ChinAI #58: Making Knives Better & Landscape of China's Intelligent Manufacturing
I'm bringin' non-sexy AI back (yeah)
Welcome to the ChinAI Newsletter!
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 and subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (those who can pay support access to all content for all).
Feature Translation: Bringing Non-Sexy AI Back — The Landscape of Intelligent Manufacturing in China
I know I say this every week, but this is one of the favorite translations we’ve done at ChinAI. With an eye toward how China can close the gap with Germany and Switzerland!!!, it’s a deep dive into the nitty gritty production-floor level details behind slogans like “industrial intelligent transformation” and “smart manufacturing.” The author is jiqizhineng’s co chief-editor utada (宇多田, a pen name and a reference, I think, to the Japanese American singer-songwriter), and many of the insights come from Zhou Tao, the CEO of Shuzhilian — a "data industry chain integrated services” company based in Chengdu, China.
Three meta-points, which correspond to the bolded phrases in the paragraph above:
1) Other countries outside the U.S. and China exist and matter in the world. If you are only reading and following people who see China solely through the lens of competition with the U.S., it’s time to take off the blinders. This piece highlights the gap between China’s industrial manufacturing level with that of Germany, Switzerland, and Japan, specifically as it relates to manufacturing techniques/craftsmanship/workflow (工艺).
2) This piece dives into the unsexy details of key aspects of the production line (anybody care about making knives?) and how under-appreciated aspects of the manufacturing workflow (“how about twenty paragraphs on machine visual quality inspection?”), a nice reprieve from analysis that use “AI,” “chips & other sexy buzzwords of the day to serve as a catchall for anything and everything (see my Twitter rant on how the tech abstraction problem plagues analysis of these concepts).
3) This piece features voices and companies from outside Beijing, Shanghai, Guangzhou, Shenzhen — It draws a lot on insights from Zhou Tao of Shuzhilian (数之联), which works to improve machine visual inspection, a process built on a) cameras and optical sensors to collect data on objects passing through the production line, b) machine learning models that identify the type, location, and size of defects.
THE BIG PICTURE: When the author was in Switzerland, a friend recommended a less-than-$20 peeler knife that “can rapidly dispatch any cutting tool you have used” (see photo at top of the issue) How come China cannot make it? The author writes, “When many people say that they want to chase Germany, Switzerland and Japan, they are not only talking about achieving ‘a higher production efficiency,’ or ‘slightly reduced costs.’ What we have to chase are improvements in 工艺 (techniques) and then quality improvements.”
ONE WAY TO SLICE THE PROBLEM: reducing the amount of defects in production lines that now span over 100 devices. According to Zhou Tao, "The defect rate for an important product on a Chinese leading enterprise’s production line is about 1%, while the defect rate of similar products on German, South Korean and Swiss production lines can be .2 or .3%."
DIG DEEPER: The core problem in knife tool production is the very severe nature of non-standardized data. A common knife has thousands of different pieces and when placed in different production and processing environments, the data that machine visual inspection models are dependent on is also different. The piece makes a distinction between simpler problems (e.g. using a Hall effect sensor to check for wear and tear of large knives) vs. thornier problems which require multi-dimensional data such as torque, vibration, etc. as well as higher-order data models.
WHAT TO LOOK FOR: A few high-end Chinese enterprises have truly carried out (a transition) to so-called digital factories. "For example, companies such as Foxconn, BOE, Tianma, COMAC, SAIC, etc., have production lines, especially in newly built factories, where the entire management and processing workflow have been opened up and linked together, generating multi-dimensional data…all categorized, recorded, and kept in good condition and high quality." The key to more widespread upgrading of the manufacturing chain rests with the mid- and low-end factories.
ChinAI Links (Four to Forward)
In my ramblings last week, I laid out the notion of “info/knowledge arbitrage” — I think the notion applies to highlighting underrepresented voices in China-related analysis. Aside from the notion that we should support more female and people of color voices in this space because equity has intrinsic value, there’s also a more instrumental and calculative reason we should do it. There are systemic factors that make it so good work from, say, female historians are not getting recognition, so the smart people who are looking for an info/knowledge edge should be reading more of their work to take advantage of this market inefficiency. For instance, the best thing I listened to this week was a UPenn Center for the Study of Contemporary China interview with Professor Shelley Rigger on Taiwan and the Global Order.
This week’s must-read is a War on the Rocks piece by Laura Schousboe, a PhD fellow at the Royal Danish Defence College. In line with the “Bringing non-sexy AI back” theme of this week’s issue, she brilliantly outlines the pitfalls (e.g. presentism, “ignoring the boring stuff”) of writing about revolutionary defense technology. H/t to Uhlrike Franke, a summer Governance of AI Fellow at FHI, for pointing me to this piece.
I really enjoyed Episode 43 of Tech Buzz China, a podcast co-hosted by Ying-Ying Lu and Rui Ma, on the gaming live streaming industry in China — not just because I have watched more hours of League of Legends live streaming than I care to disclose but also because it dives deep into the history of Chinese esports and livestremaing platforms such as Douyu.
From this week’s feature translation "In the process of collecting data, although everyone thinks that these data are important assets, they have not thought out clearly where they can be used…(data’s) value is completely determined by the application scenario." Data is another buzzword that falls prey to the tech abstraction problem. Matt Sheehan provides an illuminating five-dimensional framework for understanding data as an input in this MacroPolo analysis.
Lingering Bits and Pieces
*Last week’s translation, with Jordan Schneider, was previously featured in an abridged version done by Jordan for Technode.
Chinese Phrase of the Week: 隔行如隔山 (ge2hang2 ru2 ge2shan1) - someone from a different industry knows no more of the secrets of the craft than they know of another country
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