ChinAI #73: Dispatch from the Beijing Academy of AI Conference (BAAI) 2019

Plus, thank you for getting us to 100+ paying subscribers!

Welcome to the ChinAI Newsletter!

We live in a crazy-wonderful world where 100+ people now pay to support the ChinAI newsletter. A good chunk of subscription fees go to supporting contributors’ translations/analysis like this week’s amazing readout of the Beijing Academy of AI Conference which took place earlier this month. As extra incentive for subscriptions: subscribers will get access to a ChinAI library document with all the ChinAI translations grouped by category (like a library!).

As I’ve mentioned in passing in previous issues, last year I struggled with severe neck pain. I don’t think it’s something that will ever completely go away, but I’m grateful for the days now where it doesn’t consume every waking moment, and I’m especially grateful that it made me realize that I couldn’t churn out 10,000 word translations every week anymore, and needed to rely on a community of ChinAI contributors. The payoff is a much more balanced lifestyle and the chance to feature amazing contributions/translations like this week’s.

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).

Feature Translation: Beijing Academy of AI (BAAI) Conference 2019

The BAAI, established in Nov 2018 as an implementing body for the Beijing Zhiyuan Action Plan (a Beijing-specific AI plan), hosted its first global AI summit last week. A total of 1500 people attended the conference, including more than 100 top experts from the U.S., UK, Japan, Canada, Singapore, the Netherlands, and China. A ChinAI contributor who wishes to remain anonymous attended the BAAI Conference last week, and wrote up the following dispatch:


  • BAAI was set up by the Beijing Municipal Science and Technology Commission and Haidian District government, with the support of the Ministry of Science and Technology and leading academic and industry players such as Peking University, Tsinghua, the Chinese Academy of Sciences, Baidu, Xiaomi and Megvii.

  • More analysis on BAAI can be found in Thomas Lehmann’s piece in the recent DigiChina report “AI Policy and China.” ChinAI #52 covered BAAI’s Beijing AI ethics principles.

  • The conference featured a range of technical presentations and conversations, as well as discussions on AI ethics and governance. The write-up presents excerpts from opening session speeches/slides from: Vice Minister of Science and Technology Li Meng and Vice Mayor of Beijing Yin Yong; Huang Tiejun, dean of BAAI Zhiyuan Research Institute, who gave an overview of BAAI’s first year of work; Gao Wen, Director of the Peng Cheng Lab (PCL) in Shenzhen.

MAIN TAKEAWAYS from anonymous ChinAI contributor:

  • Immense efforts are being made by local governments to implement the national New Generation AI Development Plan and address shortcomings in China’s AI ecosystem. The work of BAAI and PCL is ambitious and wide-ranging, bringing together the resources and expertise of leading figures and organizations across academia and industry.

  • To address shortcomings in top talent and fundamental breakthroughs, BAAI aims to provide funding (~USD 71,000 per year) to 300 AI scholars identified over three years. The dean of BAAI said in his conference speech that the youngest of the current scholars is not young enough: “The two largest contributors in the fields of computing and artificial intelligence, Turing and Gödel, made groundbreaking contributions when they were 24 years old. Therefore, we hope that more young students in the future will be able to take the lead and in the future become the backbone of China's artificial intelligence innovation.” As an aside, the UK recently announced a Turing Fellows program that similarly aims to support top AI talent.

  • PCL is trying hard to reduce China’s dependency on American platforms and frameworks. PCL’s CloudBrain aims to provide shared access to an exascale supercomputer combined with an open source framework and machine learning tools. Its Zhihui code hosting site was framed in large part as a response to the risk of the US cutting off access to GitHub. However, I am sceptical that these projects will deliver significant results. As an article by Helen Toner and Lorand Laskai in a recent report by DigiChina said, “The network effects that arise because researchers want to use the same frameworks as their collaborators (and because frameworks with more users are generally better maintained over time) mean it could be increasingly difficult for a Chinese company to come from behind and dethrone established frameworks.” Exchange ‘frameworks’ for ‘code hosting platforms’ here and the same logic applies.

REFLECT FURTHER on BAAI’s and the Shenzhen-based Peng Cheng Lab (PCL) actual value-add — Jeff’s thoughts:

What is BAAI actually contributing? In his speech, the dean of BAAI’s Zhiyuan Research Institute claims that Zhiyuan Institute released the world's largest object detection data set [Objects365]. The ICCV paper that introduces this dataset is a Megvii paper (supervising author is the chief scientist at Megvii). They acknowledge support from the National Key R&D Program of China, Beijing S&T Program, and BAAI. The other two are established funding schemes, but I’m not sure what BAAI is bringing to the table here (maybe just more funding or some sort of coordinating power?). My cynical take is that it’s more of a “PR-based, let the bureaucrats believe they are doing cool AI things” contribution. BAAI claims they are planning 10 joint AI labs with industry (two already established with Megvii and Jingdong). Again what is BAAI contributing to the joint lab? Why wouldn’t Megvii and JD just partner with Tsinghua or a university instead? To be clear, I’m not saying BAAI is not doing good work on the AI ethics and safety side — we’ve covered that in a previous issue — I’m just skeptical of their value-add to leading-edge technical research.

Gao Wen, director of PCL, gives more detail on hardware deficiencies in his slides: Xilinx and Altera have nearly 90% of the FPGA market, NVIDIA have nearly 70% of the global GPU market; Boston Dynamics robot Atlas relies on the company’s huge edge in high-precision sensors and motion control algorithms. It’s not just chips, it’s also high-end sensors (e.g. analog and radio frequency sensors).

PCL is the more interesting body to me: It’s a Provincial Research Lab opened in 2018 by the Guangdong Provincial Government and funded and managed by the Shenzhen Municipal Government. In partnership with the Harbin Institute of Technology (Shenzhen), PCL also cooperates with local industry and research institutions such as the Shenzhen branches of Peking and Tsinghua Universities, the National Supercomputing Center in Shenzhen, Huawei, Tencent, and ZTE.

PCL’s CloudBrain (p. 8-10 of full translation) brings real value-add on through access to compute: CloudBrain 1 is a large-scale cluster system with 100 Petaflops of computing power, including NVIDIA GPUs, Huawei GPUs, and Cambrian AI chips. A machine of 1000 Petaflops will probably be built next year, which can be used by universities, research institutes, and SMEs for training models. The goal of 1000 Petaflows (an exaflop) is generally considered a big milestone for compute over the next few years, which the DOE is heading towards.

Still, I think the importance of these supercomputing benchmarks is often overstated. It often depends on whether the “supercomputer” is defined as a fully integrated supercomputer with low latency or a bunch of distributed compute. CloudBrain seems to fit the latter case: at present, the distributed computing resources connected to CloudBrain 1 include Zhongshan University’s Tianhe-2 supercomputer in Guangzhou, Hefei Brain-Inspired Computing Center’s server cluster (which they spent nearly 100 million RMB on), and PCL’s supercomputing center in Shenzhen.

***The full translation contains the anonymous ChinAI contributor’s notes from the conference, extracts from speeches, and translations of presentation slides:

FULL TRANSLATION: BAAI 2019 Conference Write-Up

ChinAI Links (Four to Forward)

Must-read: The Early History of AI in China (1950s-1980s) by Jieshu Wang

Jieshu, PhD student in the Human and Social Dimensions of Science and Technology Program at ASU, takes us through the tumultuous development of cybernetics and AI in China, including a fascinating section on how Qian Xuesen (father of China’s space program) was convinced that AI was associated with “exceptional human body functions (e.g. ESP, telepathy). Jieshu argues that the establishment of the Chinese Association for Artificial Intelligence in 1981, the first official research institute in China dedicated specifically to AI, was critical to the development of AI in China. The piece is a welcome reminder for us to not get caught up in the presentist view of AI in China.

Should Read: Scary China by Yuan Yang in Chinese Storytellers Newsletter

Reflecting on the development of the “Scary China” narrative, Yuan (FT’s China tech correspondent) writes “We as journalists should take some responsibility for the rise of this meme. As an example, I recall that it was early 2017 when the phrase “AI arms race” started to gain prominence. Although I can’t remember exactly how the phrase entered the collective consciousness of our English-speaking newsrooms, I can guess at the reason we all reached for it: it sounded scary. Scary things are important, and important things are newsworthy.” The question she poses to the Chinese Storytellers group: How do we create narratives that extend our emotional palettes beyond fear?

Should Read: AI Principles: Recommendations on the Ethical Use of Artificial Intelligence by the Department of Defense

Longtime readers of ChinAI know that I’m not the biggest fan of what I lovingly call the military industrial complex. But this Defense Innovation Board (DIB) document on AI principles is worthy of praise, drawing on a 15-month study that involved public listening sessions at major universities. The 74-page supplementary document draws attention to emergent effects (unintended escalation and speed) of AI systems, which are especially important in my opinion. Full disclosure: I participated in a stakeholder interview session, providing input to the DIB on this initiative.

Should Read: Assessing the State of AI R&D in the US, China, and Europe — Part 1: Output Indicators

Stefan Torges provides a clear-eyed comparison between the state of AI R7D in the U.S. and China, finding that for almost all indicators the US institutions are the global leaders. I particularly liked the comparisons between the assumptions and methodologies driving various indicators. Stefan also draws from my testimony before the U.S.-China Economic and Security Review Commission, which provides a systematic framework to assessing the fuzzy concept of “national AI capabilities.”

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 Rhodes Scholar at Oxford, PhD candidate in International Relations, Researcher at GovAI/Future of Humanity Institute, and Research Fellow at the Center for Security and Emerging Technology.

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 or on Twitter at @jjding99