Computer scientists used football footage to develop a sophisticated kind of artificial intelligence.
Artificial intelligence on the football field. I’m Bob Hirshon and this is Science Update.
Football season is upon us, and coaches all across the country are racking their brains to figure out how to win the next game. They often review game footage to refine their strategy. Now, a computer can do that too. Oregon State University computer scientist Alan Fern and his colleagues want to make artificial intelligence that can analyze complex visual scenes. So they programmed a computer to track football passing plays, learn from them, and create new ones.
This computer was able to design much better plays, after having watched all of this previous video, than plays it could design without watching any of that video.
Fern says the current program still couldn’t outwit a serious fan, much less a Vince Lombardi. But he says it may eventually analyze the movements of all kinds of groups, from shoppers in a store to soldiers on a battlefield. I’m Bob Hirshon for AAAS, the Science Society.
Making Sense of the Research
There are a lot of things that computers can do a lot faster and more accurately than people: for example, solve mathematical equations, or search thousands of documents for a particular word or phrase. However, some things that come as second nature to people remain extremely challenging for computers—or perhaps more accurately, for computer programmers.
One of these is processing and analyzing complex visual scenes. Humans can watch a dance performance, a crowd reacting to an emergency siren, or football play, and make sense of the movements of dozens of individuals doing different things and affecting one another. We also can judge what's working well and what's counterproductive. We don't even think about how complicated this is.
But when you consider a football passing play, for example, it's very difficult to set up formal rules that can be put into a computer program. First of all, the computer has to keep track of 22 players, who are dressed alike and often block one another from being seen. Then it has to track where the ball goes, and whether or not the play was successful. From the snap to the catch, there are fakeouts, blocks, and other complex maneuvers that each team uses to outwit or overpower the opponent.
Fern's team designed a program that can, to a limited extent, keep track of all this. Moreover, it can learn: by feeding it footage from more and more games, the computer was able to design better and better passing plays for a video game. So in effect, it was doing what a coach does when he watches game footage to improve his strategy.
However, the computer program still isn't nearly as sophisticated as even a casual sports fan. It's worth remembering that a computer lacks the life experience that helps people understand football, from memories of playing the game itself, to a broader understanding of how ball games work, to a lifetime of experience watching other people execute coordinated tasks. Everything has to be programmed from scratch. Given that, Fern's algorithm represents a significant achievement in artificial intelligence.
Although Fern would like to improve the system to the point where it could actually give coaches an idea or two, the applications don't start and end with football. Military commanders, for example, have a far more serious motivation to design coordinated operations for a group of people battling motivated opponents. If a computer system could run countless battle scenarios and come up with strategies for beating the enemy, it could become a valuable asset to national defense. Other possible venues for the computer program include business (for instance, tracking the movements of customers in a store, depending on how items are laid out) and counterterrorism (studying surveillance footage for unusual and suspicious movement patterns in crowds.)
Now try and answer these questions:
- Why was this program designed?
- Why is it hard for a computer to understand football plays?
- Why do you think the researchers chose to study football in particular?
- Can you think of other applications for this kind of artificial intelligence?
You may want to check out these related podcasts:
The Science Update Computer Composer describes another AI system—one that can generate original music.
The Science Update Averaging Faces looks at another skill that comes easily to people but hard to computers: facial recognition.
The Science Update lesson Computer Composer describes another AI system—one that can generate original music.
The Science Update lesson Averaging Faces looks at another skill that comes easily to people but hard to computers: facial recognition.