Topics
Artificial Intelligence and Business Strategy
In collaboration with
BCG
In collaboration with
Jeff Miller moved from a career in government to focus on the health and safety of NFL players. Due to the nature of their sport, they are more prone to injuries, including concussions, than other professional athletes. By leveraging the game’s unique elements, such as the volume of cameras filming on-field activities, the NFL can gather large data sets to track injuries, and then segment their likelihood of occurrence by position and design equipment and update game rules to minimize them.
Jeff joins the Me, Myself, and AI podcast to describe how the NFL has partnered with tech companies, including Amazon Web Services (AWS), to expand its technology fluency and implement these solutions. Listen to today’s episode to learn about specific equipment the NFL has produced and rule changes it has implemented to better track and protect its players, as well as one element of the game Jeff believes technology cannot replace.
Subscribe to Me, Myself, and AI on Apple Podcasts or Spotify.
Transcript
Sam Ransbotham: How does one major sporting league use AI to minimize player injuries? Find out on today’s episode.
Jeff Miller: I’m Jeff Miller from the National Football League, and you’re listening to Me, Myself, and AI.
Sam Ransbotham: Welcome to Me, Myself, and AI, a podcast on artificial intelligence in business. Each episode, we introduce you to someone innovating with AI. I’m Sam Ransbotham, professor of analytics at Boston College. I’m also the AI and business strategy guest editor at MIT Sloan Management Review.
Shervin Khodabandeh: And I’m Shervin Khodabandeh, senior partner with BCG and one of the leaders of our AI business. Together, MIT SMR and BCG have been researching and publishing on AI since 2017, interviewing hundreds of practitioners and surveying thousands of companies on what it takes to build and to deploy and scale AI capabilities, and really transform the way organizations operate.
Sam Ransbotham: Hey, everyone. I’m kicking things off today with a discussion with Jeff Miller, executive vice president of communications, public affairs, and policy at the National Football League. Shervin, what do you think about that? “Kicking things off.” Did you catch that? Pretty good, huh?
I do these things for you, Shervin. You have to appreciate them.
Shervin Khodabandeh: How long did it take you to write that?
Sam Ransbotham: Oh, it was about half an hour. Anyway, Jeff is also responsible for player health and safety, and we’ll learn … about how that relates to AI. Jeff, it’s great to have you with us. Thanks for joining us.
Jeff Miller: Thank you for having me. I’ve been looking forward to this.
Sam Ransbotham: For those of us in the U.S., we’re hearing this episode just before the Super Bowl this weekend. We’re eager to learn how the capabilities of artificial intelligence have changed the game of football. But first, let’s start [with some things that are] more basic. Can you tell us, Jeff, about your role and how you came to focus on player health and safety?
Jeff Miller: Sure, [I’m] happy to. I’ve been at the NFL now for about 15 years, and I think I’ve had about 15 jobs during the course of that time, each one evolving on the last.
I started working in Washington, D.C., coming off of Capitol Hill, and one of the primary topics that the government was interested in during my tenure in Washington on behalf of the NFL was health and safety issues, long-term effects of concussions, what the league was doing to intervene, what the league was doing with retired players, what the league was doing with any number of different sorts of injuries.
With that as a primary topic, we ended up building out a portfolio with partners and others around health and safety, and looked for a handful of different strategies to affect the safety of the sport. And since I was involved on the policy level at that point in Washington, it became my portfolio. Over time, that portfolio grew and changed and became more innovative and sophisticated with a number of great people to help do that work. My job also began to evolve as I began to communicate around the health and safety of the sport, and then eventually added the communications portfolio and some other odds and ends to my work.
That’s 15 years in 15 seconds.
Sam Ransbotham: That’s great. Tell us what kinds of initiatives you’re working on. I think details would be fun here.
Jeff Miller: On the health and safety side, the primary thesis of our work is to try to find ways to reduce injuries. Seems pretty simple. NFL football and football at all levels is a contact sport, and unfortunately, one of the effects of that is the possibility of injury, which is true in all sports, but of course, in football, tackling and blocking can lead to injurious behavior. So can high-speed running and the amount of demands that we put on the human body.
Work that began several years ago around trying to reduce concussions, specifically, has morphed into innovative strategies to try to reduce any sort of injury. So everything that you can think of, from hamstring strains — which bedevil our players in greater numbers than any other injury — to shoulder strains, to high ankle sprains, all of those have component parts to them and strategies against them where we rely upon experts in the spaces, researchers, biomechanical engineers, to help us understand how the human body works; innovators to come up with advanced ideas around equipment; and of course, trainers and doctors to try to intervene; and, should an injury occur, we have [ways] to get them back sooner.
We share research, we publish research, we work with first-class institutions around the world and try to share our learnings in as transparent a way as we can, because in any one of those spaces, any injury saved is a benefit to the game. And more importantly, any injury saved in bulk, by creating better equipment, changing the rules of the game, has a huge benefit to a player in the short term and his long-term health. It’s become a big priority for the league. With some of the more advanced tools that we have now in understanding how the human body works and how our game is played, we’ve had the ability to intervene in numerous and really interesting ways.
Shervin Khodabandeh: You’ve alluded a few times, Jeff, to advanced tools, and I would assume a fair amount of that [are] tools around computer vision and computation and AI and that kind of stuff. Maybe comment a bit on that front?
Jeff Miller: Concussions, of course, have been a long-term concern for the league, and to understand that, the first thing we did when we sat down with a group of biomechanical engineers and doctors and trainers was to ask ourselves a question: “OK, well, what’s causing these?” You can look at, given the fact that we’ve got a couple of dozen cameras in every stadium on every game, more so than probably any other sport, you can watch the injuries happen in real time. I can’t think of another place in the world outside of the world of sports where you’re going to get the benefit of understanding how a human being reacts to force. And that’s in essence what we were looking at.
We would literally sit in a room and look at all of the different injurious events and the causes of them, and try to bin them — by position, by play type, maybe by behaviors, certain blocks, or certain reactions that players had. And try to understand them better if we put them into different themes or bins. We would come back to our rulemaking committee and say, “Hey, this is what we see” and hope that they would make changes. As we were going through that process, we were acting as the early version of a computer there that could potentially bin them for us.
All players wear sensors — GPS trackers in essence — in their shoulder pads. That’s how you get some of the advanced analytics that you see either in the game or that we look at for research purposes [afterward]. Or we created our own bespoke instrumented mouth guard to understand the forces on the head. You could put sensors in a helmet, but helmets move around, so to get a real specific understanding of what the brain sees, we had a number of players wearing instrumented mouth guards, which had terrific advanced technologies. We understood any number of other elements about players’ positioning, velocities, etc. We were able to collect all of this information. And instead of sitting there — me as a history major and a bunch of engineers who know a lot more about science than I do — some of us came up with this great idea that we should go out to Silicon Valley and share this data set.
Maybe three or four years into this work — so probably four or five years ago — we went out and sat down with some of the tech giants and said, “Here’s what we have.” We had reconstructed against 150 different variables at that point, about 800 or 900 different on-field concussions dating back several seasons with all the video that we had. [Now] increasingly with some of the other technologies and sensors that I mentioned, we’re talking another 500 or 600, 700 concussions since. [With the information] broken down, these companies said, “Well, what you have is this incredible data set. What you need to do is get yourself organized, and we can help you do that.”
Ultimately, we chose to work with AWS on this. And they said, “Well, let’s help you organize all of this data that you have, and let’s stop sitting in a room watching concussions” — and any other injury, by the way, because we did the same thing for high ankle sprains — “and come up with ways that we can create models that can then understand for you where the risk elements are.” That was my introduction to computer vision and artificial intelligence.
Our models have gotten more sophisticated by the year, and I think that the league has benefited tremendously. Not only do we understand more about the injury but we’ve changed rules. The equipment has become vastly better as a result of understanding how these injuries occurred — not to mention the teaching and the coaching and the other elements that go into changing how players behave, which has led to real-world results in a very short time around injuries.
Shervin Khodabandeh: This is quite illuminating because people hear about these rule changes and policy changes, but they don’t really know the richness and the analysis and all of that data and information behind it. Thanks for unraveling that for us.
Jeff Miller: A lot of the data that I just mentioned we collected included how frequently a player at a specific position is hit in a particular place on the helmet. Not every player on the field experiences the game in the same way. If a quarterback, for example, is to suffer a concussion, the vast majority of those occur when he is tackled, sacked, and he’s holding onto the ball because he has to, of course. He doesn’t have an opportunity to brace himself. So he ends up getting hit and he doesn’t brace himself with his arm or anything else, and the back of his head ends up hitting the turf.
The majority of concussions of quarterbacks occur in that manner. Other players, like an offensive lineman, never get hurt that way. They have repetitive head contact but primarily at the crown of the forehead. Yet their equipment is exactly the same. It doesn’t make any sense. And a wide receiver, if he’s to get a concussion, it’s usually a very high-velocity impact because he’s out in open space and he achieves incredibly high speeds, as do the defenders. So when they hit him or he hits his head, those things come at very high velocities, much higher velocities than an offensive lineman, who is dealing with more head contact but it’s at less force. Yet they’re all wearing the same equipment.
So taking all of that data, as we better organized it and measured it, [we went] to the helmet manufacturers and [said], “Help us build helmets that protect players where they get hurt.” And to develop laboratory testing, which our engineers did, [we said,] “OK, well, we know what happens on field when guys get hurt, so why don’t we test helmets against those conditions for when they get hurt?” And then we can tell players which helmet will do a better job of protecting them because the lab testing should show that. What’s remarkable, as we stay on equipment for a moment, the injury rate on field and the testing of the helmets in the laboratory have a remarkably consistent correlation: a better test in the lab, the fewer injuries on field.
Now on other injury types, take a lower extremity injury and ankle sprain — a very common injury — or a high ankle sprain, which is really sort of a nasty injury that causes a lot of missed time. Now you’re talking about the interaction between a surface and a cleat. So we have to understand those and the forces there, and that takes a different set of things because you can’t really out-rule cleats or surfaces. Those aren’t things that the Competition Committee can take a look at. But, instead, you can build better ones and you can train better, and you can think of [different] ways to intervene. So the different injury types have different strategies and different levers that you would pull against them.
Sam Ransbotham: What’s the role of training here? What you’ve talked about is on-field injuries and, you know, things that happen within the game. But it seems like there’s also a role for these technologies in helping improve, let’s say, exercise regimens. Maybe we could avoid these injuries in the first place, not so much by equipment but by better, I don’t know, hamstring exercises, for example. Is any of that going on?
Jeff Miller: All of that is going on. The football schedule is odd because unlike other sports like baseball or basketball where they play regularly throughout the week — three games, four games, baseball more — we play once, right? So there’s a lot of time in between, which is usually for rehab. Instead, to get players ready as opposed to recovering from the games and the contacts, we spend a lot of time on field getting ready for the regular season in August with an intensity of practices. And, in fact, players show up for training camp and jump right into it, or did three or four years ago.
And we know that because, believe it or not, in the first two weeks of training camp, meaning the last couple days of July and the first week or 10 days of August, we’d have more injuries in that period than the rest of the season combined. And those are where soft tissue injuries would occur. You’d have the hamstring strains and the adductor and groin pulls and all of that sort of stuff. And we knew that not only because of the injury numbers that we collected but also because we’d have sensors on players, and we could measure their loads. So we would understand looking at the data when we would aggregate and be like, “Well, this doesn’t make a lot of sense.”
A lot of players are in better shape year-round now than they were 50 years ago when football was a part-time job and they were out doing other things during the rest of the year. Now, they’re coming to training camp in shape but still not necessarily in the way they need to be for the first regular season game, which comes six weeks later. So we’ve changed the beginning of that training camp period with the data that we’ve collected and the analytics that we’ve done to create mandatory limits on a day-by-day basis of how much players can be on the field for practice. [We’re] increasingly sharing tools with the clubs so they can manage the loads of the players themselves and look at the players, not just on a team basis or on a position basis but on an individual basis. Is this player running hot in terms of the number of loads compared to other positions in the group [and] compared to the league as a whole? Do you want to pull back on successive days?
We mandate off days. We limit, in the first couple of weeks, how much time [they can spend] on field, on an acclimating basis, so it increases. And then when we put pads on, because we keep them without their shoulder pads and helmets the first few days, we rerack that and start all over again. So they get used to playing with the pads, which add weight and add a different element of force to the game when they start tackling each other, and we re-ramp them. Believe it or not, the first year we saw that, I think, we had a decrease of around 30% lower extremity injuries in year one that carried through into the regular season. And we’ve seen that [be] consistent in successive years.
[These are] huge savings on the hamstring strains for anybody who’s ever had one. You know, they don’t go away, there’s nothing you can do for it, just rest. A player wants to play. So there’s always the chance of recurrence, and those sorts of things happen. By simply analyzing the data and taking a look at the first couple of weeks of the preseason, we decreased injury rates substantially across the league.
Shervin Khodabandeh: Jeff, the question I have is, these initiatives have been going on for quite some time. Recently, you brought in better instrumentation, AI, more technology. Did you see a step-function sort of improvement, or was it gradual how that introduction of AI changed the trajectory of your speed to insight, how quickly new information and new insights were coming up? What was your experience?
Jeff Miller: In football, I think we’re in an incredibly innovative time and a bit of a culture shift as it relates to engagement with some of these tools, AI specifically, around the health and safety of the game and relatedly around changing rules and behaviors that lead to greater injury types.
My experience has been that the culture of the league requires us to explain to coaches and players why we want to make a recommendation for something. So if I can show them, as an example, that the speeds on the contacts on the kickoff play are 20% slower, and therefore some of the forces are lessened when there are blocks and tackles on the new kickoff as compared to the last one, and therefore you should vote for this play because you’re going to see a lower number of injuries, that gets the sort of attention and a level of intensity around the conversation and seriousness around our rulemaking body much more so than if I came to them with a half-baked idea that was based on a theory as opposed to a proof point.
Likewise, I mentioned the helmet innovation earlier based on all the data points. When we go to a player or an equipment manager and a trainer goes to a player, and we say, “This helmet, based on all this explanation that I just offered you, is going to make you 20% safer than this other one,” the adoption changes. It just does. We visit with all 32 clubs on an annual basis around their injury rates and such — and I’ll get back to some of the tools that we’ve created that we share with them so they can do this work on their own — but we showed them their injury rates in the preseason and how those numbers pulled into the regular season and then demonstrated for them how to cut those rates down. [We] showed them that it’s something like three of the past five Super Bowl winners have had injury rates in the preseason among the lowest third in the league. That gets a coach’s attention.
So we’re telling you to do this, and we’re showing you that it works, and then you transition to it. Sports science and other things at the clubs play a role in that, too, because they’re getting smarter and better at that. Therefore, there are great two-way conversations, but we have the benefit of being able to aggregate information from 32 teams. They can only keep one. So their ability to receive this and give us feedback on it has made us all better. But I think it’s been over just the last few years that the speed with which people have used these newer tools and the applications of them have really picked up pace.
Shervin Khodabandeh: What’s the next frontier? Is it keep doing, do more, accelerate, or is there sort of a next big bang, big frontier in this space?
Jeff Miller: It’s sort of both hands. One is take the advances we’ve made in certain equipment like the helmet and translate that to surfaces. We created finite element models of surfaces, the synthetic surfaces. [We] want to make them play with less traction, and, therefore, hopefully, fewer injuries. [We] do the same thing with cleats. As you innovate in those spaces and some of our partners, shoulder pads, and other things, too. It’s an application of a path that we’ve created … and, obviously, [there are] continued rules changes as we see areas where there are higher risks, like the kickoff play and others. That’s one.
Path two — I think this is probably the more important one in terms of having [an] immediate effect — is to put the tools that we’ve created into the hands of the people who use them on a day-to-day basis, which are the clubs, right?
The clubs have substantial and sophisticated athletic training and sports science staff. And so we’ve tried to create a tool, which we call the Digital Athlete, in conjunction with our friends at AWS, that puts all of what we know in the hands of the people who can use it. So they log in to the Digital Athlete application, and they can take a look at a player, his loads, compare them across the league, look at their injury rates, compare those across the league, take a look at things, increasingly, like helmet detection. There’s a model there that picks up through computer vision every time somebody hits their helmet. And you can take a look: Does that have a correlation with eventual injury if you’re an offensive lineman or any other player?
We’re working on pose, which should help us understand human movement better. And we’re getting pretty close to being able to say to an athletic trainer in a club, “Hey, look, this player’s rehabbing from an ankle sprain, but his pose when he runs now isn’t what it was before or it isn’t in best practice.” You can toss in functional movement screens into that as we try to integrate more and more of those data points to better understand human movement. It’ll be an opportunity to understand whether a player has recovered from injury but also probably an opportunity at some point to improve human performance, to make sure that he is doing all the things you would hope that he would do on the preventing injury front, as well as perform better. To have those in the hands of the people who see the athletes day-to-day and work with them is so much more valuable than me having access to it. That doesn’t get the job done. I can do things at a larger level, maybe strategically, but operationally, that’s where we’re going to see the biggest advances — people owning it themselves.
Sam Ransbotham: One of the things I think about here is that you’ve mentioned, for example, equipment changes, and I like that. But as I think back on maybe what happened — I’m not sure if this has truly happened or if it’s lore — but when people started wearing seat belts more often, they started to drive riskier because they had the seat belt to protect them. So when you’re saying that, for example, padding increases or cleats increase or helmet changes, suddenly it seems like people might be pushing more risk and willing to go harder and willing to ride that line harder.
Shervin Khodabandeh: Sam, are you suggesting players aren’t giving their 100% already?
Sam Ransbotham: Oh no, I’m not saying that they’re not giving a hundred percent.
Jeff Miller: I’m more interested in how Sam drives.
Sam Ransbotham: Shervin’s the bad driver here. I think that’s pretty established from past episodes. I do worry that there’s so much, I think, incentive to make the great play and to run faster and to push harder. What keeps people from pushing and using up all that buffer you’re creating with equipment and using it up in different areas?
Jeff Miller: It’s such an important question. And one that we’ve wrestled with a great deal. I’ll give you one case study. In addition to better helmets, we put in a device called the Guardian Cap, which is a soft exterior device.
Sam Ransbotham: We’ve seen those.
Jeff Miller: It’s been useful for us. You see them a fair amount in the high school level. We shared with the company some of the same data I talked about earlier with the sorts of forces that we see at the NFL level. And they created one that dealt with NFL-type forces, [where] obviously the speeds are faster and the men are larger than you would see in the high school space. So it was a new device, and we mandated it for different player positions increasingly over the last three years. And we’ve seen about a 50% reduction in concussions in the preseason practices for the players who have been mandated to wear it.
And we’ve seen no ill effects, but, behaviorally, we were worried about what you mentioned, which is when you take the thing off or even with the thing on, have you changed somebody’s behavior? And we didn’t see an increase in concussions when we took them off for games or when the regular season started, in those player positions. There’s probably a little bit of an effect on their behavior because there’s another, you know, quarter inch or half inch they’re not used to dealing with when they make impact. But we didn’t see the forces increase when we were able to measure them in practices, and we didn’t see the injuries increase when we took them off. But it continues to be an issue that we would be concerned about.
One way we measure as players have moved into better helmets is we still look at the speed: Has their behavior changed in terms of the high-speed helmet collisions, the kind that cause injury? And we haven’t seen them go up. I think they need to go down more than they have, but we haven’t seen players change behavior as it relates to that.
The notion of invincibility, which is something we talk to players about all the time, that’s true for any 22-year-old, right? But it’s certainly true if you’re 6'4" and 250 pounds and can run like the wind, that you are invincible. That’s always been a challenge, because there are other players who are willing to go make the play. And that’s, for me, where we get into officiating and making sure that the rules are enforced and, where necessary, players are penalized or teams are penalized for acting in ways that are, while aggressive, also indicative of injurious type behavior that needs to be ruled out.
Sam Ransbotham: One of the things I think you’re also getting at is this short term [versus] long term. You know, one of the things I think you’re thinking about is perhaps a longer-term view of the sport in general and the health of the sport when what you’re saying [is], “Oh, I’ll trade a concussion,” which may be a 30-year problem for a person, versus an ACL [injury], which may be a one-year problem for a person. It seems like you’ve got a lot of data to try to get on top of that, and it may take time for that data to flow through, but you’re at least collecting it now.
Jeff Miller: You know what you can’t simulate? You can’t simulate the creativity of a coach to try to figure their way around the rules. We modeled this. Some other leagues have done versions of what we’re doing on the kickoff play. Just by way of background, the kickoff play had the highest injury rate of any play year over year in the highest rate of concussions. It should come as no surprise because the players reach their highest velocity and, therefore, the impact speeds and the closing speeds when they contact each other, for a block or for a tackle, are at their highest rate.
And it’s bedeviled the league forever since we’ve been able to collect the data. You had roughly double the injury rate — and major injuries, ACL injuries, the kinds of things that you lose players for a year — and concussions up to four times, depending on the year, are runner pass plays.
Again, it’s not a shock when you look at the dynamic of that play. So, what do you do in the abstract? Well, you try to decrease the speed and the space. We had a sense as to what the speeds would be. And we had to guess at what the formations would be and what the strategies would be. So we would share our modeling with the coaches — in this case, the special teams coaches, the coaches who coach the kickoff and the punt, field goals, and extra points. And they come back to us with a thousand ideas you couldn’t imagine, unless you know the game so intimately.
They’re like, “Well, the team’s just going to do this. They’re just going to kick the ball this way to that guy, and the whole play’s going to look ridiculous. Or they’re just going to double-team this guy, and this guy’s going to swing around and do that, and you know, you’ll have five touchdowns a game.” Like, “OK, well, obviously we can’t do that.”
We’ve learned a lot this year already. The kickoff return rate, which is part of what we were trying to do, is get the kickoffs back in the game. [At last year’s] Super Bowl, there were 13 balls kicked off — halftime, beginning of the game, and then after scores. Not a single one was returned. It was a dead play.
That’s useless. Everybody was frustrated by that. At the same time, when the balls were returned, you were looking at a two to four times injury rate. So nobody was rooting for the ball to be returned, necessarily. So everybody was ready to do something different. And now I think our return rate is up about 65% from a year ago. It still could use a little more goosing to get it higher. The injury rates are down to what they were on a runner pass play pretty much, pretty close. So the modeling on the return rate is about right. The modeling on the injury rate is really close to what it would be because we were able to model out what the speeds would be, starting where the players would start because you know how fast they can get moving once you know what the distance is between them from a stand. But when coaches have tried different tactics, and I think we’ll see more of it over the intervening time, that is going to outsmart any model that has no idea what the input is.
Sam Ransbotham: People are creative.
Jeff Miller: Thank goodness.
Sam Ransbotham: Well, that’s maybe a great way for us to end this, [as] I think about the idea that you brought up here at the end about human creativity in a world of artificial intelligence. That’s a nice balance to the idea of all the technology here. Thank you for taking the time to talk with us, and we’ve enjoyed learning about it. Thanks.
Jeff Miller: Thank you for having me.
Sam Ransbotham: Thanks for listening today. We’re joined next time by Barbara Wixom, principal research scientist at MIT’s Center for Information Systems Research. We’ll be talking about the opportunities and challenges of researching artificial intelligence. Please join us.
Allison Ryder: Thanks for listening to Me, Myself, and AI. Our show is able to continue, in large part, due to listener support. Your streams and downloads make a big difference. If you have a moment, please consider leaving us an Apple Podcasts review or a rating on Spotify. And share our show with others you think might find it interesting and helpful.
