ChinAI #160: Sensetime's listing: a precautionary measure?

Plus, more plagiarism drama in Chinese CS Academia

Greetings from a world where…

Shohei Ohtani had the most extraordinary — if not greatest — season in baseball history

…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: A Theory For Why Sensetime Chose Now to Go Public

CONTEXT: Back in August, Sensetime, China’s largest AI start-up, filed to list on the Hong Kong exchange (ChinAI #154). This piece (original Mandarin) provides more context for Sensetime’s decision to list, including an intriguing explanation for why the choice could be seen as a precautionary measure. It’s written by Yang Jingzi for 放大灯 (enlarging light), a S&T media platform under guokr, a popular Chinese S&T education community.

KEY TAKEAWAYS:

  • Facial recognition has become an indispensable part of daily life in China, as evidenced by the emergence of the four computer vision dragons — Sensetime, Megvii, Yitu, and Cloudwalk. The article also points out that the price of a face photo on the black market is .002 RMB.

  • Why did Sensetime choose now to go public? The piece first debunks two possible explanations: a) Sensetime is short of funding; b) it needs to raise its popularity/name recognition. On the first factor, Sensetime’s losses are overestimated because Hong Kong’s financial disclosure standards differ from the mainland, which means direct comparisons between the four CV dragons are futile. Sensetime also has a healthy amount of cash on hand. Nor is Sensetime in need of a reputation boost.

Instead, the piece argues that Sensetime chose to list now because it felt pressured by structural trends in China’s AI ecosystem: “The time to go public is fleeting, and Sensetime didn’t dare to miss it.” Here’s the supporting evidence:

  • Investment hotspots have shifted away from computer vision and AI toward healthcare and advanced manufacturing. According to IT Juzi data, financing events in computer vision peaked in 2015 , and there’s been a sharp decline since (graph below). *Note the caption on this graph says “computer data” financing rounds (2013-2021), but I read it as a typo for “computer vision.” The Chinese word for data (shuju) and for vision (shijue) are very similar.

  • The policy environment around facial recognition is also tightening. Yang cites legal cases that have raised public awareness about protecting facial information, as well as Tsinghua University Law School professor Lao Dongyan’s comments about the risks of facial recognition (see ChinAI #77 for a translation of some of her critiques). China’s Personal Information Protection Law will also come into force in November, raising the level of scrutiny on facial image collection and analysis.

Without reading the minds of Sensetime’s directors, we’ll never know if the decision to list was a case of “now or never.” Still, this article makes a convincing case that Sensetime’s decision can be analyzed in the context of overall trends in China’s computer vision industry.

ChinAI Links (Four to Forward)

Should-read: Three Cases of Academic Misconduct (in Mandarin)

Jiqizhixin reports on three cases of academic misconduct from the past couple months:

1) A paper published at the 2021 International Conference on Computer Vision, co-authored by graduate students at Hong Kong University of Science and Technology and Nankai University, plagiarized a paper published at the 2021 International Conference on Machine Learning. See more details in English here.

2) Three students, including a Beijing Institute of Technology master’s student, copied a NeurIPS 2020 paper submission word-for-word and uploaded it to arxiv. This news event became the No.1 hot topic in Zhihu Science.

3) Latest case to blow up: Two researchers from the Shanghai Key Lab of Intelligent Information Processing, and School of Computer Science, at Fudan University (an elite Chinese university), have been accused of plagiarism. Their 2017 paper published in the Chinese journal Computer Applications and Software bears striking similarities with a 2008 paper by IBM and University of Michigan researchers.

Should-read: White Paper on Trustworthy Artificial Intelligence

Let me plug again this CSET translation, co-authored by China Academy of Information and Communications Technology and JD.com, because I recommended it before getting a chance to take notes on it. Here’s a few points that caught my eye:

  • “AIIA (China’s AI Industry Alliance) also released the first batch of commercial AI system trustworthiness assessment results in 2020, involving 16 AI systems of 11 enterprises, providing an important reference for user selection; among the latest AI legislation proposals issued by the EU is a proposal that an authoritative third-party organization carry out trustworthiness assessments and other such measures.” *Would love to see those assessment results.

  • “Research on trustworthy artificial general intelligence (AGI) must be laid out in advance. At present, whether it is AI governance or trustworthy AI, most work is carried out for weak AI technology and applications. AGI and even superintelligence have not garnered sufficient attention. Once these emerge, they will be major events tied to the destiny of humankind and require a forward-looking layout, such as exploring the development path of AGI through the development of cutting-edge technologies such as super deep learning (超级深度学 习) and quantum machine learning. At the same time, we must also carry out research related to trustworthiness when exploring strong AI.” *Important section about AGI.

Should-read: Yiqin Fu’s newly discovered Chinese-language resources

Includes an interesting anecdote about Tencent’s text-to-speech capabilities.

Should-listen: Four years on Substack: A conversation with Bill Bishop, Substacker #1

To mark the four-year anniversary of the first-ever Substack publication, Bill Bishop’s newsletter about China, Hamish McKenzie of Substack interviewed Bill about his story:

I had known Bill for almost a decade from my previous life as a reporter and was a regular reader of Sinocism. Around the time that we came up with the idea for Substack, Bill had been telling his readers that he was planning to introduce a paywall for the newsletter. I jumped into his inbox and suggested that he be Substack’s first publisher. Happily, he agreed! 

Chris and I promptly flew to Washington, D.C., where Bill had recently relocated after 10 years in Beijing, and started figuring out how we could build the first version of the product around his needs. By October 2017, Bill was ready to launch, and on the 15th of the month he enabled paid subscriptions. By the end of that day, he had brought in six figures of revenue, heralding the arrival of two businesses at once: his own, and Substack’s. 

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 chinainewsletter@gmail.com or on Twitter at @jjding99