ChinAI #51: China's AI "National Team"
A "Collective Work Report" on five AI innovation open platforms
|Jeffrey Ding||May 20, 2019|| 6|
Welcome to the ChinAI Newsletter!
These are Jeff Ding's (sometimes) weekly translations of writings on AI policy and strategy from Chinese thinkers. I'll also include general links to all things at the intersection of China and AI. Please share the subscription link if you think this stuff is cool. Here's an archive of all past issues. *Subscribers are welcome to share excerpts from these translations as long as my original translation is cited.
I'm a grad student at the University of Oxford where I'm based at the Center for the Governance of AI, Future of Humanity Institute.
A “Collective Work Report” of the five major “National Team” members — BAT, iFlytek, and Sensetime
In November 2017, the Ministry of Science and Technology (MOST) assigned Baidu (autonomous driving), Alibaba (smart cities), Tencent (medical imaging), and iFlytek (intelligent voice) to lead the development of four national AI open innovation platforms; in September 2018, Sensetime (intelligent vision) was selected as the fifth member of this “national team” (guojiadui). This week’s feature translation is on a Leiphone report from an AI Expo held in Suzhou on May 10th, where each of the companies reported on the progress of their respective platforms (a “collective work report” of sorts). This fits with the Chinese government’s recognition of the importance of openness in spurring research and diffusion in the AI field, related to a previous white paper on AI Open Source Software, which we covered in a previous issue.
I find these “national team” platforms super interesting. Of course, they can tell us about leading Chinese companies in various domains of AI. But the contrast between a “national team” vs. “national champion” model may also be significant. For one, all five were already a) strong, b) independently self-sufficient, and c) hybrid firms backed with a lot of foreign capital before they got this “national platform” assignment. These aren’t your father’s “national champions” propped up with government funding. Some have argued that this “national team” moniker is just a PR move for both sides — MOST bureaucrats and these companies get to share in the glory in helping China advance its AI dream. Relatedly, there’s a lot of competition within each of these company’s “turf” — in Sensetime’s states in its progress report that it will compete in areas (autonomous driving, smart cities) in which Baidu and Alibaba are building national innovation platforms. And everybody’s going after smart cities.
A few tidbits from the translation about each company’s platform progress:
iFlytek: iFLYOS is its open platform, with 920,000 registered developers as of Dec 2018; its speech recognition capabilities now cover 23 dialects (some Chinese dialects are hard problems for speech recognition because there’s not enough data); it claims to have 100,000+ personal voice banks and 90+ customized corporate voice banks on the platform
Sensetime: SenseParrots is its core technology platform, which it positions as a competitor to Berkeley’s Caffe2, Facebook’s PyTorch, and Google’s TensorFlow. We don’t get many statistics on SenseParrots before the report goes on to talk about Sensetime’s forays in augmented reality, autonomous driving, and smart cities.
Baidu: Apollo is its open source platform for autonomous driving, now in its seventh version as Apollo 3.5. There’s a “1 to 3” data exchange principle in which partners open some data to Baidu, and Baidu will open up 3 times of its data to give back. It has a “Dual Hundred Plan” that will invest 10 billion RMB (100 yi) to 100 autonomous driving-related businesses. It highlighted deep collaborations with carmakers Neolix, Ku Wa Saodi Che (酷哇扫地车), and FAW-Hongqi.
Alibaba: Feitian (飞天) is the cloud computing platform for the city brain. I’m not as familiar with this domain but I’m not sure where the “open innovation” happens. My intuition is there’s not really a big open source aspect of city brains and Alibaba is positioning general cloud services as the “open” component, but others should correct me on this. Notable that Alibaba Cloud and its DAMO Academy seem to be taking the lead on this effort.
Tencent: Miying is Tencent’s AI imaging product. Similar to the Alibaba case, I’m not sure what the “open innovation” aspect of Miying is — Tencent reports on its vast repositories of medical knowledge datasets that its feeding into Miying but it’s not clear that these are open to other companies to benefit from. Anyways, report makes clear that Tencent is investing a lot of talent and resources into Miying, which focuses on five major diseases: colorectal, lung, breast, cervical cancer, and various diseases of the fundus [eye].
This Week's ChinAI Links
Tech Scroll Asia, a FT and Nikkei joint newsletter is putting out some good content.
Helen Toner, director of strategy at Georgetown’s Center for Security and Emerging Technology shares her observations from the last few years of talking with AI scientists and policymakers in the US and China on an episode of the Rationally Speaking podcast. She references a review essay that Helen, Remco, and I co-authored in Foreign Affairs: "Beyond the AI Arms Race: America, China, and the Dangers of Zero-Sum Thinking"
I got linked to this podcast from the excellent Chinese Effective Altruism (EA) newsletter, which gives updates (in both Mandarin and English) on AI safety/EA-related issues in China. Also give some love to AI safety researcher Rohin Shah’s Alignment Newsletter (now featuring Chinese translations by Xiaohu Zhu) — alignment refers to the problem of building powerful AI systems that are aligned with their operators.
Richard Lester, MIT’s Associate Provost and Professor Nuclear Science & Engineering, writes a wise and measured letter about MIT’s relationship to China. Many, not all, of these insights could be applied to the US-China relationship.
Thank you for reading and engaging.
Shout out to everyone who is commenting on the translations - idea is to build up a community of people interested in this stuff. You can contact me at firstname.lastname@example.org or on Twitter at @jjding99