75 Fans crush ESPN’s predictions by using Swarm A.I.

What is an “expert?” It could be a master chess player, planning 10 moves ahead.  Or a skilled doctor making a critical diagnosis. Or a veteran mechanic who knows everything about your Jeep. Specialists who posses deep knowledge on a given topic are incredibly valuable to society.

But, what if EVERYONE could be an expert?

New technologies are enabling regular people to rival the experts by pooling their knowledge, insights, and intuitions.  The process is called Swarm Intelligence and its incredible potential is just beginning to be unlocked.  This inspired us to ask the bold question:

Can Fans out-predict ESPN on College Bowl Games?

To test this, researchers at Unanimous A.I. asked a “human swarm” of 75 sports fans (from Reddit) to predict 9 Bowl Games and to name the eventual national championship. They did this using UNU, an online social platform that allows groups of people to answer questions, together, by pooling their wisdom as a collective swarm.  This is what it looks like when a swarm makes a prediction, each magnet controlled by a unique user:

Rose Bowl 2015 Stanford by 7

To compare performance, researchers also asked those same 75 fans to make individual predictions using a standard survey poll.  This allows them to quantify how much smarter the Swarm Intelligence might be, as compared to individuals making predictions alone.  Researchers also collected predictions from major sports experts, including ESPN and FiveThirtyEight.

So, who’s smarter: Experts, Individual Fans, or Swarms of Fans?

The results are amazing. If you’d like to see predictions and confirm they were all made in advance of the games, we’ve posted them on this blog as well as across Reddit.  For your convenience, you can see the Swarm’s picks: here.  ESPN’s expert picks: here.  FiveThirtyEight are here and here.

THE BASIC RESULTS:

  • Individual Sports Fans (across 75 surveys): 4.64 / 9 correct picks against the spread (52% accuracy)
  • ESPN College Football Experts (posted on their site): 5/9 correct picks against the spread (56% accuracy)
  • An UNU Swarm (75 fans working as one): 6/9 correct picks against the spread (67% accuracy)

WAIT, THERE’S MORE

In addition to the 9 bowl games, predictions were made for the Playoff Champion. These predictions were the hardest, because they were done before the semi-final games were even played. So there was no guarantee that a team would make it to the final, much less win. Here’s the pick:

College FB Playoff Champ - Alabama

EPSN, on the other other hand – picked Oklahoma to win it all.  So, they’re already out.  This means we can look at their accuracy across all ten predictions.  This puts ESPN at 5/10. Despite all their “expertise”,  with 50% accuracy, ESPN experts did no better than a coin flip in picking winners.

Nate Silver’s FiveThirtyEight put just 3 picks on record for college bowl games.  These picks were supported by 10 pages of stats and figures, but did even worse, ending up with 1/3 correct on their picks.   Thus, FiveThirtyEight scored a paltry 33%.

TO SUMMARIZE:

  • FiveThirtyEight = 33% accuracy and already failed in the final playoff Prediction.
  • ESPN = 50% accuracy and already failed in the final playoff Prediction.
  • UNU Swarm (of fans) = 67% accuracy, with its pick of favored Alabama still alive

HOW IS THIS POSSIBLE?

When groups come together as a real-time swarm, not only are they able to pool their intelligence in real-time, they can fill the gaps in each other’s experience and vary their input based on their confidence in their answer. So, the answers produced in UNU reflect not just the knowledge of the group, it also maximizes the group’s confidence and conviction in their answers.

The remarkable thing is that people are smarter working together as a swarm, than they are on their own.  This is clear when we look at the predictions made by the 75 individual sports fans who filled out a survey.  Just over 95% of those average sports fans would have been better off if they had placed bets using the swarm’s predictions, rather than their own individual predictions.  That quite an endorsement.

Of course, the researchers used “average sports fans,” but what would happen if they used a swarm of experts?  Could UNU produce a super-intelligence that aims at near perfection?  Could swarming change the very nature of what it means to be an “expert?”

SHOW ME THE MONEY!

It’s no secret that gambling is a big reason sports are so popular. ESPN’s Chalk is an full website dedicated to helping readers beat Vegas.

Photos taken at Las Vegas.

So, what would happen to your hard-earned money if you bet with average user, the experts, and the swarm? To convey these numbers, we calculated the theoretical results of $10 bets placed Against the Spread (ATS) for all picks.  Here’s how it shook out:

RESULTS vs VEGAS:

  • Individual Sports Fans (across 75 surveys): 4 / 9 wins ATS , yielding a cash loss (-16 % ROI)
  • ESPN College Football Experts: 4 / 9 wins ATS, yielding a cash loss (-16 % ROI)
  • An UNU Swarm (75 fans working as one): 6/9 correct picks ATS, yielding a big gain (+27 % ROI)

In other words, the “human swarm” earned (+27%) on every dollar bet, across 9 games, while ESPN lost (-16%) over the same 9 games, a 43% improvement. And, the swarm still has a shot at winning the Playoff, (which ESPN already lost). This means the swarm could boost its ROI to well over 30%, a stunning feat for even seasoned sports gamblers.

** 1-12-16 UPDATE: After Alabama’s victory over Clemson in the title game, it’s possible to calculate a final ROI for both ESPN and the Swarm. After placing ten $10 bets (  9 bowl predictions and 1 prediction on the eventual champion before the playoffs started ) the numbers look like this:

  • ESPN (EXPERTS): is left with $75.79 out of $100, for a -24% ROI 
  • UNU (SWARM of FANS) finishes with $133.97, for a 34% ROI

Betting with the Swarm of 75 average fans yields a 58% improvement in ROI vs listening to the ESPN experts.**

WHAT DOES IT ALL MEAN?

Although this is just one study predicting College Football games, many prior studies also show that groups, when working together, tap into levels of human potential that we’re just beginning to comprehend.  Follow this blog, and we’ll keep you up to date on all the latest results.  Also…

Want to try swarming?  Now you can.  The UNU platform is now open to BETA USERS, with swarms forming on a range of topics from sports and politics, to movies and music.  If you’d like to join the Beta program, just fill out the form below:

SIGN UP


 

This Post Has 2 Comments

  1. Pingback: 5 – Can A.I. Turn Average Fans into Experts?

  2. Pingback: 4 – 75 Average Fans Crush ESPN’s Bowl Picks by Using A.I

Comments are closed.