ChinAI #94: Cloudwalk -- A "National Team" Member Unlike Any Other
Breaking down one of China's "Four CV Dragons"
|May 18, 2020||2|
Greetings from a land that is always in the process of becoming…
…as always, the archive of all past issues is here and please please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay support access for all AND compensation for awesome ChinAI contributors).
Full Translation: Respect the Latecomer Wave — Cloudwalk Receives 1.8 billion RMB in financing from national funds
Context: Cloudwalk, one of the big four Chinese computer vision startups (Megvii, Sensetime, and Yitu are the others), raised an additional 1.8 billion RMB in financing this past week, putting its valuation at more than 25 billion RMB. This piece by Wu Xin for jiqizhixin (Synced) examines why Cloudwalk is “a member of the national team unlike the others (与众不同国家队).” We discussed the “National Team” concept in a past ChinAI issue; this article shows how there’s important distinctions among the national team members themselves in terms of ties to the state.
Much of Cloudwalk’s financing this round came from government funds such as the China Internet Investment Fund (established by the Central Cyberspace Affairs Commission and the Ministry of Finance in China)
It’s not just about facial recognition and surveillance: per jiqizhixin, Cloudwalk is the largest AI supplier in China’s financial industry (serving banks including three of the “big four” state-owned commercial banks of China) — with nearly half of the market share. This involves facial recognition for identity authentication, facial scans to pay, precision marketing, etc. Part of a broader “smart business” vertical which has become the largest source of revenue for Cloudwalk in the past two years.
How is it unlike the other “national team” members? It was incubated in the Chongqing Research Institute of the Chinese Academy of Sciences, participated in a national strategic guiding technology (category A) project in which it was responsible for face recognition research and application, plus “it is also the only AI unicorn that has been invited to formulate national-level standards, public security bureau standards, and industry standards in facial recognition.”
Why is it unlike the other members of the “national team?” Short answer comes from the preface of the article: “With the advent of the “new infrastructure” era, “purely domestically funded enterprises (纯内资)” are expected to stand out from the competition.” Also: “In the opinion of some people in industry circles, the "purely domestically funded" company style retains more national autonomy for original AI technology and massive amounts data, and also provides a protective screen for some security and financial projects that involve national security.”
Where companies list give you some sense of their genetics: Wu Xin writes, “There are rumors that Cloudwalk will be listed on the Sci-Tech innovation board (STAR Market), and the listing time may be at the end of 2020.” This Star Market, a pet project of President Xi Jinping’s, is meant to encourage investment in domestic tech companies and make it easier for mainland investors to trade in these companies, “after complaints that Chinese mega stars like Alibaba, parent company of the South China Morning Post, chose to list in the US rather than at home.” See this SCMP article for more on the STAR Market.
Reflections: Mark Cohen on trajectories
I got a lot of good feedback from folks after last week’s musings on trajectories and U.S.-China relations. I want to start featuring more reflections from ChinAI readers — especially welcome those that directly rebut/disagree with my own. This week, Mark Cohen, Distinguished Senior Fellow and Director of the Berkeley Center for Law and Technology, offers his thoughts on discursive bandwagoning in the context of intellectual property issues. One of the points he emphasizes — how China’s lack of transparency fuels these trajectories — is really important and was lacking from my post last week. Mark runs the excellent ChinaIPR blog and previously served as Senior Counsel for the U.S. Patent and Trademark Office. Thanks to Mark for offering his views (very lightly edited by me):
Excellent article on discursive bandwagoning. I would probably go a bit further into how pernicious and self-destructive this type of "engagement" has become, be a bit less harsh as there are many well-meaning people who just don't have an adequate understanding of what is going on, recognize that some of the narratives can be accurate — albeit overbroad and perhaps harmful — and also recognize that we may all have become victims of it and that there can be competing contradictory bandwagon memes.
This phenomenon can be tracked not only in social media but also especially in trade/tech contexts through business surveys (US-China Business Council, American Chambers of Commerce/EU Chamber business surveys, etc). Generally, I think these surveys show a desire to support whatever priorities the administration has identified. They also create a vicious cycle of feeding into the meme - for example, forced tech transfer or trade secret theft, were rarely important issues before the administration identified them, and they they became important memes as the administration pushed forward with retaliation/trade war, and now they have sunk back down to a lower category. See The Trump Administration and China IP Diplomacy: Old Wine In a New Bottle?
Moreover terms evolve. "IP Theft" was an ill-defined concept and originally meant pervasively tolerated trademark and copyright infringement. It later came to include trade secret theft, and later cyber intrusions/market barriers and ultimately also included patent infringement - despite various legal and definitional impediments to using it in those ways (e.g., there is no criminal remedy in most countries for patent infringement). These terms tend to expand and contract over time. I have described these repeating, ill-defined memes as an "echo chamber." See: https://chinaipr.com/2019/05/12/the-600-billion-dollar-china-ip-echo-chamber/
The danger has always been however that broad-brushed approaches lose nuance, are not a useful substitute for well-conceived strategies, and tend to exhaust the public's attention with a hyperbolic perspective. We also victimize ourselves and deprive ourselves of meaningful avenues to make progress: "collaboration" vs "confrontation" or "decoupling" - when did this become a binary? Why don't we just do what suits our national interest based on prevailing circumstances? Many academics are also thinking along these lines, and they risk their insights becoming compromised by false choices. See: https://chinaipr.com/2019/07/17/collaboration-or-confrontation-beyond-the-false-dichotomy-in-us-china-ip-relations/
I also think part of the fault for this rests with China. Lack of transparency and lack of confidence in Chinese data have often created suspicion. China's trumpeting of its own data has led to the outright rejection of any data by foreigners, and reliance by the foreign community on qualitative (anecdotal) information in lieu of ascertaining reliable data. The result is that China has also contributed to the impoverishment of discourse around it, or to more polarized perspectives. It is not surprising that the Section 301 Report that launched US-China trade war similarly is a data-free zone. (From my testimony before the U.S.-China Congressional Commission: "In seeking to address the impact of Chinese industrial policies on protection of IP, I believe that we should increasingly utilize big data type analyses, which are also left out of the 301 Report.”
I can't speak for the AI issues, but this has certainly been my experience on IP issues. At the US Embassy, some Chinese friends said I was the "the mouse in the accordion, receiving pressure from both sides" 风箱里的老鼠，两边受气. That is probably the right place to be.
ChinAI Links (Four to Forward)
Must-read: A Guide to Writing the NeurIPS Impact Statement — by Center for the Governance of AI
The NeurIPS conference, a premier venue for machine learning research, now requires submissions to include a statement of the “potential broader impact of their work, including its ethical aspects and future societal consequences.” This (unofficial guide) provides some helpful suggestions on how to write the statement. By a team of GovAI researchers and machine learning researchers (Carolyn Ashurst, Markus Anderljung, Carina Prunkl, Jan Leike, Yarin Gal, Toby Shevlane, Allan Dafoe)
Should-read: The AI Powered State: China's approach to public sector innovation — Nesta Essay Collection
China is innovating with AI in public services at breathtaking speed. Nesta’s new essay collection explores how China is using AI in public services, with a focus on practical applications and the ethics of AI - and reflects on what policymakers in other parts of the world can learn from China’s experience. My piece in the collection, titled “Promoting Nationally, Acting Locally: China’s Next Generation AI approach,” highlights the emergence of AI ecosystems in Hangzhou and Hefei. Thanks to Hessy Elliott convening and organizing this collection.
Xu Xu, a PhD candidate at PSU, in the American Journal of Political Science (pay-walled) leverages variation in the implementation of China’s Golden Shield project to study the consequences of digital surveillance in dictatorships. From the abstract:
I first develop an informational theory of repression and co-optation. I argue that digital surveillance resolves dictators’ information problem of not knowing individual citizens’ true anti-regime sentiments. By identifying radical opponents, digital surveillance enables dictators to substitute targeted repression for nonexclusive co-optation to forestall coordinated uprisings. My theory implies that as digital surveillance technologies advance, we should observe a rise in targeted repression and a decline in universal redistribution. Using a difference-in-differences design that exploits temporal variation in digital surveillance systems among Chinese counties, I find that surveillance increases local governments’ public security expenditure and arrests of political activists but decreases public goods provision. My theory and evidence suggest that improvements in governments’ information make citizens worse off in dictatorships.
Should-attend: Two Webinars this Week
GovAI is launching our Governance and Economics of AI webinar series on May 20th 1700-1815 BST (0900-1015 PT, 1200-1315 ET. The first one features Joseph Stiglitz, Diane Coyle, and Daron Acemoğlu in a discussion about COVID-19 and the economics of AI. Registration here. More information about this and future events here.
I’ll be doing a webinar this Thursday May 21 (1000-1100 PT, 1300-1400 ET) with the Tech Buzz China team. The pitch: come for discussion about the “unsexy” aspects of China’s AI landscape and basically anything folks want to ask about during the Q&A. Registration here. More information about this and the awesome work that Tech Buzz China is doing here.
Thank you for reading and engaging.
These are Jeff Ding's (sometimes) weekly translations of Chinese-language musings on AI and related topics. Jeff is a PhD candidate in International Relations at the University of Oxford and a researcher at the Center for the Governance of AI at Oxford’s Future of Humanity Institute.
Check out the archive of all past issues here & please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay for a subscription will support access for all).
Any suggestions or feedback? Let me know at email@example.com or on Twitter at @jjding99