By Dan Kedmey
October 9, 2014

Andrew Ng is holding still as a co-worker carefully angles an iPhone camera in front of him. The app onscreen shows the silhouette of a head, and Ng’s face fills the space inside. The two must be aligned for the program to read his features, decipher them and retrieve a headshot of his nearest celebrity look-alike from the web. More often than not, the app fetches a dead ringer.

This is more than a parlor trick for Ph.Ds. Ng, 38, is working in one of the most contested and potentially lucrative frontiers in tech: so-called machine learning, or teaching computers to teach themselves. The app Ng is toying with is just one small demonstration of its potential. As everyone from automakers to health care providers grapples with the vast amounts of data generated–by themselves, by us–many engineers have come to believe, like Ng, that the only reasonable way to avoid drowning in the deluge is to build machines that can train themselves to parse it.

“I’ve always thought you can make computers more intelligent, free us up from a lot of the more routine work, and then we can spend our time on more worthy pursuits,” says Ng. Artificial intelligence is also, he argues, the key to dominating the next wave of Internet businesses.

A race to build computers that learn is already under way. Google and Facebook have gone on a hiring binge, and talented grad students as well as top professors have grown accustomed to six- and seven-figure salary offers from big tech firms. In May, Ng who until earlier this year was a Stanford University professor, took over the research efforts of another web giant: Baidu.

If Americans know the name at all, it is likely as the “Google of China.” In fact, Baidu is Google on growth hormones. The company has a virtual lock on its home country’s 618 million Internet users–more than double the number in the U.S. The majority of those users run searches through Baidu’s servers, which house one of the fastest-growing data sets in the world. (The firm’s version of Wikipedia, for example, has nearly double the number of articles as the English-language version.) If Baidu’s growth continues, it might soon possess one of the most valuable information reservoirs in history. “We dominate search,” says CEO Robin Li. But he argues that in order to tap it for all it’s worth, the company first needs the right mind.

Ng, who grew up in Singapore before arriving in the U.S. for graduate study, earned the engineer equivalent of rock-star status in 2012 when he created a program dubbed Google Brain. Using 16,000 computer processors loaned to him by the search giant, Ng and a team of researchers created a neural network, or artificial brain, that could watch YouTube videos and teach itself distinctions onscreen–the difference between a human’s face and a cat’s, for instance.

It may sound odd, but Ng’s work was a major breakthrough. Neural networks are computer programs that try to understand the world in much the way people do, in layers from abstract to concrete. Feed an image through one set of the program’s “neurons” and it may recognize blobs of color. In the next layer it may sort out shapes, and so on, until it reaches complex concepts like a cat’s maw. Ng’s innovation was to pack networks with enough processing power to get drastically smarter over time.

Since Google Brain, similar programs have been used to improve more prosaic technology. Google Search and Apple’s Siri, for example, take spoken commands with fewer misunderstandings as a result of Ng’s research. “Anything half related to machine learning, if I wanted it done, I would want to hire him to do it,” says Geoffrey Gordon, a computer scientist at Carnegie Mellon.

Now Ng is overseeing Baidu’s international R&D efforts from a brand-new, $300 million research center in Sunnyvale, Calif. Inside, office chairs still have promotional tags dangling beneath them advertising “adjustable lumbar support” to workers who have yet to arrive. Ng’s plan is to fill them with some 200 employees, many poached from Facebook, Twitter, LinkedIn and Google. Out of this half-vacant office, Baidu has already promised breakthroughs that will “transform the world.”

But Ng’s new office has no visible signs of a big artificial brain under construction. There are no refrigerator-size computers, no bundled cords snaking across the floor. There’s not even an employee in a white lab coat. Just beige carpets, white walls and generic office plants. Squint and you could almost mistake Baidu’s cutting-edge lab for an office of tax accountants.

The really cool equipment is in Baidu’s Institute of Deep Learning, on the fringes of Beijing, where the company is building one of the world’s most sophisticated neural networks. It will be kitted out with 100 billion connections, or 100 times as many synapses as Google Brain. Ng and his team of researchers can command its stacks of servers from just about anywhere in the world. From here, Ng will attempt to feed Baidu’s ocean of data across layers of neurons to make image recognition sharper, make voice dictation more perceptive and, the company hopes, make searching the web about much more than typing text into a blank box.

Spotlight Story
From 'Love Is Blind' to 'The Circle,' Netflix Is Getting Terrifyingly Good at Reality TV
The streamer is building a reality empire on a foundation of viewers’ most addictive impulses.

Baidu isn’t the first foreign company to pin its hopes on absorbing some of Silicon Valley’s talent. Over the years, many opened Bay Area outposts with similar expectations–often with mixed results. Samsung, the massive Korean competitor to Apple, is currently building a 1.1 million-sq.-ft. (102,000 sq m) research facility in downtown San Jose. What makes Baidu’s foray unique is that it has no plans to launch products and services in the U.S. The company wants minds, not customers.

Still, recruitment in the U.S. could be difficult. Baidu will have to compete fiercely for a pool of talent that is surprisingly shallow. Among the vanguard of top thinkers in this emerging field, most have already been recruited, according to Michael Littman, a computer-science professor at Brown University.

There are also skeptics about what neural networks can really achieve. “There’s more to human intelligence than just finding patterns,” says Ray Kurzweil, Google’s engineering director and a well-known futurist. Ng himself admits that he has at times been skeptical, steering students away from the idea. “It’s easy to hype them up,” he says. Yet the potential of Baidu’s brain has refired his imagination. Ng is hesitant to predict what one of the world’s largest neural networks might accomplish when it hums to life. But one thing is clear: he’s no longer steering talented young programmers away from the idea.

This appears in the October 20, 2014 issue of TIME.

Contact us at editors@time.com.

Read More From TIME

EDIT POST