Thomas Lee:
Adapt to self- driving cars, a scientist advises.
Scientists have long tried to build computers that can think like humans: Eliza, Deep Blue, Siri to name a few.
Andrew Ng probably occupies a more rarefied place among artificial intelligence experts: He spent plenty of time exploring how both humans and machines learn. Ng not only chairs Coursera, the online education startup he co- founded in 2011, but as chief scientist of Chinese search engine Baidu, he oversees the company’s efforts to develop self- driving cars.
So if anyone knows how humans stack up against computers, it would be Ng. And judging from my recent conversation with him, humans remain on top of the proverbial food chain — at least in some ways.
“Humans have a learning machine called the brain,” Ng said. “By age 6, humans can reasonably move and communicate with a sophistication” that would take even the best computers years to master.
At the same time, he noted, “Amazon makes better book recommendations than my wife does.”
But the constant need to compare humans with computers falsely suggests they exist on an equal level, said Ng, who is also a professor of computer science at Stanford.
Take Coursera. The company is helping to revolutionize higher education by offering thousands of free online courses from 120 universities and colleges, including the University of Michigan, Duke and Johns Hopkins. The idea is to instantly expand access to education for students across the world who otherwise can’t afford it.
“We realized we had an opportunity to transform education,” Ng said. “I used to teach 400 people at Stanford. I did the math, and for me to teach 100,000 students, I would need to teach for another 250 years.”
But critics say online classes dilute the quality of an education best delivered by personal interactions between teacher and student in a classroom. They fear that computer- based instruction could one day replace teachers. The controversy prompted the Commit-
tee on Institutional Cooperation, a group of provosts from large universities, to issue a paper in 2013 that called for schools to embrace the technology — but on their terms.
“There is no conceptual or empirical data to support the contention that online classes lag in quality relative to the traditional classroom experience,” the report said. “It is more accurate to say that online classes have advantages and disadvantages, just as is the case for regular classroom instruction, and that these advantages and disadvantages play out in different ways for different subjects, and for different kinds of students.”
“The question ‘ Which is better?’ is oversimplified,” it said.
The same debate applies to self- driving cars and whether computers are “better” at driving than humans. Though Ng is confident that autonomous vehicles can reduce traffic accidents and fatalities, computers cannot yet recognize a construction worker’s hand gestures.
“We need a shift in how we think of autonomous vehicles,” Ng said. “Right now, we have a traffic system designed for humans, and we can’t expect computers to behave like humans. A horse is not the same thing as a train.”
Unless artificial intelligence makes an exponential leap in ability over the next few years, government regulators should change the system to accommodate self- driving cars instead of the other way around, Ng said. For example, let’s equip that construction worker with a wireless beacon that the autonomous vehicle can recognize.
“Today, we are saddened if a pet dog runs onto a railway track and is killed,” Ng recently wrote in an essay for Wired magazine. “Yet we do not blame the train for this accident, because we understand it is the nature of the machine to run on its tracks.”
“We accommodated that nature by developing new ways to behave around trains — such as not standing on the tracks — along with new infrastructure like railway crossings with distinctive lights and bells to keep people out of danger,” he wrote.
“We can’t expect computers to behave like humans.” Andrew Ng, Baidu scientist and Stanford professor