46 How can you tell a good seeing-eye dog from a great one? Easy, explains Bar-Ilan computer scientist Dr. Reuth Mirsky: Great ones know how to disobey. “Intelligent disobedience,” or the ability first to determine an unsafe command, and then to disregard it or insist on a different action, is a critical quality of the animals entrusted with the safety of the visually impaired. And that is precisely the quality that Mirsky seeks to develop in Golda, a seeingeye robot. “Robodogs have several key advantages over the real things,” says Mirksy, whose Goal Optimization using Learning and Decision-Making (GOLD) Lab—hence the robot’s name— studies cooperative artificial agents that can learn, adapt, negotiate options, and in turn be great teammates. “For starters, they don’t need to undergo the expensive and intense training that real dogs do. It’s simply a matter of installing the guiding software. They also don’t need to use the bathroom, to sleep, and most of all, to retire. The opposite: They can simply be updated over time and as new technologies emerge.” Pointing out that her quadruped aims to be more platform than real puppy, Mirksy explains that robodogs are actually a basis on which devices are interconnected, data is collected, and new technologies can be added all the time. By integrating GPS into Golda, for example, it can enhance its owner’s navigation of the world—and, she hopes, his enjoyment of it, too. “Real seeing-eye dogs can’t plan a route for their owners to a certain destination, but a robodog can. This can make life so much better for the visually paired.” And perhaps for the rest of us, too: Mirsky and her students are also training Golda to be a “social robot,” or a robot designed to interact with humans or other robots by following social rules. This means, Mirsky says, that Golda would know not to bump into someone while guiding its owner on a busy street, or to trod on someone’s flowers, even if it’s the easiest way to get Even Better Than the Real Thing The Goal Optimization using Learning and Decision-Making Lab
RkJQdWJsaXNoZXIy NDU2MA==