Machine Madness: Can AI Crack The Perfect Bracket?

Welcome to this week’s Deep-Fried Dive with Fry Guy! In these long-form articles, Fry Guy conducts in-depth analyses of cutting-edge artificial intelligence (AI) developments and developers. Today, Fry Guy dives into whether AI can get us closer to predicting a perfect March Madness bracket. We hope you enjoy!

*Notice: We do not receive any monetary compensation from the people and projects we feature in the Sunday Deep-Fried Dives with Fry Guy. We explore these projects and developers solely to showcase interesting and cutting-edge AI developments and uses.*

Is your March Madness bracket busted? If so, AI may be able to help …

Every year, millions of college basketball fans across the nation fill out their March Madness brackets, making predictions for all of the NCAA Tournament games. Each person makes their game predictions in hopes of becoming the first ever person to get every pick correct. Some people resort to analytics, some people rely on fandom, and others simply flip a coin. However, regardless of how you make your picks, some (if not many) are likely to be dead wrong. It doesn’t matter how much of an expert you are at basketball; within the first few games, your bracket is likely to be busted.

Nobody has ever predicted every game of the 64-team tournament correctly. In fact, the best performance in a March Madness bracket was achieved by Gregg Nigl in 2019. He correctly predicted the first 49 games of the tournament, setting a record for the longest streak of correct picks in a March Madness bracket. His bracket remained perfect until the 50th game! With 67 total games to predict, however, this still fell significantly short of perfection. People have wondered for a long time whether it will ever be possible for someone to predict a perfect bracket. Here is where AI may come into play.

Given that AI is so good at making predictions about language, math, and other things, it seems intuitive to think that it may also be effective at predicting the outcome of college basketball games. Could AI be the key to cracking the code and crafting a perfect bracket?

WHY ARE MARCH MADNESS BRACKETS SO DIFFICULT?

After the first round of the 2025 NCAA Tournament on Thursday and Friday, only 181 brackets remained perfect of the 24.4 million on ESPN. As the games continue to be played, this number is quickly dwindling. However, this kind of stat is not unique to this year. Every year, after the first round (32 games), nearly everyone’s bracket is busted. In fact, this year is the first time since 2019 that any brackets have remained perfect after round 1. The reason for the difficulty? It’s simple math.

The number of combinations for a March Madness bracket are approximately 1 in 9.22 quintillion (1/9,223,372,036,854,775,808). Let’s put this in perspective. If you filled an entire football stadium knee-deep in sand, mixed in one special grain, and someone gave you one chance to pick that exact grain—your odds would be better than getting a perfect bracket.

Now, it’s true that most people don’t guess entirely randomly. For instance, most people won’t pick the worst team to beat the best team in the first round. To account for this, it’s been speculated that those with “some basketball knowledge” have odds of around 1 in 120.2 billion to predict a perfect bracket. So, better … but still astronomical odds.

Given that the odds are not in our favor, are we hopeless? Perhaps not, thanks to machine learning. AI may come to the rescue and save humanity from its yearly defeat. How AI might help people get closer to a perfect bracket is best understood when we look at some of the common mistakes in human predictions and reflect on how AI models may improve on those faults.

AI KNOWS BALL

Poor performance in March Madness brackets is often due to a variety of factors, each of which AI may be able to improve upon. Let’s look at them one by one.

1. Our judgments are clouded by fandom and emotional influence.

Most of the time, college basketball fans are unable to step outside their own biases when filling out March Madness brackets. Duke fans are going to have a bias for Duke, and they will probably never pick North Carolina to make it very far. And Tennessee fans are not going to have the Vols losing to Kentucky or Alabama. These sort of biases can make it difficult for college sports fans to step outside their own emotional ties to teams and choose with their heads rather than their hearts, sometimes leading to poor decisions.

Unlike us humans, AI doesn’t have emotional ties or affiliations to teams. It doesn’t stand in the crowd and scream at the refs or jump around and celebrate wins in front of the TV. It’s a prediction machine, which means it is not hindered by these emotional elements which may corrupt selections. This may lead to more data-driven insights, unhindered by biases.

2. We often ignore analytics or pay attention to the wrong statistics.

During March, sports commentators and fans love bringing out random facts about teams. These can include how well a team shoots the ball, how effectively they rebound, or even how many 12 seeds have beaten 5 seeds in the past 20 years. However, it’s often very difficult to determine which of these statistics are relevant and which are not when making predictions. We are also not exposed to all the facts—we are often only exposed to a select few. This limits our ability to make fully-informed decisions. These limitations are often exposed when one turns on the TV to watch the team they picked only to find out that the team doesn’t have a player over 6’6”. People find themselves saying, “I wish I would’ve known that when I made my prediction!”

This is not a problem for AI. Unlike a human, AI can process all of the available data about every team and, over time, discover which statistics are most related to game outcomes. AI will never be surprised by a statistic.

3. We don’t know all of the teams.

How much do you know about college basketball teams like McNeese, Wofford, High Point, or Drake? Probably not much. These teams often don’t play on national television, and they are not talked about on SportsCenter like some of the name-brand schools. For this reason, many of these teams fly under the radar until the tournament and then shock the world by winning games against more well-known schools. It’s unlikely that even the most dedicated hoop fans know about all 64 teams in the bracket. Sure, people can look up statistics about these teams, but this is not the same as following these teams throughout the season.

AI doesn’t just watch ESPN or listen to the popular radio stations; AI is able to learn from all data related to both well-known and lesser-known schools. AI is able to analyze the data and statistics to become an expert on every single team, even the ones you have never heard of. This gives the model the potential to make more informed predictions than the average human or even the sports analyst.

4. We often don’t analyze all games equally, especially in later rounds.

Most people spend their time studying the first round of March Madness. However, filling out a bracket takes time, and this can cause people to get mentally fatigued and a bit lazy when studying matchups in the later rounds. This is often why people who start really well begin to fall apart as the tournament goes on. Unlike a human, AI does not get fatigued or have a limited attention span. AI will put just as much analytic effort into the first game it predicts as it does the 67th. This allows for an intentional selection of every game.

A PERFECT (BUT MEANINGLESS?) BRACKET

As we have seen, as AI continues to evolve, it may be able to improve upon many of the shortcomings of humans to better the chances of getting a perfect bracket. But there is one question we have yet to address: If AI predicted a perfect bracket, would that even be impressive? Part of the reason it is admirable when people do well on their March Madness predictions is because we recognize that we all have these human limitations. For a computer that can process tons of data and run millions of simulations, such predictions become less impressive, it seems. After all, who are you going to give the credit to?

So despite AI’s potential to predict brackets with more data-driven insights, even its correct predictions may not be that impressive. And at the end of the day, filling out a bracket is meant to be fun, not flawless. The chase for a perfect bracket will always be delightfully maddening … which is exactly why they call it March Madness.

We asked Perplexity AI to give us March Madness predictions. This is what it gave us: