Swarm A.I. Predicts Oscar Winners with 89% Accuracy

It was an unpredictable night in Hollywood at the 89th Academy Awards. Despite heavy favorites like Viola Davis and Mahershala Ali taking home Oscars, the shock of the night was Moonlight surprising victory for Best Picture and Warren Beatty’s stunning error in handing the trophy to La La Land.

And yet, despite all the uncertainty, one thing is certain: for the third straight year, groups of average movie fans were able to amplify their intelligence by working together as an Artificial Swarm Intelligence. Using their UNU platform, Unanimous A.I. created an “A.I. expert” out of a group of amateurs and that A.I. performed as well, if not better, than the vast majority of professional movie critics.

How did they do it?  

Unanimous tapped the intelligence of 107 average movie fans, asking them to predict the winners in 16 major categories, covering everything from Best Picture to Best Costume Design. Across those 107 individuals, the average person got just 54% of their picks correct (when working alone). Since most categories feature a pair of favorites, this means that the average person did slightly better than a random coin flip in picking winners.

But, when working together as a Swarm A.I., that same group was able to greatly amplify its performance to an impressive 75% accuracy. In fact, this group of movie fans, connected by A.I. algorithms, did better at predicting the Oscars than the experts at Variety.

Of course, getting the answers right is only half the battle. It’s just as important to know when to make a prediction, and when not to. Some categories, like this year’s Best Actor race, are so hard to predict they may not be worth making a forecast. For that reason, researchers at Unanimous A.I. asked the Swarm A.I. system to also register its confidence in each prediction, using a scale of $0 to $100 as a proxy.

As you can see, the Swarm A.I. did incredibly well the nine categories where it showed high confidence. In fact, the only incorrect pick came in the Best Picture category where Moonlight’s victory can be considered, according to VOX, the “biggest upset in Oscar history.” As the Los Angeles Times noted, “a movie doesn’t earn a record-tying 14 nominations and not win Best Picture.” Still, where the experts were certain La La Land would win, the swarm displayed below average confidence. And that kind of insight is borne out in the swarm’s stunning 89% accuracy in its Best Bets.

If you’re not familiar with how Swarm Intelligence arrives at predictions, the replay below should be useful. Each magnet you see is a user in a virtual negotiation that combines real-time insights from real people, distributed all around the world, using A.I. algorithms modeled after swarms in nature. Each individual participant becomes a “human processor” within the emergent Swarm A.I. system, providing knowledge, wisdom, and intuitions that are combined with the input from others to achieve optimized predictions, decisions, and forecasts. And, in this case, the swarm correctly predicted that The Salesman would win Best Foreign Film, despite Toni Erdmann being favored.

The full list of the swarm’s picks were published last week here, and are scored below. As you can see, the Swarm did remarkably well in its “Best Bets”, but as expected, did not do as well in the picks that yielded low confidence.  In other words, the table below shows that the Swarm A.I. was a good judge of what it could confidently predict, as well as what it could not predict.  That’s the definition of insight.

After three years of predicting Academy Awards, we now know that Artificial Swarm Intelligence gives deep insights on Oscar night. Not only can thinking together as a Swarm A.I. allow regular movie fans to improve on their own performance dramatically, it empowers them to consistently outperform the experts.

Keep in touch with the swarm by dropping us a line below…