Quiet, McEnroe. The Robot Called It.
Sunday, September 7, 2008
Disputes over line calls used to be one of the main joys of tennis -- this, after all, is John McEnroe's game. But fans rarely see players explode in rage anymore. In high-profile matches, human umpires have largely been replaced by a machine called Hawk-Eye, a kind of computerized ump that stitches together video footage from several high-speed cameras to produce a three-dimensional simulation of the ball. Hawk-Eye's decisions are final: When a player challenges an umpire's call, the system displays its view of what just happened, then displays a judgment on the screen -- in or out -- that the human umpires are compelled to accept.
Hawk-Eye represents one extreme in the growing adoption of technology to solve disputes in sports. On the other side, you've got Major League Baseball, which has long resisted any kind of instant-replay system. Pro baseball recently played its first games under a new rule that lets umpires review video in the limited scenario of "boundary calls" -- essentially determining whether a home run was really a home run. Unlike tennis, MLB, the NFL, the NBA and the NHL all give human officials the final say in interpreting instant-replay footage.
The trouble is, technology can introduce as much uncertainty as it solves. For one thing, photography doesn't give clear-cut answers. From one angle, a ball may look in, while from another it looks out.
Hawk-Eye is also demonstrably fallible. According to a fascinating paper by Harry Collins and Robert Evans of Cardiff University, the system's manufacturer reports its average error as 3.6 millimeters. The International Tennis Federation, which tests the line-calling equipment, allows for Hawk-Eye to be off by as much as 10 millimeters in some situations. This means that if a ball lands nine millimeters out, Hawk-Eye might call it in by one millimeter.
It isn't terrible that Hawk-Eye is sometimes wrong -- after all, humans often make mistakes too. What is odd is that the system's designer, Hawk-Eye Innovations, has never explained these failures. Hawk-Eye uses up to six cameras placed around the court, but the graphic that it shows to judges and TV viewers does not include actual footage from any of them. Instead, the system creates a composite of what it thinks happened to the ball. Collins and Evans argue that these composites subtly trick viewers. The simulation takes on an air of reality, even infallibility, when in fact it is only a statistical estimate. At the very least, the researchers say, Hawk-Eye should report its confidence -- that it is X percent sure of its ruling. They also push for a more general "health warning": When CBS broadcasts Hawk-Eye simulations on TV, it should remind viewers: "This is only a virtual representation of reality. It's not what actually happened."
Tennis -- with its two players, small field of play and bright-line demarcation between balls that are in and balls that are out -- is a comparatively easy sport for computers to umpire. In other sports, refs have to take into account many more variables before making a call. To be able to tell whether a football runner is down before he fumbled, for example, a computer would have to somehow keep track of every player, which one has the ball, when the player with the ball gets hit, and when and how the ball comes loose. That task would likely require an array of sensors and sophisticated image-processing techniques -- probably not yet a possibility.
Beyond the technological obstacles, the age of Hawk-Eye presents a larger philosophical problem. Sometimes the computer makes a call that no human -- not the fans, not the umps, not the players -- can quite understand. Late in the 2007 Wimbledon final between Rafael Nadal and Roger Federer, Nadal hit a deep ball that Federer let go, thinking it was out. The umpire thought so too, and TV replays showing the ball from Federer's side seemed to confirm it -- the ball looked a good half-inch out. But when Nadal challenged the call, Hawk-Eye called the ball in. On the Hawk-Eye Innovations Web site, the company's representatives posted an explanation that blames the dispute on the limited perceptive capacities of TV cameras and human eyeballs. When a tennis ball smashes into grass at high speeds, it compresses, skids for 10 centimeters or so and then takes off, the company said. Hawk-Eye's fast cameras were sensitive enough to see the ball just clip the base line, while TV cameras and viewers caught only a blur while the ball skidded away from the line, making them think the ball was out.
Got that? Because it's so perceptive, Hawk-Eye makes obsolete every assessment tool that humans have ever used to adjudicate sports disputes: our eyes, our TV cameras, even perceptible marks on the ground. In their paper, Collins and Evans argue that this is too precise. By erasing all of tennis's ties to human perception, Hawk-Eye renders the game interpretable only to computers. That's fairly ridiculous: After all, computers aren't paying to see two human beings hit a ball over a net. People are.
Farhad Manjoo is Slate's technology columnist and the author of "True Enough: Learning To Live in a Post-Fact Society."