Research

Writing an AI Playbook for Sports

In its first symposium, the Columbia-Dream Sports AI Innovation Center brought together a wide range of researchers and practitioners to chart the future of AI and sports

October 14, 2024
Grant Currin

Hundreds of researchers, students, industry professionals, and fans gathered in Morningside Heights Sept. 12 for a symposium of AI and sports, hosted by the Columbia-Dream Sports AI Innovation Center. This new industry-academic partnership aims to push boundaries at the intersection of AI, sports, and technology by advancing research and workforce development in sports technology.

“Our goal is to make sports better for everybody,” said Anantha Sundararajan, chief data officer at Dream Sports and co-director of the center, in his opening remarks.

In panel discussions, lightning talks, and a fireside chat, experts from sports and technology discussed how technology can improve many facets of athletics, from performance and officiating to media distribution and the overall fan experience.

“We are here to imagine the intersection of AI and sports, which we know is already deeply multidisciplinary,” said Shih-Fu Chang, dean of Columbia Engineering. “We have speakers today who are working on machine learning, data science, optimization algorithms, business, biology, economics, and many other fields.”

Performance and management

One key theme that emerged during the symposium was the power of data analytics to improve individual athletic performance. Jill McNitt-Gray, a professor in the Departments of Biological Sciences and Biomedical Engineering at the University of Southern California, described her work with U.S. track and field athletes and their coaches to turn data from practice facilities into actionable information that helps win medals.

“One of the things we’ve learned is that each athlete is unique — no one is average, so we need to think about each individual’s capabilities,” she said. “The challenge is figuring out how to use the data about the different ways it’s possible to win an event, match that data with the capability of each athlete, and then come up with the individual training plans,” she said. 

Mark Broadie, a professor at Columbia Business School and pioneer in the use of analytics in golf, explained why AI systems are necessary for conducting rigorous statistical analysis of team sports.

“When you have many players in motion, the data is different because you're tracking the continuous motion of 10, 11, or 12 players on each team,” he said. “The question becomes, What is the expectation of scoring a goal when the ball is in a certain place and these players are in these positions on the field and moving at these velocities?”

Answering that kind of question shows why AI is key to modern sports analytics. 

“You're looking at a large, large amount of data to build these expectation models,” Broadie said. “That’s where AI comes in.”

Image
Performance and Management
Christina Chase, managing director at MIT Sports Lab, speaks as part of the panel on fan engagement

Collecting the data

The importance of collecting and managing high-quality data was another important theme that emerged throughout the day. Speakers and audience members continually discussed instrumentation technologies that capture training and play in minute detail. Systems like Hawk-Eye, a computer vision and tracking system, are fundamental for developing models to understand what happened on the field — and for predicting what will happen next.

“I think the people who are going to be ahead in five years are the people who are really thinking about their data collection,” said Natalie Kupperman, an assistant professor of data science at the University of Virginia. “AI is going to be so crucial in the modeling side, but AI isn't going to help us as much on the instrumentation side yet, and so we need to be really thoughtful when we think about the cameras we're putting on fields.”

Kupperman, who consults for teams across several different sports, emphasized that camera-based technologies are poised to overtake wearable technologies in tracking individual performance, too.

“Most of those cameras are at the professional leagues, but they are vastly trickling down into collegiate sports and trickling down into high school and amateur sports,” she said. “Velocity-based training is often now using optical technologies” to track players’ movement using cameras rather than wearable devices.

Image
Project motivation
Vineet Goyal, associate professor of industrial engineering and operations research at Columbia, delivers a lightning talk: Dynamic State Dependent Catalog Optimization Approach for Contest Generation 

Fan engagement and experience

Another key theme at the symposium was the use of AI and other technologies in improving the fan experience. In a fireside talk with Columbia Engineering Computer Science Professor Vishal Misra, Krishna Bhagavathula, the NBA's chief technology officer, emphasized the importance of personalization in serving viewers clips and stats about their favorite teams. 

“We think of technology as a medium to actually take the incredibly beautiful game of basketball and showcase it across our fanbase,” he explained, noting that the goal is to engage fans on whatever platforms they’re already using. He said that the NBA’s digital transformation, from traditional television viewing to personalized mobile content, is allowing for an unprecedented level of interaction and customization.

AI has become a significant driver of this transformation, particularly through the NBA's League Pass service. This system employs AI to create personalized experiences for users, making content recommendations based on a growing number of datapoints.

“We rely on a combination of implicit and explicit data points to build content that's more personalized," Bhagavathula said, citing the example of fans who watch specific teams or players more frequently. He added, “We are creating that delightful moment for you...where you're like, wow, this app actually knows me.”

Beyond content personalization, AI has reshaped how fans engage with statistics and game data. Machine learning has enabled the league to generate advanced metrics that were previously unavailable, such as “hustle stats” that track intangible aspects of the game. In Bhagavathula’s view, data isn’t just useful for the teams — it’s also a valuable type of content for super fans who “obsess over every play” and use stats to “validate thier support of a player or team and win debates with their friends.”

Bhagavathula summed up much of the day — and sent a wave of laughter through the audience — with one observation: “There’s an extremely thin line between being a data geek and a sports fanatic.”


Photo Credit: Timothy Lee/Columbia Engineering

Stay up-to-date with the Columbia Engineering newsletter

*indicates required