Where Baidu Goes from Here

M.E. Strickland / Contributions by Wang Fei

It was something of a rough start. During Baidu’s Create 2017 conference, while COO Lu Qi was on stage, a live video was brought up of Robin Li riding to the event in a car driving itself on Beijing’s Fifth Ring Road; that is, he was in the passenger seat, while the driver’s seat was empty. It was intended as an attention grabbing demonstration of Baidu’s self-driving technology, and in that, it succeeded. The Beijing police, on the other hand, were not pleased. They promised to investigate what they saw as illegal driving, and jokes at Baidu’s expense followed online.

In all likelihood, the case will blow over. The more pressing question is whether the conference will yield the results the company is hoping for. Baidu’s Create 2017 was touted as an AI developer conference, a showcase of the company’s two latest and ongoing headline projects, DuerOS and the “Apollo program.” The former is a voice interaction system with the goal of weaving together networks of things, the latter a complete system for self-driving cars. Both are forms of AI. And the company’s future now rides on them both.

For some years, China’s tech industry orbited around the BAT, the triumvirate of Baidu, Alibaba, and Tencent, who stood head and shoulders above all other internet companies. Their position has been in many ways even more hegemonic than that of Google, Apple, Facebook, or Amazon in the US. But over the last two years, Baidu has increasingly looked like the odd one out. Its market capitalization is barely a fifth of Tencent’s or Alibaba’s, the latter two now vying with one another for the position of largest company in Asia as Baidu can only look on from a distance. Meanwhile, other competitors have come up from behind, threatening to unseat it from even third place.

Of course, the company’s stock performance is only a symptom of its issues, not a cause. Baidu was founded in 2000 as a search engine and, like Google in the US, its name soon became synonymous with search inside of China. Despite its extension into other services (cloud, video, etc.), online search has remained its core business and identity ever since.

And that is the trouble. To be in search means, in actuality, to be in the advertising business, and Baidu’s revenues therefore rest on internet users’ attention. Its peers Alibaba and Tencent might not seem like direct rivals—most of Alibaba’s business is centered around ecommerce, while Tencent focuses on instant messaging and games—but they have managed to wear away at Baidu’s business nonetheless. Because Alibaba has so totalized ecommerce in China, few people now search for products through Baidu; they go straight to one of Alibaba’s retail platforms instead, and then search for what they want to buy. And Tencent’s flagship app WeChat is not “just” a messaging app. The most widely used app in China by far, it has been compared to a kind of miniature OS of its own, one which now has its own built-in search functions. Either way, more and more users have been led to bypass Baidu in finding what they want.

Then, there was the scandal. Last year, a college student died after taking an ineffective experimental cancer treatment that he had discovered through an ad on Baidu, and in his last days, he laid blame both on the hospital offering the treatment and on Baidu for what amounted to false advertisement. Subsequent investigations found that many of the paid results on Baidu led directly to apparent scams, and so under public outrcy and government scrutiny, Baidu was compelled to revamp its policies and crack down on ads from what might be less than sterling sources. Necessary though it may have been, Baidu’s revenues still took a hit.

On top of that, Baidu has seen some shuffling of its executive team, and the loss of some of its star researchers, most notably AI specialist Andrew Ng (his abrupt, and still unexplained, departure earlier this year temporarily dented the company’s value by $1.5 billion).

But Baidu’s prospects are not so bleak as all that. It is still a multibillion dollar company with considerable assets, and despite much of the bad news above, its share price has been fairly stable over the last year, suggesting there is still some confidence in its future. Nor has it been complacent, either. In fact, Baidu made the decision to move into AI some time ago, and has had an AI and self-driving research lab in Silicon Valley for several years now, calculating that this would be the next big step in the tech industry.

That optimism was on display at the Create conference, Lu, the COO, delivered an energetic and expansive presentation, repeating what has now become almost axiomatic in China’s tech circles: that AI is the next in a series of historical industrial revolutions.

“Baidu is setting AI as the foundation” for all its other endeavors, he said, from its existing search business to its new projects. At one point, Lu grew more animated, declaring that AI is a historical opportunity for China in particular: a chance to lead the world.

Interestingly, Baidu is making both DuerOS and Apollo open source (the projects are already on Github, free for the taking). It would not be an unusual move among US tech companies—Google, most notably, has made its TensorFlow system for neural networks open source—but thus far not many Chinese companies have been so liberal about sharing their code as Baidu.

And that, perhaps, is DuerOS’s real selling point. Lu emphasized that it is a system for conversational interfaces that “can be embedded in any platform to listen, speak, and provide services,” and it is currently equipped with over 100 “skills.” Baidu is providing a simple hardware module, at a cost of little more than two dollars, which will allow any developer to easily integrate DuerOS into an existing speaker, instantly transforming it into a smart home assistant.

Baidu also used the conference to announce that it was acquiring KITT.AI, a natural language processing startup in Seattle that has the support of investor Paul Allen. Baidu already has a cutting-edge speech recognition system of its own (one which, it proudly notes, made the breakthrough of achieving a human level of accuracy even before Microsoft’s AI achieved the same feat). The addition of KITT.AI should therefore reinforce what are already strong NLP capabilities in the company.

Then, there is Apollo. Self-driving cars are not the most logical venture for a company than was born in the search business, Google’s own example notwithstanding. Even for companies that have far more experience with hardware, it is a difficult area to crack. Yet Baidu made clear it is not trying to become a car manufacturer itself. Rather, it has assembled a roster of more than 50 partners, spanning traditional carmakers, parts suppliers, and even tech companies like Microsoft and NVIDIA to collaborate on its Apollo system.

Here again, Baidu means to take the open source route, banking on the hope that it will lead to wider adoption, and that that will in turn foster quicker development through shared research and resources.

Lu likes to compare Apollo to Android: “Apollo is the Android of cars, but even more open, and more capable.”

When Robin Li at last took the stage, he sought to draw in the developer audience with the assurance that they were all partners, not competitors. Li believes that the mobile era is passing. There has not been much in the way of groundbreaking innovation in smartphones in recent years, leaving developers with little room in which to make their mark. And that, perhaps, is just one stage in an even larger shift in tech.

“The developers of the PC era were a little like cowboys, they were lonesome heroes. One person could shake up the world.” That, apparently, is no longer the case, and Li seems to believe that any further innovation is likely to be crowdsourced (or at least crowd-tested).

Baidu has a good shot at making both DuerOS and Apollo succeed. It has been hard at work on both projects for some time, it has a number of successes and technical achievements in AI already racked up, and both its acquisitions and open sourcing are strategically sound moves.

But it is all still a gamble. With DuerOS, for example, Baidu faces competition from its already overgrown rivals, with both Alibaba and Tencent having recently announced their own home smart speakers. Alibaba’s, dubbed the Genie, was unveiled just this past week. Each of the BATs, therefore, will have its own smart home system on the market in short order. How they will fare against one another is difficult to gauge. Alibaba and Tencent seem to have more discrete products in mind, while DuerOS aims to be a more general purpose IoT operating system, and its open source nature gives it the potential for wider adoption.

This is a venture that both Alibaba and Tencent could likely afford to lose. For Baidu, however, the stakes are higher. As its core search business is being eroded by powerful rivals and inescapable changes in how users and advertisers operate online, it needs for at least one of the two projects, either DuerOS or Apollo, to take root and claim a sizable share of its prospective market. The Create conference was thus not just another product launch, but a public signal of the company’s determination to fundamentally reorient itself. Baidu’s standing has already slipped. If it is to survive in the longer term, much less recover its former influence, it will need to transform.