Las Vegas Raiders head coach Pete Carroll watches as players stretch during the 2025 rookie mini camp at the Intermountain Health Performance Center on Friday, May 9, 2025, in Henderson. Bizuayehu Tesfaye Las Vegas Review-Journal/TNS
In 1968, Stanley Kubrick released his movie "2001: A Space Odyssey" and creeped out an entire country with the idea of a future controlled by artificial intelligence. In 2025, Zac Robinson is facing the idea of watching football and discussing strategy with a computer, and he is a little creeped out, too.
Robinson, the 38-year-old Atlanta Falcons offensive coordinator, worked as an analyst for Pro Football Focus before starting his coaching career in 2019, a stint that convinced him of the value and potential of advanced analytics. But there is a wide gulf between the math used to optimize fourth-down decisions and a voiced AI agent telling you to look out for the weakside linebacker while you're sitting alone in your office on a Tuesday night.
"I don't know," Robinson said, considering the scenario. "I'm a little scared."
He and other NFL coaches will have to get comfortable with that soon. Instead of Hal 9000, think of it as the Bill Walsh 3000, which could be assigned to watch the rotations of the secondary while a human coach focuses on the front seven.
"I'd have to see what that looks like," Robinson said, adding that he might get frustrated with a computer "barking" at him. "But if it ends up being a cool tool, that'll be interesting."
Ryan Paganetti got his job in part because of AI. He was hired by Las Vegas Raiders coach Pete Carroll in March as the team's "head coach research specialist," but the job may be better understood as the AI coordinator.
"I don't think when I was hired, the idea was, 'This is our AI guy,' but there is no doubt whatsoever that I am going to be using AI every single day," Paganetti said. "And probably in increasingly larger amounts every month that goes by."
In a league in which teams are constantly looking for an edge, the next big one won't be coming through the draft or free agency, Paganetti believes, but from artificial intelligence tools that are on the verge of transforming how coaches think about the game and do their jobs -- and maybe even which coaches still have those jobs in a decade.
"It almost might be a blockbuster moment where some coaches, their roles are replaced entirely," Paganetti said. "That's an issue in all sorts of industries where AI is just better and more accurate. I think that is going to happen with the football industry, to some degree.
"I feel pretty confident saying some team is going to win a Super Bowl in the next few years utilizing AI at a very high rate, significantly higher than it has ever been used before," he said. "It's really an opportunity to differentiate yourself from a team that might have a more talented roster or better coaches or whatnot. There is going to be more and more separation with teams that are bought in."
Carroll is fully bought in. The NFL's oldest head coach is perhaps its biggest believer in its youngest technology. "Everything you can think of is possible right now," the 73-year-old said.
Carroll's early adopter status isn't surprising, considering his history, which includes head coaching stints with the New York Jets, the New England Patriots, the Seattle Seahawks and the University of Southern California, where last year, he taught a class called "The Game of Life." As part of that class, Carroll spoke with author and new-age guru Deepak Chopra.
"He talked about AI giving him the opportunity to interview himself, talking to himself through AI so he was actually questioning his own person and being answered by his own person in return," Carroll said. "Some of it does feel like science fiction -- I get that -- but AI is around the corner for us."
Nearly three decades ago, IBM developed the supercomputer Deep Blue to face off against world chess champion Garry Kasparov. Kasparov won his first match against the machine in 1996, but Deep Blue won the rematch the next year, and humans have not provided a chess challenge to computers since. Computers have also since mastered the ancient Chinese board game Go, which involves exponentially more possible moves than chess.
Football presents a much tougher computer problem than chess or Go for myriad reasons, but many experts agree that some of the analytical functions done by human coaches could be done better, or at least more efficiently, by artificial intelligence, and the current rate of improvement in the industry suggests that moment might not be far away.
While the world ponders a future in which computers can generate their own decisions, the technology is still almost entirely machine learning and brute computing power rather than humanlike intelligence. "Think of machine learning as a technique for achieving artificial intelligence," said John Guttag, a professor at the Massachusetts Institute of Technology.
The large language models that power most AI and machine learning "don't know how to watch football yet, but I think with some work, they can be taught to watch football," said Udit Ranasaria, a senior researcher at SumerSports, one of a handful of companies developing AI tools with the potential to reshape professional football. "We can get to a place where we have something like ChatGPT that understands what's happening in the NFL."
It probably won't take long, Guttag said. He leads MIT's Computer Science and Artificial Intelligence Laboratory Data Driven Inference Group and has copresented papers about the uses of machine learning in the NBA and MLB.
"A big branch of artificial intelligence from almost the beginning has been computer vision, trying to get computers to see things and figure out what is in the image," Guttag said. But football is a more complex problem for computer vision than basketball, baseball or soccer because of the proximity of players to the line of scrimmage and the variance in personnel.
"Fourth-and-1 with Mike Vick and Alge Crumpler looks a lot different than fourth-and-1 with Kirk Cousins and Kyle Pitts," said Omar Ajmeri, CEO and co-founder of Slants, which uses machine learning to pull scouting information from football film.
Current AI is capable of "watching" game film from two teams, formulating a game plan and printing out call sheets for offensive and defensive coordinators, said Vishakh Sandwar, half of the winning team at this year's Big Data Bowl, which is sponsored by the NFL. "It's just a matter of the quality at this point," he said.
The winning project from Sandwar and his fellow New York University alumnus Smit Bajaj created an algorithm that can identify coverages based on the computer's "visual" analysis of defenders. The model, which used technology developed by SumerSports, achieved an accuracy level of 89% based only on pre-snap alignments. It adjusts in real time as defensive players move and can identify which ones are the worst offenders in giving away coverages before the snap. It also allows coaches to create custom looks by moving defenders on a digital whiteboard.
AI "is very good at piecing together relationships in very, very high-dimensional spaces," Bajaj said. "With languages, it's able to piece together and understand that based on the entire history of the internet, this is the word that is likely to come next."
Over time, Guttag said, the technology will continue to get better.
"And what you'll do is say, 'Here are all the series that led to first downs. Here are all the series that didn't lead to first downs. What are the important differences?' -- without hypothesizing before," he said. "You'll just let the AI machine learning look at all that data and say, 'Here are some interesting differences.' One of the great things about machine learning is it finds things you didn't know were there."
An AI agent could assist in play-calling during games, but NFL rules ban that sort of assistance from kickoff until the clock hits zero. During the week, everything in the AI realm is in bounds, although the league continues to monitor developments, at least the ones it knows about.
"There's still an extreme level of secrecy," Paganetti said. "Even people who work in analytics have very little idea what people working in analytics for other teams do sometimes because it's considered company secrets.
"We know what the scouts do on the other team: They scout. We know what the coaches do on the other teams: They coach. But when it comes to the actual contribution of the analytics department of another team, it's really open-ended."
Tennessee Titans coach Brian Callahan believes AI acceptance around the league will vary.
"Anytime you are talking to a football coach who has done one thing for a long time, it takes time for that to take hold, but I do think there is a much more open mind to all of those things: data, analytics, new processes," Callahan, 41, said. "Yeah, there will be some pushback in some spots, but there are a lot of other spots where guys will look at it as something that can really help."
Guttag is less optimistic about buy-in, pointing out the resistance coaches showed to accepting the math behind fourth-down decision-making, perhaps the most rudimentary form of machine learning introduced to the game.
"Anyone who knew any math at all knew they were behaving stupidly, and yet they continued to do it," he said. "It's kind of remarkable."
The next wave of AI will make the fourth-down bot look like an abacus.
The NFL is using an AI application called Digital Athlete to help teams predict injuries, but the upcoming coaching applications are where NFL fans are most likely to see results.
"With things like 'What play should you run against this look? What blitz should you run against this alignment?' -- those are areas where AI can really move the needle or come up with ideas that you might otherwise never have thought of," Paganetti said.
Teams that are using some sort of AI in their weekly preparation are likely using it now only at the most basic level. Carroll, at least, plans to be on the cutting edge soon.
"It's just such a wide-open domain to kind of figure things out and do things new, take advantage and utilize everything you can think of," Carroll said. "That's something I like, man. If you're not curious, you're not growing. The last thing I'm going to do is ignore AI."
This article originally appeared in The New York Times.
Copyright 2025