[SI] Could a computer coach an NBA team?

[rquoter]BOSTON — OK, so the question in the headline is obviously a ridiculous one. But the subtext at the MIT Sloan Sports Analytics Conference is that coaches, as in-game decision-makers, do a bad job compared to what a computer could do in that role.

It’s not because coaches aren’t brilliant at what they do, said Tarek Kamil, an executive director for online strategies at InfoMotion Sports Technologies and a consultant to various college basketball teams. (Kamil gave a brief presentation on simulation models at Sloan today). It’s because no human being, no matter how smart, could cycle through the thousands of variables at play during any given moment in a basketball game and always come up with a decision that will increase his team’s odds of winning that game. And so Kamil thinks that the day will come — sooner than you think — when an NBA team will have an assistant coach sitting on the bench with an iPad or a Tablet and giving the head coach instructions on what moves the team should make and when. “It’s going to be a huge fight,” Kamil said. “Coaches notoriously have a had a god complex.”

The computer will ideally simulate the outcome of a game thousands of times and produce in-the-moment suggestions — implement a full-court press, play a certain five-man group, take out a superstar whose heart rate has been elevated for too long, etc. That last bit about heart rate isn’t a fantasy, according to Kamil. Teams, right now, have the capability to measure how quickly individual players get tired via heart rate monitors players wear during practice. And Kamil’s company produces a special ball that measures everything about a player’s ball-handling — how high and how fast he dribbles with each hand, the sort of spin he puts on the ball, how his ball-handling changes when he gets tired and lots of other stuff. If you put all of this data together, teams could know at any moment whether Dwight Howard has crossed the line where his fatigue is problematic and whether you should force Dwyane Wade to his right after he has played 40 minutes.

This is the computer as coach, and if sports really go in this direction (and they already have, to a small degree), human coaches will naturally evolve more into what Kamil calls “father figures” as computer models take over decision-making. Again, this is already being done, though not everyone is in awe of a computer’s coaching abilities.

At the Sloan conference last year, Avery Johnson bemoaned the fact that in the 2007 playoffs, when Johnson was the head coach of the Mavs, he went small in the first round against the Warriors because the secret advanced stats system the Mavs used at the time suggested he do so. The Mavs, of course, lost that series. Advanced stats also say New England Patriots’ head coach Bill Belichick was right to go for it on fourth down deep in his team’s own territory in a 2009 game against the Colts. The Patriots failed to convert on that fourth down, and Belichick was (briefly) roasted. And Kamil knows the computers will, at some point, tell a coach to remove LeBron James from a game late in the fourth quarter because his heart rate data says he is too tired to play as effectively as the team needs him to. “If the odds say you should take out LeBron James, you’re going to be on the hot seat,” Kamil said. “But over the long-term, you will be more successful if you play the odds like this.”

During a panel on Friday, Mike Zarren, an executive with the Celtics, said these sorts of advanced models have proven that baseball managers regularly do the wrong things — they order too many sacrifice bunts and stolen bases, for instance. But Zarren said advanced analytics in basketball have so far proven that, in general, basketball coaches are doing the right things.

Basketball is different from baseball, though. Baseball is a game of discrete, static states that change with each pitch. Experts have already analyzed what managers should do in just about every possible state; if, say, their team is trailing 3-1 in the bottom of the seventh and they have a runner on first, managers (if they want) can already research whether having their baserunner steal second increases their odds of winning enough to justify the risk involved.

Basketball is more fluid, since fatigue is more of an issue and the game involves 10 guys interacting at the same time. That has led some folks to suggest you cannot use quantitative methods to make in-game decisions the same way you can in baseball or football — that there is too much of a human element to consider, too many “gut feeling” decisions, too many intangibles.

But a lot of people here believe you can quantify almost anything, including basketball, and that a computer could indeed make a better in-game coach than a human. That would not only change the role of a coach; it would also change things like how a team has to practice. The computer model may suggest your team should start pressing at some point in the third quarter, but you can’t do that if your team doesn’t practice a full-court press in a serious way.

Kamil predicts that some team will have the iPad/Tablet assistant coach soon, and that will either be one of the teams ahead of the advanced stats curve — the Mavs or Celtics, for instance — or a bottom-feeder desperate for a new edge. He also predicts the league might consider banning this sort of thing.[/rquoter]

http://nba-point-forward.si.com/2011/03/05/could-a-computer-coach-an-nba-team/

 

I think the Morey connection is interesting in that he might look to replace Adelman, should he be on his way out, with a young coach, who's on his way up. Someone with fresh ideas, someone Morey can work with. Morey's yes-man if you will.