Inside Baidu’s Bid to Lead the AI Revolution
Presumably, Robin Li wanted attention last summer when he decided to launch Baidu’s bid for the future of self-driving cars from the front seat of a car that was driving itself. He wanted to draw attention to Apollo, the company’s new set of artificial intelligence-driven tools, which Li hopes will come to power vehicles everywhere. Having launched China’s dominant search engine, Li is a celebrity in his home country. But even Li didn’t anticipate the amount of attention he would get. Automated motoring is still very much forbidden in China, and Li was livestreaming a video that showed him breaking the law. That’s how he became the subject of his own viral video. “I didn’t realize it would catch a lot of attention because it’s not really allowed to have a self-driving car,” he says now.
He can laugh about this now, three months after the fact, as we review the experience from top floor of Baidu’s old headquarters, a seven-floor building in Beijing’s Haidian District. A newer, larger building, complete with a two-story slide in the lobby and a conference room shaped like a bear claw, is just a 15-minute drive away—neighbors, by Beijing standards. In both, employee identification badges have been replaced by facial recognition technology. Order a green tea from the vending machine, and you can pay for it by looking at a camera. These futuristic campuses offer a glimpse at the scope of the computing power the company has amassed.
I traveled to Beijing to chronicle this tenuous moment in Baidu’s history. At 18 years old, Baidu has built the country’s dominant search engine, a business substantial enough to make it one of the most important tech companies in all of China. And yet: It’s hard to be a Chinese search company in 2017, when Chinese people increasingly navigate the web through apps, not via a browser. As WeChat and Alibaba deftly transformed their companies to suit mobile, Baidu missed this shift. It has been struggling to catch up ever since. To ascend to future dominance, Baidu needs to find a new way to grow—and fast. Fortunately, the world has provided Li with just such an opportunity: “the era of artificial intelligence,” he tells me. Li is betting Baidu’s future on the promise that he can own the future of artificial intelligence, in Asia and beyond.
So far, North American companies have been earlier to invest in AI, and first to introduce both the new technologies and the resulting products. Many of AI’s most forward-thinking researchers are in Silicon Valley or Canada. The large US tech companies were first with everything from the technologies that will enable self-driving cars to smart speakers such as Google Home and Amazon Echo.
But Li has reasons for thinking that there are advantages for a company trying to stake claim to AI in Asia—even as an underdog. It’s still not clear how artificial intelligence will reshape our lives, but it is clear that the change is coming. And those best positioned to deliver that change—and reap the spoils it will inevitably introduce—are those that master and advance the underlying technology. That’s where Baidu is competitive. Like America’s Big Five, Baidu has substantial computing brawn, a suite of AI-powered services called Baidu Brain, and a fast-improving voice assistant platform called DuerOS. “We are one of the few companies who have the capability to develop this type of technology,” says Li.
Baidu’s biggest advantage is one of place and time. Li is introducing his strategy within a culture that has few ethical hang-ups around AI development. In the West, where people are concerned about the biases with which we program our algorithms and the speed at which they’ll disrupt traditional career paths, new technology emerges more slowly. In China, it’s the reverse: There’s public pressure for companies to move as fast as possible. In July, the Chinese government issued a development plan that aims to make it the world leader in AI by 2030. Then there’s the raw fuel that powers the algorithms: data. There are close to 731 million people online in China, well more than double the number of connected Americans. When those millions search and watch videos and make payments, they leave a digital trail of information powerful enough to make any AI researcher salivate.
All of this means that China is poised to be a hotbed of AI development in the near future. Li believes he’s setting Baidu on course to own this next revolution—one that, in turn, will vault Baidu to its rightful place in the stratosphere. Soon, Li intones, his company will deliver the AI technology that infuses everything and every system—from medicine to entertainment to cars—with intelligence. “In human history, humans invented tools, and then had to learn how to use them,” Li tells me. “In the future, devices will need to learn human.”
It’s a revolution that Li can’t afford to blunder. After all, he missed the last one.
In China, people often talk about the singular controlling power of the BAT: the three companies that control the technology industry. Along with Alibaba and Tencent, they’re talking about Baidu. Li launched the company after a tour in the United States. He’d gotten his masters in computer science from the University of New York at Buffalo, and he spent two years as a staff engineer at Infoseek. Upon his return, he and cofounder Eric Xu built Baidu into the biggest search engine in China, and took the company public in 2005.
It seems like a remarkable success story. But for many years, as Tencent and Alibaba have grown, Baidu has been losing ground. This stems from one big screw-up. While Alibaba, Tencent, and others developed products like WeChat that enticed Chinese internet users to fall in love with their phones, Baidu clung stubbornly to desktops. The company missed the shift to mobile. By the time Li began directing resources to mobile development, Chinese internet users were already wed to apps. They could shop and book flights and chat from within WeChat, so why did they need to pull up a browser and search Baidu? “During the mobile age there’s not much we could do,” says Li. Search benefits from the network effect: The more people use it, the more data the company can collect about what they want, and the better it becomes. But if people don’t use it, it loses its edge quickly. Says Li, “Even if you had the best technology, it doesn’t matter because data is not available to you.”
Baidu tried to make up for its misstep by pouring hundreds of millions of dollars into ecommerce services, like food-delivery and online malls, but those largely flopped as well when consumers didn’t embrace them. Then in 2016, Baidu faced a crisis. A college student with cancer died after following a treatment he found through an ad on Baidu. Authorities tightened regulations on medical ads, which had been an important source of revenue for the company.
By the end of 2016, Baidu needed a turnaround. It needed new leadership with a vision for streamlining its technology into a bold strategy. Li was convinced it needed his old friend Qi Lu.
Among engineers, Qi Lu is a star—so much so that, like Madonna, he’s known in the field by his first name: Qi. He’s a computer scientist with a sharp intellects whose career experience maps directly to the birth and evolution of the internet. At Microsoft, he was, among other things, CEO Satya Nadella’s top deputy. He helped lead the company’s AI strategy. He commands the respect of the most respected practitioners in the field, people like Canadian computer science professor Yoshua Bengio, who first agreed to consult for Microsoft at his request. He calls Bill Gates a friend. But Qi has been just as good at identifying and nurturing young talent. As Jesse Lyu finished up YCombinator in 2015, he remembers emailing Qi, whom he’d met at Demo Day, for advice. Qi invited him to join him at Microsoft’s Sunnyvale office on a Sunday afternoon.
With more than 40 patents and a PhD in computer science from Carnegie Mellon, Qi has spent several decades working on search-related technology, starting in the late 90s at Yahoo. He met Robin Li 20 years ago, when they were both working in the Valley. “Over the past 10-15 years, we have seen each other on a regular basis to exchange ideas,” says Li. “I talked about what I saw in China, and he talked about what he saw in technology in big companies like Microsoft.”
Last fall, when Qi left Microsoft to recover from a biking injury, Li saw his opportunity. He recruited Qi to Baidu. From Qi’s perspective, the timing was spot-on. Working within the Chinese market was an advantage too great to pass up: The perfect storm of raw material, government support, public will, and talent. These sentiments are echoed by technology’s top brass across China. Business leaders are getting remarkable pressure and support from the government to move quickly. I recently spoke to the prominent venture capitalist Kai-fu Lee, who has written a book on artificial intelligence and is building a studio of AI startups in Beijing. He said the government is making a good deal of funding available to support AI. “Even small cities are putting together $100 million programs,” he said. “If you’re an AI company and you want to set up, tell them how much money you want.”
For the most part, the general public supports this effort. I asked Li whether there was a concern among Chinese business leaders that AI could lead to unintended consequences. “It’s not that debated in China,” he said. “The government is more focused on the positive impact of AI and I also agree with the government.” Li believes that AI isn’t that different from other new technologies. He compares it to the advent of ecommerce, explaining that at first, people were afraid to put their credit cards online. “Things change. You have to get used to it. Humans can always find ways to manage or deal with those potential threats,” he says.
This atmosphere was a strong draw for Qi, who arrived in Beijing in January, assuming the role of vice chairman and chief operating officer. While Li remains in charge of strategy, Qi runs the company’s day-to-day operations. They meet twice weekly, and talk regularly. Those conversations are important; they will determine the future of Baidu.
Qi joined a company that had already been sinking resources into artificial intelligence and its precursor, deep learning, for six years. In 2014, Baidu named the gifted researcher Andrew Ng as its chief scientist, and put him in charge of its newly opened Silicon Valley research center. Ng attracted scientists from around the world who wanted to work with him. He set the strategy on DuerOS and Baidu Brain, and he played a role in the earliest stages of Baidu’s self-driving efforts. Qi planned to embrace and expand these efforts.
Baidu’s challenge was focus. The company was also doing a bunch of other things. It had three automotive units, for example. Qi spent his first few months on the job reorganizing Baidu around two priorities: strengthening the company’s core business by shoring up its approach to mobile, and transforming it into an AI leader. He shut down units like the mobile medical division, while ramping up other areas like the company’s corporate venture arm, Baidu Ventures. He consolidated the self-driving efforts into one unit that reports directly to him. “I [also] run R&D, sales, and marketing, because I want to make sure that our overall strategy is fully, fully in sync,” he told me when we spoke last summer.
Early on, Robin Li asked Qi to check out another idea. He’d been wondering what would happen if the company opened up its self-driving technologies. Essentially, it would offer car makers a de facto brain for their cars in exchange for access to their data, which would help train its algorithms. Says Li, “[Qi] did his due diligence and came back to me and said, ‘This is a good idea. We should do it.’” They called the program Apollo.
Usually thick with smog, the Beijing air was crisp and clear on the day of Baidu’s tech conference. People filed outside the downtown conference center, also known as The China World Summit Wing and Kerry Hotel, for a glimpse of the company’s most ambitious project. There, in the parking lot just a few feet down from the piles of orange rideshare bikes, was a small pod-like bus, nearly as tall as it was long. It was surrounded by a red rope, and a security guard kept onlookers from getting too close. The bus is the first of a number of vehicles that will soon propel themselves across China on their own. Built jointly by the Chinese manufacturer King Long and Baidu, the bus will begin carrying passengers on short, designated routes by the middle of next year. It’s powered, as the logo on the bottom of the vehicle reads, by Apollo.
By 2019, Li says Baidu’s partners will mass produce semi-autonomous vehicles, which will have many self-driving features but still require human interaction. By 2021, says Li, autonomous cars will be in full production.
As a set of free software tools available to anyone, Apollo is Baidu’s biggest bet, and its most important project. Qi estimates he spends 40 percent of his time managing this effort. Baidu is hoping to do for autonomous vehicle manufacturers what Android did for smartphone makers: create an environment in which anyone has a shot at building a good product. The Apollo platform consists of core software, a number of cloud services, and self-driving vehicle hardware such as GPS, cameras, lidar, and radar. If it works, it could help Baidu catch up to companies that have been developing technologies for much longer. Alphabet’s Waymo, for example, has been in development since 2009.
From the start, building an open ecosystem for developers has been challenging. Baidu already had a couple of close corporate partners. Robin Li worried that if the company opened up the platform and its pre-existing partners were treated the same as other automakers, they would be upset. But when he approached them privately, he discovered that they were pleased to hear the idea. For them, it meant that Baidu’s technology would get better faster. “We see a lot of signals other companies don’t see. We can feed those signals to them,” says Li.
There was another, tougher tech concern. “Internally, the codes were not that clean,” says Li. If Apollo made them available, he was concerned people would criticize the engineers for it. It would take several months to clean the code up, and Baidu’s engineers already felt they were running behind. But the work it would take immediately, he felt, would allow them to speed up their efforts long-term. The engineers got to work immediately. By July, Apollo was ready.
An open ecosystem is a particularly smart approach in China, where, rather than a few dominant players, there are more than 200 car manufacturers. Unlike the United States auto companies like Ford or General Motors, it’s less clear strategically which companies in China make good partners. Now, any one of these companies can download the software to begin manufacturing on its own. Consider Chery, a Chinese auto manufacturer headquartered in Wuhu, which is understood to be one place where China has relaxed its policies so that companies can experiment. The 20-year-old auto company has 3,500 engineers, explains Huang Yong, the executive director of Chery’s Intelligent Vehicle Technology Center, but it lacks the technological know-how to enable autonomous driving. He says the decision to work with Baidu was clear: “They are the leading player in this area.”
Less than half a year after Baidu announced Apollo, more than 70 companies have signed on as partners. The roster includes plenty of international companies, such as Ford and Nvidia. Of course, just as Android’s free operating system eventually resulted in more ads for Google as more people used Google’s products, Li anticipates that Apollo will help Baidu make money. To start, Li expects the company will sell services, such as mapping or simulation training. But once Baidu begins to digest the data Apollo will collect, the opportunities are limited only by the bounds of its executives’ imaginations.
For Apollo and other projects to work, what Baidu needs most is smart engineers. As US companies have discovered, it’s very hard to keep the most talented among them. For the past three years, Baidu had a global AI celebrity leading its AI team: Stanford’s highly esteemed Andrew Ng, who was also a cofounder of the celebrated Google Brain. But several months after Qi joined the company, Ng stepped down, citing other projects he wished to pursue. He has since released an online course, and there are reports that he’s working on a startup.
Ng’s replacement is a Chinese engineer named Haifeng Wang. Based in Beijing instead of Sunnyvale, Wang is a different type of leader. Ng was an academic superstar and charismatic presenter who regularly drew researchers looking to break new ground in the field. Wang, according to people who’ve worked with him, is a strong, pragmatic engineer. Says Kai-fu Lee: “He gets stuff done, executionally.”
His appointment is indicative of another critical shift in China right now. Chinese companies once had to look to North America to hire AI researchers. That’s changing. “The gap between China and the United States is rapidly closing,” says Qi. The North American market remains vital. (After acquiring a small startup called Kitt.ai over the summer, Baidu opened a Seattle office.) Yet increasingly, Qi believes the company’s engineers will be homegrown.
There is evidence of this all around me at Baidu World, where the company has programmed an afternoon of work sessions and lectures with Wang and other engineers. Developers crowd into the hallways outside the conference, comparing notes on their current projects as they wait for the doors to open for the morning keynote. There’s rumor Baidu will be announcing a new hardware product, and an engineer tells a reporter he’s heard it’s a smart speaker. There are developers as far as I can see, and these developers have nothing but opportunities. Baidu must convince the most talented among them that its set of problems are the most interesting, and its chances of success are the most assured.
Steve Jobs’ most anticipated announcements were often preceded by a familiar phrase: “There’s just one more thing.” He’d slip in the long-awaited product, as if an afterthought. An iPod. An iPhone. A MacBook. So as Baidu’s Jesse Lyu, the young product guy behind the company’s smart speaker, finished up his recent speech at the company’s annual Baidu World tech conference, he was channeling Jobs’ force when he flashed a confident smile, and said, “I have just two more things!”
The joke didn’t register. Maybe it was because the reporters, developers, and analysts crammed shoulder-to-shoulder inside the downtown Beijing convention center weren’t Jobs aficionados. Maybe it was because we had just sat through a 45-minute description of the object Lyu was presenting—a tower of boxes in bright primary colors—in which he provided great detail on its plastic casing, including a several-minute interlude by the Swedish designer (in Swedish), but he never explained exactly what they did. (Cliff notes: Imagine a fancier version of the Amazon Echo.) His attempt at one-upping Jobs was lost on the restless crowd, which had already started to thin by the time Lyu got to his two more things: a small dog-like robot and an orange robot that could bob and sway along to music.
The effect was clear. Baidu’s products were cool. But they were no iPhone.
Then again, the global battle for AI dominance will not be won by the business with the best speaker or the first autonomous vehicle any more than the vast mobile market belonged to the first company to stake claim (that would be Nokia).
Baidu doesn’t need to one-up Jobs in order to succeed in its ambitions. Baidu’s robots may fail to stun a contemporary audience, but Robin Li is doubling down on a future beyond 2017. In that future, Baidu is not a series of products, but rather an engine that belongs inside everything—an engine that powers Baidu back to dominance in China, and possibly far beyond.
Photo direction by Michelle Le.