If you really want to know who will win Oscars on Sunday, whose opinion would you trust?
According to Unanimous A.I., you should look for answers from a group of 53 people with absolutely no expertise at all–who made each of their predictions in under one minute. Using a software platform that brings people together to make collective decisions, Unanimous A.I. has been remarkably accurate in its predictions–of everything from the Super Bowl to the Iowa Caucus. It’s latest challenge? Predicting the Oscars.
While most AI research is focused on simulating neural networks, Dr. Louis Rosenberg, CEO of Unanimous A.I., said his company is interested in “another way that nature has of doing intelligence”–swarms. Birds, bees, fish, all belong to swarms that are able to make incredible, life-saving decisions as a group. And since humans haven’t evolved to do this, Unanimous A.I.’s platform, called UNU, allows humans to collaborate and “swarm” artificially.
So how does it work? Imagine a group of five people want to pick a restaurant for dinner. Three vote for Indian. One for Italian. One for barbeque. A simple vote would go to Indian. But that’s not the way a swarm (or real life) works. The Italian-food voter may have the best argument for their pick. Or two people may be vegan. The decision doesn’t rest on any individual votes, but how the group weighs options together.
“Wisdom-of-the-Crowd algorithms are known to outperform experts in many domains,” said Roman Yampolskiy, director of the Cybersecurity lab at the University of Louisville. Marie desJardins, AI professor at the University of Maryland, Baltimore County, also believes in the power of the crowd.
“The use of ‘swarm’ is a clever kind of crowdsourcing,” said desJardins. “Instead of each user voting just once, independently of the other voters, it basically lets the entire community of users ‘see’ what everybody else is advocating for.”
Why does it work? “If a user can see their first choice has no hope of winning, they can at least advocate for their second choice,” she said. “Sort of like a really fast, real-time runoff election.”
Unanimous AI did a pretty great job predicting the Oscars last year. A sub-group of 7 people (all non-experts) were able to guess 11 out of 15 categories correctly (73% success). All decisions were made in under a minute. Experts at the New York Times, it should be noted, had a 55% success rate.
Predictions for the 2016 Academy Awards
The 2016 Academy Awards are on Sunday and here are Unanimous A.I.’s predictions for the winners of the major awards (click the hyperlinks to see the swarms in action):
Best Picture: The Revenant
Best Actress in a Leading Role: Brie Larson (Room)
Best Actor in a Leading Role: Leo DiCaprio (The Revenant)
Best Director: A.G. Iñárritu (The Revenant)
Best Actress in a Supporting Role: Kate Winslet (Steve Jobs)
Best Actor in a Supporting Role: Sylvester Stallone (Creed)
It’s worth paying attention to where differences lie (in bold). Based on results from Gold Derby, a website offering a “hyper-focused analysis,” New York Times reporter Cara Buckley predicted a different winner for Best Actress in a Supporting Role: Alicia Vikander, co-star of “The Danish Girl.”
While there are other ways of predicting, like simple polling, surveys, and prediction markets, Rosenberg thinks Unanimous A.I.’s software shows more promise. “Prediction markets actually behave more like polls,” he said. “The problem with polls, and surveys, and markets is that they’re sequential. Once somebody votes or places a bid, everyone else sees what happened and it influences everyone who comes after them.”
With a swarm, he said, “people aren’t working sequentially–they’re working simultaneously. They aren’t having these swings and bubbles, so we get answers very quickly.”
Most people in the swarm, he said, haven’t seen all of the movies. But it works because people “fill in each other’s gaps in knowledge.”
And if a group of non-experts can predict the Oscars this accurately–and quickly–it is easy to imagine the implications for what else it could do. Currently, Rosenberg is seeing the most interest in the software from marketing companies who want to learn what their customers might respond well to. But it also could have implications for predicting results of something even higher-stakes: an election, for example.
The 3 big takeaways for TechRepublic readers
1. Swarm AI looks to natural swarms to see how collective decisions are made, and uses this in its own artificially constructed human swarms.
2. Unanimous A.I. has predicted events with greater accuracy than experts and polls.
3. The swarm AI method is fast–under a minute–and is composed of a group of non-experts.