Episode Overview

Podcast: Pablo Torre Finds Out
Episode Title: How Artificial Intelligence Is Already Changing Sports
Guests: Daryl Morey (President of Basketball Operations, Philadelphia 76ers), Sendhil Mullainathan (MIT professor, behavioral economist, MacArthur "Genius" awardee)
Date: March 13, 2025

This episode, recorded live at the influential Sloan Sports Analytics Conference, dives into the transformative impact of artificial intelligence (AI)—specifically large language models (LLMs) and machine learning—on sports. Host Pablo Torre guides a sharp, often humorous discussion with Daryl Morey and Sendhil Mullainathan, aiming to separate hype from reality and probe how AI is reshaping everything from team strategy to front office decision-making and even fan engagement.

Key Discussion Points & Insights

1. The State of AI in Sports Decision-Making

• Daryl Morey shares that AI and machine learning have long been deployed in the NBA front office:

  • Used for forecasting draft picks, player performance, and computer vision analysis (e.g., identifying pick-and-roll plays) ([05:11]-[07:20]).
  • LLMs currently assist in coding, documentation, and minor predictive analytics but remain only one "vote" among many in crucial decisions ([07:20]-[09:04]).
  • “We’ll treat [AI models] almost like one scout. Now over time, if it’s shown that they have a better track record than scouts, we might weight the LLMs…as one or two votes out of the process.” — Daryl ([08:03]).

• Pablo summarizes: “I would just like NBA Central to aggregate what Daryl just said, which is that he asks ChatGPT for help on trades.” ([09:04])

  • Daryl groans at the media fodder: “Please, not aggregate.” ([09:13])

2. What Is the True Frontier for AI? Beyond AGI

• Sendhil Mullainathan challenges the fixation on AGI (Artificial General Intelligence):

  • "[Artificial general intelligence] sounds unambitious... We already have people who can think like people. What we need is algorithms that can do things people could never do.” ([09:44]-[10:09])
  • Tells the story of The Wizard of Oz authorship, where an algorithm definitively identified which chapters were written by which author—something literary critics failed to agree on. Shows that AI can reveal patterns humans don’t see ([11:10]-[12:15]).

• Key Insight: The real promise lies in algorithms revealing “signals” that humans miss and in self-scouting—models that analyze the decision-makers themselves ([13:45]).

3. Human-AI Collaboration & The Limits of Algorithms

• Discussion of “Centaur Chess” (human + computer teams) versus fully automated systems:

  • Initially, humans aided chess AIs, but eventually, pure machines outperformed any combination ([14:22]-[14:57]).
  • Sendhil: “We will never beat an algorithm in a closed world problem. Most phenomena in the world are not really closed world problems, they're open world problems.” ([15:17])
  • Pablo reframes: “If we're playing Centaur chess, the computer doesn't know we can still flip the board.” ([16:50])

• Entertainment vs. Efficiency:

  • Pablo asks about balancing competitive advantage, entertainment, and fan engagement — “Is there a way... to align everybody on what the best outcome is, which involves entertainment?” ([17:28]-[18:22])
  • Sendhil shares that “storylines” and the emotional experience of sports are “part of the objective function”, not just wins ([19:12]-[20:32]).

4. The Emotional Economics of Sports

• Daryl recounts fans’ reactions to playoff outcomes, noting that entertainment value, not just wins, shapes the mood ([21:00]-[22:41]):
  • “The way we measure it was objectively better... and that surprised me.”
• Sendhil says negative emotions are vital for sports culture, “The thing that is dangerous is disinterest... Anger is good. You got something to talk about.” ([22:39]-[23:51])

5. Parity and Automation in Sports Analytics

• As LLM (AI) technology becomes widely available, Pablo wonders if Daryl’s unique value is eroding ([24:06]-[25:34]).
• Sendhil: “Some experts absolutely should be worried... (but) the genius of someone like Daryl... are people who can take an ill-formed problem and actually formulate it in a way that it becomes tangible and actionable.” ([25:34]-[27:44])
• The bottleneck is shifting from analysis to problem formulation.

6. Strengths and Flaws of LLMs

• LLMs can now analyze the NBA’s collective bargaining agreement and solve complex legal/data questions at a near-expert level ([28:33]-[29:28]), but:
  • Error rates and inscrutable mistakes remain; “It’s not just their error rate, but the inscrutability of their errors.” — Sendhil ([29:28])
  • Daryl’s “Monty Hall problem” as an evolving test for LLM sophistication ([30:00]-[31:03]).

7. Self-Scouting & AI for Decision-Makers

• Sendhil proposes “turning the camera” on the executives themselves, not just the athletes ([40:01]-[40:30]):

  • Summarizes research finding that AI models predicting judges’ bail decisions outperformed the actual judges themselves ([40:39]-[42:09]).
  • Key variable influencing judge decisions? The defendant’s mugshot, with factors like “full or fat face” and being “well-groomed” dramatically affecting outcomes ([43:05]-[44:09]).

• Daryl notes the “bad idea whack-a-mole” aspect of executive work, where narrative-based ideas must be constantly scrutinized and checked ([33:58]-[35:47]).

8. The Importance of Data over Analysts

• Both agree the next arms race is in data collection more than analytical prowess:
  • “I wrote a paper years ago on ‘you need better data, not better analysts.’ The data is going to be the unique thing, not the analysts.” — Daryl ([38:53])

Notable Quotes & Memorable Moments

• AI as a Decision Tool:
  • “We use models as a vote in any decision... over time, if it’s shown that they have a better track record than scouts, we might weight the LLMs or use multiple LLMs as one or two votes.” — Daryl Morey [08:03]
• On AGI:
  • “AGI...is a shockingly unambitious goal.” — Sendhil Mullainathan [09:44]
• Algorithms See What Humans Miss:
  • “Algorithms are fantastic at picking up signal on the kind of things we didn’t even know to notice for.” — Sendhil [11:38]
• “We will never beat an algorithm in a closed world problem. ... But most phenomena in the world are open world problems.” — Sendhil [15:17]
• Entertainment over Wins:
  • “The thing that is dangerous is disinterest. ... Negative emotions are fine. The worst thing is disinterest.” — Sendhil [22:43]
• Problem Formulation as the Expert Edge:
  • “The genius ... are people who can take an ill-formed problem and formulate it in a way that it becomes tangible and actionable.” — Sendhil [27:34]
• LLMs & Their Flaws:
  • “It’s not just their error rate, but the inscrutability of their errors.” — Sendhil [29:39]
  • “That’s probably true of most humans too.” — Daryl (on the shallowness masked by language fluency) [32:56]
• Self-Scouting:
  • “We have a self scout...we flip that on its head and self scout and say, when they play us, what are they likely saying about us?” — Daryl [12:47]
  • “A decision aid like that could help me understand myself.” — Sendhil [44:20]
• The Data Race:
  • “I wrote a paper years ago on you need better data, not better analysts. ... the data is going to be the unique thing, not the analysts.” — Daryl [38:53]

Important Timestamps

• [03:00-04:02]: Introduction of Guests & Their Credentials
• [05:11-09:04]: Daryl Morey explains the current use of AI in NBA operations
• [09:44-12:15]: Sendhil explains why AGI is “unambitious”; Wizard of Oz authorship story as analogy
• [13:45-17:25]: Self-scouting, closed vs. open world problems, “Centaur Chess” analogy
• [18:22-23:51]: Entertainment vs. efficiency in sports; the role of storylines and fan emotions
• [24:06-27:44]: Parity in data access, will AI erase executive edge?
• [28:33-32:34]: LLMs, legal analysis, error problems, the Monty Hall test and LLM limitations
• [33:58-35:47]: “Bad idea whack-a-mole”—the creative and critical tasks of a GM
• [36:15-39:12]: Supervised learning explained; importance of novel data collection
• [40:01-44:44]: Turning AI on decision-makers; lessons from criminal justice research
• [45:12-46:57]: Final takeaways from each speaker

Tone & Style

The conversation flows with wit and self-awareness—Pablo keeps a breezy, searching tone (“Can I just grossly aggregate what you said…”), Daryl is characteristically dry and practical, Sendhil is both erudite and playful (admitting “genuine sicko behavior” as a message board moderator). The episode’s intellectual depth is leavened by NBA fandom and honest humor about their status as self-proclaimed “nerds.”

Key Takeaways

• AI is already deeply integrated into pro sports, especially for forecasting—but it’s one tool among many and not yet a silver bullet.
• The “edge” will belong not to the best analysts, but to those who ask the sharpest questions and collect the best new data, especially in open, ill-defined domains.
• Self-reflection (and even self-scouting via algorithms) and appreciation for the emotional, narrative elements of sports will remain vital—even as AI advances.
• Expertise is shifting from pure analysis to creative, robust problem and data formulation.
• As AI’s technical ‘magic’ becomes widespread, the new frontier is understanding its limitations, its hidden biases/errors, and integrating it in ways that enhance (not replace) uniquely human roles in sports and society.