In the world of sports forecasting, the on-air expert is widely accepted as the gold-standard. Every week, highly paid analysts and former players fill hours of pregame shows with their predictions and (sometimes subtle) tips for would-be gamblers. But for all of their insider information, there is one other predictive model that these experts must often contend with: Paul the Octopus. The late, great, eight-legged wonder rose to fame during the 2010 World Cup when he correctly identified the winner of 12 out of 14 match-ups. With all due respect to the late Paul, his performance at the World Cup might be the clearest case on record for the existence of luck in forecasting.
None of this is to say that animals aren’t intelligent. Just the opposite. Thanks to “swarm intelligence,” animals in nature are capable of solving and optimizing incredibly complex problems. The point is, when judging Paul’s predictive performance, it’s crucial to track results over an extended period. Consistency over the long run is the most reliable way to verify performance.
All season long researchers at Unanimous AI have been tracking their Swarm AI platform’s ability to forecast NFL games in comparison to 120 experts from places like ESPN, Sports Illustrated, NFL.com and more. Those expert results show that the highest-performing expert achieved an impressive 70 percent accuracy throughout the NFL season. In comparison, the Swarm AI system’s 75 percent accuracy outpaced every single expert. The technology might be based on Swarm Intelligence, but this is no gimmick. Unanimous AI has produced an AI super-expert by combining the intelligence and intuition of a group of regular NFL fans.
With that in mind, let’s take a look at the insight generated by Unanimous AI for the NFL Division Playoffs. The field has been narrowed down to eight teams, and it’s time to separate the true contenders from the pretenders. The chart below lays out the predictions for Tennessee at New England, Jacksonville at Pittsburgh, New Orleans at Minnesota, and Atlanta at Philadelphia. This Google Spreadsheet shows how the swarm made up its mind.
As you can see in the above chart, the predictions have been sorted according to confidence, although the “Bet” isn’t a recommendation that anyone place a wager so much as an expression of the AI’s relative conviction. To see how that confidence is captured, it’s worth taking a moment to compare two match-ups.
New Orleans Saints vs. Minnesota Vikings, Swarm AI Prediction
Up first is the highly anticipated clash between the Minnesota Vikings and the New Orleans Saints. This contest pits the Saints’ longtime star Drew Brees against the “ultimate breakout” sensation, Case Keenum.
The Saints are widely considered one of the toughest outs in the playoffs, but they have to go on the road to Minnesota, who recorded their best regular season record since 1998. As a result, the Swarm AI system expressed significant support for both teams en route to an optimized prediction of Minnesota to win, albeit with Low Confidence. The movement of the puck shows quite clearly why researchers at Unanimous filed this game under the “Proceed with Caution” banner.
Tennessee Titans vs. New England Patriots, Swarm AI Prediction
In comparison, despite the brewing controversy between Tom Brady, Bill Belichick, and Robert Kraft, the research produced by Unanimous AI suggests that the Patriots’ on-field relationship is as healthy as ever. The dominant franchise in the NFL over the past decade or so is the current odds-on favorite to claim yet another Lombardi Trophy.
The replay below shows that this group of NFL fans was able to quickly converge on a predicted Pats’ victory, and with High Confidence. So, while it’s too early for the technology to register its own Super Bowl prediction, barring another Mariota-to-Mariota miracle, it’s a “Best Bet” that Tom Brady & Co. will advance.
The above guest post has been edited by Inverse.