March Madness and the AI Experiment

This is the final installment in a series documenting the use of Artificial Intelligence to aid in NCAA tournament bracket predictions. While the ultimate goal of winning a March Madness pool wasn’t achieved, the experiment proved valuable in refining the bracket-building process.

Improved Process, Not Certainty

The biggest takeaway from this exercise wasn’t increased accuracy, but rather an improved approach to bracketology. AI helped facilitate better thinking, even if it couldn’t eliminate the inherent unpredictability of the tournament. Despite not winning, the author’s brackets performed well, landing in the top two in one pool and top six in another out of 65 entries.

Early Success and the Inevitable Chaos

Initial success saw 13 of the Sweet 16 teams correctly predicted. However, the characteristic chaos of March Madness soon unfolded. A projected Duke vs. Florida semifinal matchup, with Duke ultimately winning, didn’t materialize, ending championship hopes. Despite this, brackets were in the 98th percentile of ESPN’s 26 million entries going into the Elite Eight.

Key Lessons Learned for Next Year

Weighting Late-Season Momentum

One crucial lesson was underestimating the importance of teams peaking at the right time. Teams like Arizona and Michigan, strong contenders from a competitive conference, exceeded expectations, and their late-season form wasn’t adequately considered. Next year, greater emphasis will be placed on current performance in March.

The Value of Coaching

The experiment also revealed an underappreciation for the impact of coaching. While players execute on the court, coaches are vital in tournament play, particularly in single-elimination formats. Dan Hurley’s success and John Scheyer’s struggles highlighted this point. Future bracket building will prioritize evaluating coaching prowess.

Acknowledging Inherent Volatility

Forecasting, even with AI assistance, excels at identifying patterns but struggles with predicting specific outcomes. College basketball involves unpredictable teenagers, prone to fluctuations in performance. The one-and-done format amplifies the potential for upsets, a reality AI cannot overcome.

AI's Broader Applications Beyond Bracketology

Bracket pools present a uniquely unforgiving test, demanding precise predictions. AI’s strength lies in narrowing down options to a strong set, rather than guaranteeing the single perfect choice. This principle extends to real-world applications like investing, planning, and research, where AI can identify promising paths and mitigate risks.

Real-World Example: Travel Planning

A recent trip to Lima, Peru, demonstrated AI’s utility in simplifying complex decisions. ChatGPT helped identify attractions and restaurants, resulting in a satisfying experience without the anxiety of missing out. The quality of AI’s suggestions is directly linked to the clarity of the user’s input.

Looking Ahead: A Refined Approach

Next year’s approach will incorporate increased weighting of late-season momentum, coaching effectiveness, and the inherent volatility of the tournament. The goal isn’t to eliminate uncertainty, but to leverage AI for more systematic evaluation and intelligent decision-making. The experiment proved worthwhile, even without a championship victory.