The NFL is sponsoring a competition called the "Big Data Bowl". The goal is to create an algorithm which can most accurately determine how many yards a running back will gain on a given play, given some prior information. Whoever comes up with the best model gets a prize of $75,000.
The competition is being sponsored on a website called Kaggle, which is community for data scientists and artificial intelligence (AI) researchers. The website has hosted many competitions, such as automatically determining home value (Zillow), or predicting a person's movie preference (Netflix).
In order to train the models, the NFL provides extensive historical data. For every given play, at the moment the ball is handed off to the running back, there is information about the position, direction, velocity, acceleration, weight, height, windspeed, offensive formation, defensive formation, etc, etc - for all 22 players on the field. The dataset is ridiculously extensive and gives AI researchers all the information they need to train their models.
I believe the goal of this effort is to basically create a new "Next Gen Stat", which measures rushing yards gained vs expected yards gained. For example, if the AI model predicts that Saquon is supposed to only gain 3 yards on a play, and he gains 8, that means Saquon overachieved by +5 yards.
I'm excited because I've been beating the drum for technology in the NFL for a long time. While this effort is just going to create a new Next gen stat, this is only scratching the surface of what AI can do for the NFL. AI research is advancing at breakneck speed and it's only a matter of time before NFL teams are hiring teams of nerds to gain a competitive advantage in a billion dollar industry. Here are a few uses of AI that I envision.
1. Reading the field pre-snap and calling an audible. Teams with veteran quarterbacks rely on the experience and expertise to read the field, and call an audible. But not every team has a Manning, or a Brady, or a Rodgers. Some teams have quarterbacks that, although they have a great arm, don't have the mental acuity to call an audible. AI models can do this automatically. They can scan the field, look at the defensive alignments, and determine if an audible is appropriate.
2. Informing the quarterback of his best receiver option. For passing plays, AI can be used to inform the quarterback of who his primary, secondary and third reads should be, based on analytics. If your receiver is going to be running a double move, and the cornerback has a history of biting on double moves, AI will tell you that receiver should be your primary.
3. Calling the play. I've always wondered what makes a good NFL playcaller. It's currently an art more than a science. It can be a science. You can train models to call plays, based on historical data. Coaches have to do extensive film study to figure out tendencies of their opponents, and exploit them. This can be done using AI.
4. Inventing new plays. Generative adversarial networks (GANS) are used to create new data. They are currently used to synthesize fake voices, and create images of people that don't exist. GAN - Some cool applications
5. Call penalties. This technology would require some more advancements in computer vision, but in five years referees can be made obsolete. (I'm sure Lions fans would be on board right about now)
Ultimately, football can become a cat and mouse game between AI systems. Just like on Wallstreet and in Silicon Valley, AI talent will become a premium.
TLDR : The nerds are coming to take over the game of football