Robin Li, Baidu’s billionaire founder Image: Hou Yu/ CNSPhoto/ VCG / VCG via Getty Image
Chinese search-engine operator Baidu, dubbed the Google of China, is betting big that smart machines will disrupt industry after industry. Of the $3 billion Baidu spent in research and development (R&D) over the past two-and-a-half years, the majority has gone to artificial intelligence (AI), says Wang Haifeng, the company’s vice president and head of the AI group.
In the past quarter alone, R&D expenses went up by 35 percent from a year ago, to $412 million, according to Baidu’s latest financial results. Like Google, Baidu wants to use the technology to refine its search algorithms, develop voice assistants, produce self-driving cars and build augmented-reality tools that may soon have broader applications in marketing, tourism and health care.
To Baidu, however, AI means more than frontier research. The company needs AI-enabled services to bring in new revenue, as its core online-advertising revenue faces mounting pressure from government restrictions, particularly in health care ads. What’s more, advertisers are flocking to ecommerce and social media platforms such as Tencent’s messaging app WeChat and Alibaba’s Taobao marketplace. According to consultancy eMarketer, Alibaba accounted for 29 percent of China’s $42 billion digital-advertising market in 2016, with Baidu’s share falling to 21 percent from 28 percent in 2015.
In the past quarter, Baidu’s operating profit declined by 9 percent to $291 million, after going down by 14 percent last year. Its other diversification efforts have produced mixed results. The company invested billions of dollars in online-to-offline services a few years ago, betting that it would earn healthy commission fees from cinemas, restaurants and other local services using its products to connect to nearby customers. But that plan didn’t gain much traction, as the Tencent-backed Meituan Dianping now holds 80 percent of this $100 billion-market, according to Beijing consultancy Analysys International. IQiyi, its online-video unit that just inked a streaming partnership with Netflix, remains popular but is facing rising content costs.
Baidu has amassed a 1,700-member AI team and built four research labs in China and Silicon Valley, with the second Silicon Valley lab ready to accommodate 150 scientists, the company announced in March. In January, it hired Lu Qi, a Microsoft veteran, as its president and chief operating officer overseeing Baidu’s AI research. The company has one advantage: Its online search app boasts 665 million monthly active users, whose behavioural patterns can be harnessed for deep learning, a branch of AI that teaches machines to learn by themselves. And because search encompasses so many different topics, Baidu has more diversified data tranches than other Chinese tech giants, says Wang Shengjin, a professor at China’s prestigious Tsinghua University.
Baidu has amassed a 1,700-member AI team and built four research labs in China and Silicon Valley
And there has been real progress in this area. The company, which claims a 97 percent accuracy rate in voice recognition and 99.7 percent rate in facial recognition, has won many international competitions in speech-text interaction with its DuerOS platform, including one where its software achieved a typing speed three times as fast as humans’. It has also been appointed by China’s National Development & Reform Commission to lead a national AI lab, as Beijing seeks to transform from a manufacturing economy to one driven by advanced technologies.
In perhaps the clearest articulation of Baidu’s ambition, its billionaire founder, Robin Li, said at last year’s World Internet Conference in Wuzhen that, “The age of the mobile internet is over. Future opportunities lie in artificial intelligence.”
But not everyone is convinced of that prospect. In voice recognition, which is the most commercialised aspect of AI, other tech giants are catching up. Earlier this month, Tencent launched Dingdang, a voice-interaction platform that is a direct competitor of Baidu’s DuerOS. DuerOS now counts speaker manufacturer Harman as well as smartphone companies Lenovo, Xiaomi and Vivo among its customers, according to Baidu.
When it comes to driverless cars, commercialisation is still decades away, partly because all relevant players need to collect far more data, says Yang Qiang, a professor at the Hong Kong University of Science & Technology. “Current data collected for autonomous driving isn’t reliable enough,” according to Yang. “And we have to figure out how to use those cars. Is it in car-sharing or logistics—or something else? We have no idea yet.”
“ The age of the mobile internet is over. Future opportunities lie in artificial intelligence.
That means Baidu’s biggest advantage is still in online search, since it has plenty of related data to write better algorithms, Yang says. In addition to adding voice and image search functions, Baidu is using AI to develop personalised news feeds in its mobile search app, which Nomura Securities analyst Shi Jialong estimates will bring 6 billion yuan (nearly $900 million) in revenues in fiscal year 2017.
Baidu also faces talent competition. After the company’s chief scientist, Andrew Ng, resigned in March, Zhang Tong, director of Baidu’s big data lab, was poached by Tencent. In the same month, Baidu’s senior vice president, Wang Jing, resigned to build his own autonomous driving startup. Baidu’s Wang acknowledges the fierce battle for brains: “AI represents an enormous opportunity—but to seize this opportunity, we need access to the best R&D talents in China and globally,” he said. “There is fierce competition throughout the industry to both find and retain these talents.”
But the commercialisation of AI is still in its very early stages, and Baidu may still stand out in the end, HSBC analyst Chi Tsang says. “Any market where there are tremendous amounts of data and the inability to process data is applicable to AI,” Tsang said. “The building blocks of AI that Baidu has can enable them to enter a lot more markets other than advertising.”