The NCAA's March Madness - the men's college basketball tournament - is already well underway. And in fact, as I write this on Saturday morning, we're already down to 32 teams - so, presumably, we've already missed half the fun. That said, there are two more weeks of excitement planned, with the championship game taking place on Monday, April 7.
Intriguingly, the men's tournament so far has had none of the "Nobody-Saw-That-Coming" type of unexpected - and unbelievable - upsets that have become an integral part of its March Madness brand. For perspective, nobody calls the NFL playoffs "January Madness," do they? In the league of Vince Lombardi, not only does one of the top-ranked teams usually win the Super Bowl, but the Cinderella teams (those 10,000:1 underdogs) rarely make it very far - at least, not after winning that one upset that nobody saw coming.
On the other hand, it may seem somewhat astounding, but there's no record of anyone ever creating a perfect bracket for the 64 teams that make up March Madness, as the NCAA itself makes clear:
The longest (verifiable) streak of correct picks in an NCAA tournament bracket to start the beloved March Madness tournament is 49, a mark that was established in 2019.
So, what does this have to do with AI? Simply put, since artificial intelligence has become ubiquitous, it was only a matter of time before it found its way into trying to crack the March Madness code. And that's something that should be a perfect job for AI, right? It's a task that involves crunching a huge amount of data in order to come up with the most probable outcomes. And that's what we've seen so far with AI in this year's March Madness: For the most part, the various AI tools have picked the most highly favored teams - just as you and I are likely to do.
The problem is this: What data could help AI predict the improbable upsets and the Cinderella(s) of a tournament? We're talking about such human - and unknowable - factors as an underdog's burning desire and quiet confidence vs. a heavily favored, highly ranked team's overconfidence verging on cockiness. You know who I'm talking about: those number one seeds whose stars are focused purely on what they're going to pack for their trip to the Final Four and how they'll thank their parents after the championship game on April 7.
Again, in this year's tournament we haven't yet seen any big upsets. And, by definition, if they don't appear in the first couple of rounds, all the would-be Cinderellas will be headed home. Which means that this year's results - at least, so far - are very comfortable territory for AI.
As MarketWatch reports, ChatGPT produced a bracket featuring a Final Four consisting of Auburn, Kansas, Duke, and Houston, with Duke winning the title. No big surprises there, as three of those four teams are No. 1 seeds, and Duke is the consensus top-ranked team in the country.
So, if this is a year that produces no Cinderellas, what does that mean? Will there be a good number of first-ever perfect brackets this year? That certainly seems possible, whether they're produced by humans or AI.
And yet, again per the NCAA, the odds of anyone getting a perfect bracket are 1 in 9.2 quintillion for anyone just guessing or flipping a coin. For those who follow college basketball, the odds are somewhat brighter, but still an almost unfathomable 1 in 120 billion. That kind of feels like a long shot to me, but crazier things have happened.
Ultimately, AI can crunch the available data, but in more than a few cases, it’s that as-yet inaccessible data that matters most. How does one measure heart and soul? Burning desire? Confidence vs. overconfidence? I have no doubt that all those data points are coming to AI, but they’re not here yet. So, for one more year at least, I don’t think AI is your answer for picking the perfect bracket.
At any rate, my money’s on Duke.
