likely.” Algorithms analyzed the forces acting on the magnet to give Miami Heat an overall picture of how the group felt as a whole rather than a collection of data points that illustrate how each member feels at one moment in time.
The result was that the franchise was able to see that among fans, price is the most important factor in renewing memberships, beating access to playoff tickets and a guaranteed seating location. The most valuable experience is the chance to take part in private events with players and coaches, which is much more important than the availability of courtside seats during warm-ups. The team ended up with a clear idea of what it needed to provide to season ticket holders in order to generate renewals and land more recommendations.
That wasn’t the first time that Unanimous had successfully turned its platform to sport. In 2016, the company brought together readers of The Boston Globe and asked them to use the service to collectively name ten MLB playoff teams. Using Swarm AI, they got nine right, as well as the last eight teams standing, the two World Series teams, and correctly picked the Chicago Cubs as the winners. The following year, baseball fans using the platform picked every division winner and wildcard heading into the postseason. A survey couldn’t have done that.
The idea behind Unanimous comes from Dr. Louis Rosenberg, a pioneer of virtual reality systems. He founded the company in 2014, and in a 2017 Tedx talk, laid out the concept underpinning the platform. What he was building, he said, was a “hive mind” that could be more powerful than the artificial intelligence systems, like Google’s Deep Mind, currently in production. His inspiration was the swarming patterns formed by groups of animals.
Species, he argues, amplify their intelligence by forming real-time systems. “This is why birds flock, fish school, and bees swarm – they make significantly more accurate decisions when thinking together in systems than they could on their own,” he explained by email. Birds form murmurations in which the decisions of individuals determine the actions of the group. Fish swim in schools to make it harder for predators to isolate individuals. Like birds, the collective decisions of multiple individuals determine the direction of the group.
“If birds and bees and fish can get so much smarter together – it should work for humans too if we can create the right technology to connect people together.”
Another way to think of what Unanimous does is to imagine the human brain. A single synapse has little that can be described as an intelligence. It passes an electrical impulse from one place to another. But put billions of those synapses together and the human brain can do remarkable things. Unanimous attempts to be a platform that brings together large numbers of human brains to work and decide together.
It’s that togetherness, coupled with simultaneity, that’s important. Votes, polls, and surveys, says Dr. Rosenberg, treat each individual as single points of data that have to be aggregated to form a statistical model. “Most people don’t know how they really feel when a question is posed to them on a survey,” he noted. “Or, if they do know, they can’t adequately express it in simple terms of some abstract scale (like a rating of 1 to 5). But in swarm, they discover their true feelings as they interact with everyone else, and the AI engine is watching their behaviors, determining where there is true conviction and where there is ambivalence, and finding the optimal combination of perspectives.”
Even markets, he adds, only create transactions between a buyer and a seller with no group interactions among participants. Each purchase is an individual choice, largely unaffected by the decisions of others.
“A swarm is a real time system where all participants are engaged at the exact same time, all of them acting, reacting, and interacting with the full population,” says Dr. Rosenberg. “This is a true crowd, although scientists refer to it as a ‘swarm’ to distinguish.”
It’s also more reflective of the way decisions are made and sentiment changed in the real world. Our purchase decisions are affected to some degree by the purchase decisions of other customers. Seeing people who are similar to ourselves raving about a brand will affect how we feel about that brand, pushing our feelings from “skeptical” to “open” or even “positive.” Websites now often feature small pop-ups that tell us that someone in Delaware or Dhaka has just bought the item we’re looking at. Those notifications build trust. They tell us that the customer base with similar interests to our own is moving in a particular direction, and they drag us in that direction too.
Crowdsourcing and betting markets work in a similar way. There too, the collection of individual decisions can affect the decisions of others. As more people place orders for a product on a crowdsourcing site, for example, they both show faith in the concept of the product and they increase the chances that the product will be built and future orders will be fulfilled. As gamblers place their bets, bookmakers change the odds, indicating the choice made by the market and affecting future decisions. As the odds fall and a favorite emerges, betters start to doubt their own decision to choose a different option, and bookmakers have to increase the odds to tempt them to do so.
Where those markets differ from Unanimous’s Swarm AI, though, is that each decision in real life is made once. On Unanimous’s platform, members of the swarm are constantly negotiating their choices, while algorithms track their behavior to understand the overall direction of the group.
The approach appears to work and not just in sport. In 2018, eight radiologists from Stanford University Medical School used Unanimous’s swarm intelligence to review a set of 50 chest X-rays and determine whether the patient had pneumonia. The same X-rays were also analyzed by an AI software algorithm called CheXNet. The study found that when using Swarm AI, the doctors reduced their diagnostic errors by 33 percent.
But Unanimous has been particular effective at using swarm intelligence to make predictions, especially about sport. In addition to predicting baseball games, the company also asked groups of basketball fans to use the platform to predict 238 NBA games across 25 weeks. Bets placed using those predictions produced a 57% return on investment. “This was a remarkable result and the first time an AI technology had demonstrated the ability to consistently beat the oddsmakers,” says Dr. Rosenberg.
The results led to a flurry of interest by sports fans. Unanimous responded by launching Sportspicker.ai, a service offering sports predictions generated by Swarm AI. But the move into sports forecast was always intended as a testbed, a way to confirm that the technology could use swarming to enhance human intelligence. Dr. Rosenberg has already tested the power of swarms in medicine and in finance. In study run with researchers at MIT, Unanimous asked groups of financial traders to predict changes in the price of financial assets such as gold, oil, and the S&P 500. The results showed a 36% increase in forecasting accuracy compared to traditional forecasting.
Build Your Own Swarm
In nature, swarms are big. Starlings can form flocks of as many as 100,000 as they swirl over fields at dusk. But it turns out that you don’t need as many human brains to come together to form an effective brain of brains. In one study conducted with California Polytechnic University, Unanimous brought together 60 groups of between just three and five people, and challenged them to take a standard intelligence test. Some of the groups worked alone and others converged on answers together using Swarm AI. The researchers found that when using Swarm AI technology, the groups reduced their error rate by more than 50 percent, a result that surprised even Dr. Rosenberg.
That result has real implications for businesses. Unanimous has built a piece of technology. It has a digital interface and algorithms that can analyze the pressure that members of the swarm can place on a virtual magnet. But if its studies are accurate, its biggest contribution has been to prove Dr. Rosenberg’s argument that swarms provide a more powerful form of artificial intelligence than either robotic synapses or individuals working alone. And while Unanimous’s platform is one way to measure the direction of a swarm, it’s not the only way.
In 2000, Southwest Airlines had a problem with its cargo operations. Its planes typically used only 7 percent of its cargo space but some airports lacked the capacity to accommodate its freight schedules. Employees would load freight onto the first plane going in the direction of the cargo’s final destination, offload it, then spend time moving cargo and filling aircraft.
Southwest applied swarming intelligence. The company copied the way that ants always found the most efficient way to find food and move it to the nest. Instead of constantly unloading and reloading cargo, they discovered that it was more efficient to leave the cargo on a plane until that plane reached the cargo’s destination. The result, according to the Harvard Business Review, was a reduction of as much as 80 percent in freight transfer rates at the busiest cargo stations and a fall in the number of cargo storage facilities. By copying rules followed by swarms of ants, Southwest was able to save about $10 million.
The Harvard Business Review also described how Jim Donehey, as CIO of Capital One, applied the same principle. He created four rules in the IT section : always align activity with the business; spend money like its your own; don’t box yourself into one way of acting; and show empathy with others. Donehey then distributed 10,000 gaming chips in four colors to represent each of the rules. Whenever someone in the IT group followed a rule, they received a chip. When they followed all four, they got a special chip. The result was that within a year, the rules had been embedded in the organization and its 1,800 employees were better able to work together as a group.
It would be great to say that the benefits of swarming can be won as easily as bringing a small group together and holding a discussion. It’s a bit more complex than that. But you don’t need a lot of people to form a swarm and you don’t always need sophisticated technology to understand or define its direction. You just need a group, a question, and a recognition that lots of brains are better than one.
Recent Comments