ChinAI #187: Should Chinese Tech Giants Tear Down their AI Research Institutes? (Part I)
Yuchen Li advocates the affirmative case
Greetings from a world where…
why am I just now getting into Peaky Blinders
…As always, the searchable archive of all past issues is here. 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).
Feature Translation: Why I support the BAT tearing down their “AI Research Institutes”
Context: Over the past year and a half, the AI research labs attached to tech giants have come under fire in both the U.S. and China. In a previous issue, I covered the fizzling out of Alibaba’s AI lab (ChinAI #126). This week’s opinion and commentary piece, authored by Leiphone reporter Yuchen Li (李雨晨) on the AItechtalk platform (AI科技评论), advocates that all Chinese tech giants should dismantle their AI research institutes. Li’s work is well-sourced and insightful, as evidenced by his excellent series on medical AI, which I translated a few months ago. (ChinAI #161)
Key Takeaways: In Li’s mind: “The AI research institutes of the Chinese Internet giants have reached a point where ‘there is no construction without destruction’ (不破不立).
The key theoretical scaffolding for his argument? Different types of organizational structures in business. The concept of an independent AI research institute is very much tied to a functional organizational structure: Image below shows the AI lab situated under the CTO position on the rightmost column; separated from the other three columns, which represent other business lines
Using Facebook AI Research (FAIR) as an example, Li acknowledges the benefits of this model, namely protection from short-term, economic pressures and the ability to conduct pure, fundamental AI research. But Li comes down harshly on the model’s shortcomings:
“Independent AI research institutes understand the latest academic trends and possess algorithmic capabilities, but the separation from business departments makes them lack engineering capabilities and data for core scenarios, and it is easy to become a thing for show (花瓶). Ultimately, this is about amusing oneself with scenarios that are less dependent on data.”
Two noteworthy examples (these seem like industry rumors that I would want to take time to verify before citing, but I think they are illustrative of the organizational point):
Li cites conflict between DeepMind, which is very much nested in Google under a functional model, and YouTube over a joint project to improve YouTube’s recommendation algorithm. Due to disagreements on how much data to share, the collaboration fell apart.
When Andrew Ng was leading Baidu’s AI lab, another group in Baidu’s search division (led by Jia Lei) was also working on voice recognition. Jia Lei’s group won out because their results were verified in engineering practices, whereas Ng’s academic achievements did not provide improvements in application. Apparently, Ng was annoyed that his solution was not widely used in Baidu and forced Jia Lei out.
Li draws compelling parallels between U.S. and Chinese AI labs:
Much of his thinking seems to have been sparked by Facebook’s decision earlier this month to disband its centralized AI department and assign researchers to various product divisions. Turing Award winner Yann LeCun, for instance, now reports to the Facebook Reality Labs (the augmented reality division), instead of FAIR. Jerome Presenti, Facebook’s AI chief, is departing the company at the end of the month.
Li connects Facebook’s situation with a similar reorganization by Tencent in the past: Like Presenti, Tencent’s Vice President Yao Xing (former dean of its AI lab) resigned earlier this year. Like LeCun, Zhang Zhengyou, the director of Tencent’s AI Lab, now has more responsibilities in a product division (robotics, in Zhang’s case).
*A lot more to unpack from this 5,000-word commentary that I plan to give space to in the next issue. Those who can’t wait can read the FULL TRANSLATION: Why I support the BAT tearing down their “AI Research Institutes”
ChinAI Links (Four to Forward)
Louisa Lim, in a longread article for The Guardian, uncovers the legacy of the King of Kowloon, a pioneer of political graffiti in Hong Kong. This piece draws on her book about the King of Kowloon, her PhD on the subject, and a six-part podcast.
Another informative Stanford Center on China’s Economy and Institutions brief on a paper that aims to prove that “access to government-collected data allows China’s facial recognition AI firms to innovate more in both government and commercial applications.”
For VentureBeat, Sharon Goldman’s article helpfully summarizes key issues, questions, and debates over how businesses are trying to measure the impact of AI adoption on their bottom line.
Abstract of an intriguing working paper by Justin Sherman: “This report examines Moscow’s efforts to move Internet governance processes and authorities to the UN’s information and communications technologies agency, the International Telecommunication Union, instead of permitting multistakeholder bodies that include civil society, nonprofits, and corporations to have a major role. This report draws on Russian primary-source documents and media, ITU documents, and other sources to describe and analyze Russia’s strategic view of the Internet, Russia’s historical efforts in the ITU, and Russia’s campaign for its candidate to take over the ITU secretary-general position in September 2022. The report details and analyzes the Russia-China relationship regarding Internet governance developments...”
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 postdoctoral fellow at Stanford's Center for International Security and Cooperation, sponsored by Stanford's Institute for Human-Centered Artificial Intelligence.
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 firstname.lastname@example.org or on Twitter at @jjding99