Football Program to Predict Which Play Coach Will Pick? : Computers: Sports fan’s software-in-progress makes defensive calls, and it’s right most of the time.

It’s gut-check time for the Washington Redskins, who are down, 17-3, in the third quarter of their Monday night game against the New Orleans Saints.

It’s also put up or shut up time for Dave Hillman’s computer software, which has been having, at best, an inconsistent evening. The computer has been programmed to assume the role of a pro football team’s defensive coordinator, which means it’s trying to predict what play the offensive team will run in any given situation.

So far, it hasn’t shown All-Pro capabilities.

In this respect, it’s emulating the ‘Skins--last year’s Super Bowl champs--who on this night are stinking up the Superdome. It’s third down and 10 to go, time is running short, and everyone in the Bayou State, not to mention the millions watching at home, knows that Coach Joe Gibbs is going to call a pass.

Everyone, that is, except Hillman’s computer program, which, for the third straight play, predicts an inside run.

“This thing will run streaky at times,” Hillman says apologetically. “It’ll hit 10 or 12 plays in a row, then throw all sorts of weird things at me.”

Shed no tears for Dave Hillman. His little toy might be shaky tonight, but it has performed admirably in the past, averaging an overall prediction rate of better than 70%. And that’s using a software program barely out of diapers, with 336 plays from only seven Redskins games in its memory.

If Hillman, 36, had the time and money, he could boot the computer up with thousands of additional bits of information, creating a technological monster that could, hypothetically, replace the defensive coach.

Don’t misunderstand: putting a lot of aging guys with thick necks out of work is not Hillman’s aim. He is something of a visionary, a dedicated trouper scouting the front lines of neural-networking and artificial intelligence. He’s trying to teach machines to think.

“You know what they’ve found with computer technology?” he asks. “Computers have never really replaced people, they’ve simply made what they’ve done qualitatively better. This software forces a coach to look at their own tendencies and say, ‘Look at what I do,’ or ‘I don’t want to do that!’ ”

*

Football prognostication might seem a frivolous way to enter the world of the future, but the fact is, Hillman came up with his program in response to a sober problem. A few years back, what he will refer to only as a “government intelligence agency” (wink, wink, nudge, nudge) asked Hillman to try to use neural-network technology--a type of artificial intelligence programming that processes diverse streams of information--to predict human behavior.

A big, barrel-chested football fanatic who develops computer-based tools for GTE, Hillman thought a program using ‘Skins games as a model might be fun. As he puts it: “I was trying to impress a bunch of guys in suits that this was a good thing.”

So he taped and charted every play from three 1989 Redskins games, noting such things as field position, down, yards to go, score, time remaining and quarter. He also included data on which play--inside run, outside run, short pass, long pass, punt, field goal attempt--was called. After feeding all this information into the computer, Hillman asked it to spit out play predictions during an actual game.

So, for example, if the ‘Skins have a second and 6 on their own 40-yard line, they’re up 14-7 with four minutes left in the half, the computer predicts a 90% chance that the offense will try a short pass, a 7% chance of an outside run, and a 6% chance of a long pass. If, however, it were second and 4, the computer shows a 79% chance of an inside run. And so on.

Does this mean the microchip stuff could replace human coaches?

“It could call the plays,” Hillman says emphatically. “The computer can’t factor in the wild card (the occasional trick play), but then again, who can? It doesn’t worry about it. There are really only six basic plays in football, and every other play is some variation. Most things in life are simpler than we’re led to believe.”

*

Big surprise! Football types tend to view Hillman’s dabbling with skepticism.

“I think it’s far-fetched,” says Greg Aiello, a National Football League spokesman. “Ultimately, the game comes down to execution, and the people on the field.”

Not that the pros and the big colleges don’t use computers. They all favor high-tech when it comes to studying “tendencies”--what an opponent will do on a given down, from a certain field position, with a particular offensive formation and players. Most teams use such data, analyzed by a computer, to develop a game plan. But Hillman’s software could call plays as the game develops.

It’s a big leap. So big that even Tom Landry, the former Dallas Cowboys coach credited with being the first to study tendencies, can’t accept it.

“You use something outside of the people on the field, and it becomes distracting,” he says. “It’s hard for me to believe you can predict all plays; you can be close, but you can’t afford a miss in a football game. We use computers to base our assumptions on, but it still comes down to the human touch.”

Well, yes and no. Hillman, an offensive lineman in high school and in the U.S. Air Force, believes that computer play-calling is undesirable because, he says, the game “boils down to 22 guys trying to kick each other in the butt--it’s mud and sweat.”

But he agrees with Robert Epstein, director of the Cambridge, Mass., Center for Behavioral Studies, who has said that, in the not-so-distant future, “There will be two intelligent species on Earth, Homo sapiens and computers, and we, the Homo sapiens, will integrate them, the computers, into every part of our lives, including sports.”

Hillman envisions educational software that would allow teachers to develop their own computer programs. And he still believes there might be practical uses for his little hobby.

“The idea of a video playbook is doable right now,” he adds. “Take live footage of a game, put it on the computer, then use it like John Madden does on TV, with all the little Xs and Os showing how the play develops.

“I could also see uses for gambling; you could set it up to play against itself. This could, for example, help bookies set their point spreads on a game. You could run a simulated game, then use the results to determine the line.”

On a scale of one to 10, 10 being the ability to predict play-calling with 90% accuracy, Hillman gives the program “a five, because the concept works.”

But to get to stage nine or 10, he must input plays from more games--plus factor in injuries, the game’s importance and whether it’s being played on the road or at home, on Astroturf or grass.

Then, of course, there’s the human element. The software is having trouble this Monday night because of a lack of raw data. On a third and 20, for example, the computer predicts an inside run (in an obvious passing situation) because that scenario isn’t in its memory. Hillman also believes the computer’s decision-making is off because of the injury factor: This year’s ‘Skins are bruised and battered, much different personnel-wise than in their Super Bowl year.

Still, the game ends with the software right 54% of the time. Drop out the disastrous third quarter and it rises to 61%. A good showing, but Hillman knows there is more work to be done.

“I’d love to refine it,” he says. “What I need is to sit down with a good defensive coach, to show them what I’m doing, and then to ask them, ‘Right before the snap of the ball, what are you thinking?’ ”