Google AI Software Defeats Human Champion In Board Game Go
It's the type of artificial intelligence that could make the difference in the development of autonomous cars.
The success of the program known as AlphaGo suggests machines can do a better job of detecting directions in which actions will be most beneficial at a much higher level than we thought. That's exactly what's needed with autonomous cars. Even the current Google car operates on techniques that have been around since the 1970's with rules that tell it what to do.
"If you're in a middle lane, and there's a cyclist who wants to turn right, and do this," Stuart Russell, UC Berkeley Computer Science Professor said.
Russell researches artificial intelligence, and says that in the real world, there are plenty of situations where a rule doesn't apply.
"For the car to understand what to do in certain situations, it has to reason about what might happen next, and how desirable that is," Russell said.
That means figuring out the intentions of other drivers. Like that guy in front of you. Is he drunk? Should you quickly pass? Should you exit the freeway?
"Those are the complicated decisions that people make by understanding the intentions of other players on the board, so to speak. This is the kind of reasoning that we don't quite know how to do yet, but it's in the works," Russell said.