Feb. 8, 2017
By: Michael Feldman
While US-based firms such as Google, Facebook and Microsoft still dominate the artificial intelligence space, Chinese counterparts like Baidu, Tencent, and Alibaba are quickly catching up, and in some cases, surpassing their US competition. As a consequence, China appears to be on a path to reproduce its success in supercomputing in AI.
As should be apparent to anyone following this space, the technology duo of supercomputing and AI are not unrelated, the most recent example being the triumph of the Libratus poker-playing application over four of the best players in the game. Libratus’s software was developed at Carnegie Mellon University, but schooled at the Pittsburgh Supercomputing Center using the Bridges supercomputer. In fact, Libratus was tapping into Bridges at night during the poker tournament, refining its poker tactics, while the human players slept. In fact, all the technologies discussed below rely on some sort of HPC platform.
But while it’s relatively straightforward, although not necessarily easy, to build supercomputing systems, developing AI software requires more cutting-edge talents. And until a few years ago, much of that talent resided inside US-based companies and universities. No more. In fact, a US government report determined that the number of academic papers published in China that mentioned “deep learning” exceeds the number published by US researchers.
Another visible indication the Chinese are catching up is the number of AI-related patents being submitted there. In an article published last week in Nikkei Asian Review, an analysis showed that Chinese patent applications in this segment rose to 8,410 over the five-year period between 2010 and 2014, represent a 186 percent increase. During that same timeframe, US-sourced AI patent applications reached 15,317, a rate of increase of only 26 percent. The article quotes Shigeoki Hirai, director general at the Japanese government-affiliated New Energy and Industrial Technology Development Organization, who believes the patent growth in China is not only quantitative, but also qualitative. "China's progress is remarkable in hot areas like deep learning," he said. "It's not like they are only growing in numbers."
Last month CNBC reported that venture capital investment in China is being spurred by AI, robotics and the internet-of-things. According to a study by tech auditing firm KPMG, VC investments there will move increasingly into artificial intelligence in 2017. The study noted that venture capital money in China reached a record high of $31 billion last year, despite a global slowdown in VC investment in 2016.
Some of that money is flowing into Chinese startups like iCarbonX, a company specializing in mining medical data and using machine learning analysis to optimize health outcomes. The company, which was founded in 2015 by Jun Wang, has since received a whopping $600 million in investment capital. Wang, who is an alum of Shenzhen-based genomics giant BGI, says he will be able to collect more data and do it much less expensively than US-based rivals working in this area. According to a write-up in Nature, he expects to get data from more than a million people over the next five years. That, he maintains, will allow the algorithms the company is developing to understand how this data correlates with disease states, and be able to dispense advice on lifestyle choices to improve the health of its users.
Other Chinese up-and-comers like iFLYTEK, a firm that focuses on speech and language recognition, and Uisee Technology, a self-driving car company, have also received some notoriety, most recently in a New York Times article. While that report focused primarily on China’s rapidly maturing AI-based military defense capabilities, it noted that much of the technology is freely flowing across borders. As a result, AI knowledge is rapidly assimilated in countries like China because much of that expertise originated with US-based multinationals and the academic community, neither of which hold a particular allegiance to US government interests.
More well-known Chinese firms like Tencent, the country’s biggest provider of Internet services, and Alibaba, the country’s largest e-retailer, are quickly ramping up their AI efforts. Last August, Alibaba announced a new AI suite, dubbed ET, which includes everything from audio transcription and video recognition to financial risk analysis and traffic forecasting. Tencent, meanwhile, has established an AI lab, which while still relatively small (about 30 researchers) by Google standards, represents just the start of the company’s push into this space. In an article published last December in MIT Technology Review, the lab’s director, Xing Yao, said he thinks domestic companies have an advantage in acquiring AI talent. “Chinese companies have a really good chance, because a lot of researchers in machine learning have a Chinese background,” he said. “So from a talent acquisition perspective, we do think there is a good opportunity for these companies to attract that talent.”
To date though, the biggest Chinese success story in artificial intelligence has to be Baidu, which commands the biggest Internet search platform in its homw country. As one of the first firms to recognize the potential of AI technology, it opened a deep learning institute in Silicon Valley in 2013, a move designed to tap into US-based expertise and computing resources. The next year it expanded its investment, to the tune of $300 million dollars, establishing the Silicon Valley AI Lab (SVAIL), which is now one of the premier AI research centers in the world.
Baidu’s pioneering work in speech recognition, with its Deep Speech and Deep Speech 2 platforms, is considered the best in the business and is quickly closing the gap between human transcribers and automated speech recognition. At the same time, the company has moved forward on many other fronts, including autonomous driving, image recognition, ad matching, and language translation (especially Mandarin to English) – many of which are now in production serving its domestic users.
Baidu also recently recently hired Qi Lu, a former Microsoft executive who was at the center of the software maker’s move into AI and bots. Lu is now Baidu’s chief operating officer (COO) tasked with overseeing the company’s business and research operations. According to company founder Robin Li. Lu’s immediate focus will be to work on beefing up Baidu’s search business with AI technologies,. For his part, Li has said he intends to make Baidu a global leader in artificial intelligence and machine learning.
Even given all that, US-based AI is likely to remain dominant for some time. Multinationals like Google, Facebook and Microsoft still have a bigger audience, and thus a bigger data collection pipeline and deployment potential than the largest Chinese web-based companies. But not that much bigger. China’s internet user base is estimated to be in the neighborhood of 800 million people and if these companies can expand elsewhere in Asia or beyond, those numbers could quickly shift. In which case, that Mandarin-to-English translator is going to be especially useful.