Yahoo Autos Before Computers Master the Art of Driving, They Need to Become More Human
Making a computer that drives more reliably than a human should be easy. Making a machine that can safely negotiate the chaos of our roads and still move at reasonable speeds—that’s the hard part. But it will be essential if robocars are ever to co-exist with human drivers.
“[Humans] make quick judgments based on not a full analysis, and it gets us by most of the time,” explains David Strayer, a professor of cognition and neural science at the University of Utah. “When we make decisions, we basically use a lot of shortcuts to get us to a good answer most of the time.” In contrast, artificial-intelligence computer systems offer “kind of a brute-force analysis that pulls together a lot of different pieces of information.”
Homo sapiens’ faith, intuition, hope, altruism, and wild-ass guesses are precisely why human drivers can negotiate their environments at high speeds. Without the willingness to break rules and act with incomplete information, machines will struggle to make sense of our unpredictability. “The challenge right now is that autonomous cars tend to treat people as merely these obstacles that are in the way, and they try to be very careful about not colliding with them,” says Anca Dragan, who runs the InterACT Lab at the University of California, Berkeley, which investigates the interactions of humans and robots. “This leads to trouble because people can take advantage of this defensiveness.” In other words, given the cautious, law-abiding behavior of a driverless car, we malicious humans can game that vulnerability for our own commuting advantage.
Eventually, though, computers may process their environment in a way we can’t comprehend. “The new thinking, which is still evolving, is this ‘deep learning’ stuff,” Strayer continues.“The computers actually develop their own algorithms and tune themselves to make good decisions. They’ve been very impressive and able to outperform humans in many respects. [Deep learning] has the potential to be very powerful and very effective. But unfortunately, we may not ever really be able to understand what it does. It’s effectively a black box, impenetrable to us. It will be abstractly representing the problem space in a way we may not need to know, but I don’t think we could know if we wanted to.”
These deep-learning machines could eventually resemble our own limbic-system lizard brain, which is largely unknown to us but saves our bacon in many situations. Before computers can master the craft of driving, though, they will need to become more human.
Auto•No•Mo'•Us: Return to Full Coverage
You Might Also Like
