How to Use AI to (Maybe) Win March Madness - 4 minutes read




For most people, predicting the NCAA’s March Madness tournament results with any degree of accuracy is a fool’s errand. The single elimination tournament, comprised of 64 teams, one of which must win six straight games to be crowned champions, is notoriously unpredictable, with unknown Davids upsetting legacy Goliaths almost every year.


Enter the machines. With the popularity of predictive algorithms and data models informing elections in recent years, it’s no surprise that March Madness has been swept up in the craze, with a surge of AIs taking a stab at predicting the big dance. Knowing this, perhaps while staring at a daunting bracket, begs a question: Should you consult an AI before filling out your bracket?

The answer, while incumbent upon who you are and your understanding of the intricacies of college basketball, is a definite maybe.

Are AIs better at predicting the tournament?

According to Matt Osborne, a mathematician and post-doctoral researcher at Ohio State University, an AI-generated model can help better inform your bracket in three ways.

The way he explains it to Lifehacker, a predictive model can give you a good idea of where things in the tournament are likely headed, but it certainly can’t predict the tournament outright. He explains that there are certain criteria that can bolster your chances of accuracy if you use an AI: overcoming your personal biases (which are rampant in sports), gauging historic outcomes, and a general estimate of probability (which he likens to traditional Las Vegas betting odds).

He delves into further detail in an email, writing:

You may have personal biases that won’t be present in the data. For example, you think that team A is good because you’ve heard of them before, but it turns out they aren’t so great this year.

Using historic outcomes can help you guess better than at random. Picking teams randomly is basically like flipping a quarter, but the chances of team A beating team B is typically not 50-50. For example, a 16 seed has only upset a 1 seed one time in the tournament’s history (UMBC against Virginia in 2018), so you’re pretty safe penciling in the top dogs in each region.

[Algorithms] can give you an estimate of the probability that something will happen (similar to Vegas odds), which can help you judge for yourself just how likely that upset pick you’re contemplating is.

Osborne says AI tools are generally useful for layman fans and well-studied experts alike. He adds, however, that the diehard fans are “probably better equipped (because of their existing knowledge) to understand the input and output of the tool, and check to see if what the machine suggests actually sounds feasible.”

How accurate are the AI brackets, usually?

Asking an AI to predict a college basketball tournament is a bit thornier than asking one to predict a presidential election. Though the latter can be subject to irregularities—as was the case in 2016—the models are fed a steady stream of polling, which is predicated upon the more reliable act of people voting. Basketball, especially in an environment such as March Madness, is more volatile.

Osborne says in his experience, “only the best brackets ever get coverage, and because the tournament is pretty random, the absolute best brackets are usually not ones created using a data science tool.” As an example, he points to the case of 12-year-old Sam Holtz, who filled out a perfect bracket in ESPN’s 2015 March Madness challenge, beating over 11 million other contestants. Maddeningly, Holtz didn’t even watch basketball with any degree of regularity, and operated off simple hunches when picking his victors. And then, somehow, he made history, filling out a perfect bracket that defied gargantuan odds that one business professor calculated were “somewhere in one in some number of quintillions.”

No predictive model has ever done that, and it isn’t likely that one will. But that isn’t to say that an AI-generated bracket won’t be useful if you decide to enter a contest. For guideposts, Osborne recommends Five Thirty Eight’s model (which is free) and Sportsline’s model (which comes with a fee).

Even with a model at your disposal, you should feel fine operating off instinct and bias as well. We’re talking about sports, after all.

Source: Lifehacker.com

Powered by NewsAPI.org