Welcome to the ChinAI Newsletter!

Greetings from a dream world where The Farewell won the Oscar for Best Picture tonight...

…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: State Grid Corporation of China: A Hidden Giant in AI

Pop Quiz: Which Chinese company was ranked second on the 2018 Fortune Global 500 List and was the only Chinese company ranked in the top 20 in AI patent applicants, per the World Intellectual Property Organization report? The answer my friend is sitting in the headline, the answer is sitting in the headline. This week’s feature translation, from Shang Li’s article for AIbizweek in August 2019, sheds light on the grid company’s AI strategy

Key Takeaways:

• The number and quality of patents cannot fully reflect the technological strength of a company, and the author acknowledge’s State Grid's patented core technology capabilities are probably not as good as the BAT and “Four Little Dragons of AI” (Sensetime, Megvii, Cloudwalk, Yitu)

• BUT State Grid has massive amounts of data (power grid operations data across entire country as well as life cycle of assets) and application scenarios — power grid demand forecasting, identifying type and severity of grid breakdowns, State Grid’s Zhuzhou subsidiary has used computer vision to identify defects in transmission lines, predictive maintenance of electricity equipment, etc. Recall that DeepMind was in talks to work with the UK’s National Grid though those have ended.

• State Grid also seems to have bought into the AI hype, releasing a white paper on the modestly named “Ubiquitously Powered Internet of Things” (泛在电力物联网) and building an electric power AI brain which claims to integrate image recognition and natural language processing

• Challenges to the diffusion of AI in power grid sector are tied up with the overall low degree of digitization in China’s electricity management system: for instance, there are 4.8 million charging points in China that are unconnected and unmanageable. Any issues with non-digitized equipment and technology results in high costs for upgrading.

• This is a huge domain that has infrastructure-strategic implications: The article states, “State Grid has 540 million smart meters across the country, which is several times the number of camera terminals in the security field.” How much ink is devoted to China’s smart surveillance compared to smart grids? Of course, surveillance brings a bunch of other nasty baggage, but if we’re talking economic and even military effectiveness, there should be a significant shift in how we think about the big levers in diffusion of AI. More precisely surveilling your population with facial recognition will not bring productivity spillovers across the entire economy; optimizing grid management will.

• The article ends with this comparison: “In the past few years, we may have focused more on AI + Internet companies, such as BAT, Xiaomi, etc., Hikvision, Dahua, AI + Finance, AI + Education ...In the future, traditional enterprise giants like the State Grid will increasingly attract attention and enter our horizons. Because the storm of AI-driven upgrading of traditional industries is coming, the arrival of the industrial Internet in the next 20 years may be as violent as that of the Internet/mobile Internet in the past 20 years!”

• Here’s what I wrote in ChinAI #70 about the Industrial Internet: “I like to choose topics where the information arbitrage ratio (relative importance divided by number of people paying attention) is very high. China’s efforts to build an industrial Internet has a very very high information arbitrage ratio.” That point stands and State Grid is a piece of the puzzle.

***The standard we try to live up to every week is to serve you a slice of China’s AI landscape that tastes different than anything you’ll consume anywhere else. I think this week lives up to that standard. If you think so too, and you think that standard is met relatively consistently, but you just haven’t gotten around to subscribing, here’s your chance to subscribe here.***

DISCUSSED IN THE FULL TRANSLATION (in the style of Believer magazine): THE UBIQUITOUSLY POWERED INTERNET OF THINGS (yes this is a thing)

FULL TRANSLATION: State Grid Corporation of China: A Hidden Giant in AI

JeffJots - Dig Deeper into State Grid

If this week’s feature translation piqued your interest, keep reading for the debut “JeffJots” section, which essentially consists of some notes jotted down after digesting a really cool chapter “The Search for High Power in China: State Grid Corporation of China” by Professor Yi-chong Xu (link to edited volume in ChinAI links section):

Her main point: explaining why and how the State Grid Corporation of China (SGCC) -- despite being under state ownership -- successfully engaged in technology innovation in ultra-high-voltage technologies, contradicting long held arguments that government intervention can at best produce lower-end imitation

History

The transmission and distribution business = natural monopoly and the electricity industry around the world was vertically and horizontally integrated until the 1990s

• massive fixed investment requirements + scale economies + non-storable nature of electricity requires precise coordination across networks in real time

In 1990s electricity restructuring was pushed on developing countries by int’l financial institutions; in China the State Power Corporation of China was created out of the Ministry of Electric Power in 1999

• In December 2002, SPCC was unbundled into two grid companies: SGCC (responsible for transmission networks in 26/30 provinces and regions) and China Southern Grid Corp (gets the rest)

SGCC Basics

• Among the top-tier SOEs under direct Party-state control and supervision but complex and diverse in terms of management, employees, and activities

• Among 1.7-1.8m employees, less than half worked at its headquarters and direct subsidiaries; it owns only about 75% of transmission and distribution lines in its service areas, as local governments and privately owned companies own the rest

SGCC success

• One of very few large SOEs to receive an “A” performance ranking from the State-owned Assets Supervision and Administration Commission

• In 2005, SGCC first entered Fortune Global 500 ranking 40th; ten years later, SGCC is world’s largest utility and 2nd largest company in the Fortune Global 500 in annual revenue, just behind Walmart

• “A truly global company,” with investments in hundreds of transmission and distribution projects across Asia and Africa

• Has deployed high-voltage transmission lines to connect China’s entire population, including those living in the most remote villages in the Tibetan plateau

• Most importantly, for Xu’s focus, it has also mastered UHV (ultra-high voltage) technologies to become a global power in this field, even proposing a plan to combine interregional and intercontinental UHV transmission grids with smart grid technologies to address serious energy and environmental challenge -- one that has the endorsement of the International Energy Agency

How did SGCC successfully engage in innovation in UHVs?

• Government policies that incentivized SOE reform and innovation + SGCC’s adoption of an “open innovation” strategy, working with a range of universities, research institutions, and multinational corporations

• SGG’s chief executive was an active policy entrepreneur, proposing to deploy UHV technologies to create a national interconnected T&D (transmission and distribution) network -- had to compete with demands from not just electricity generation and other segments of power industry 

• SGCC’s pitch: mastering UHV technologies was necessary to “take the commanding heights in global competition for the transition to low-carbon electricity” (p. 235).

• 2006 Medium-to Long-Term Plan for the Development of Science and Technology (2006-2020) included SGCC’s proposals for UHV technology in its national priority list

• In combination with 2006 MLP, SASAC adopted measures to assess central SOEs by their ability to build global brand names, global networks, and international standards

• SGCC reorganized talent across five major research institutes, including two focused on UHV technologies (AC and DC), and one focused on automation (Nanjing Automation Research Institute)

• Ramped up investment in R&D - a typical government-funded research institute normally has funding in the tens of millions (RMB), but for a SGCC research institute, it is hundreds of millions (RMB)

• Collaborations with MNCs were essential but not smooth -- multinational giants (Siemens, ABB, Toshiba) dominated the design, core technologies, and manufacturing capacities and they wanted to continue selling SGCC “one-stop” turnkey transmission substations

• Compromises made -- joint research by State Grid and ABB was conducted in China, Sweden, and Switzerland on key components such as thyristor valves

• SGCC initially had no capacity to produce equipment for its UHV projects but to keep its promise to NDRC that at least 80 percent of the equipment would come from domestic sources, SGCC helped electric equipment manufacturers and even acquired two of these manufacturers (Xuji and Pinggao)

The State of Play Now

• Chinese firms now make nearly all equipment for both UHV AC and UHV DC systems -- “even for the few core technologies that multinationals still control, they have to work with Chinese makers because no other countries produce these devices on a commercial scale. China is the only country investing in multiple varieties of UHV projects (AC, DC< and AC-DC synchronized transmission lines).” (p. 244)

• Expanded leadership in standard-setting: In 2019-12, SGCC submitted 14 standards to IEC; 11 have been adopted as int’l standards; in 2013, it provided the secretariat for 7 committees and chaired one

• Organizational Shift to Focus on “Strong and Smart Grids” in 2011 -- SGCC created a Smart Grid Research Institute (has 600 employees), regrouping projects from subsidiaries such as the Nanjing Automation Research Institute. Soon after its establishment, the Smart Grid Research Institute opened a North American branch at Santa Clara, CA, and an European branch in Berlin

• It’s not all hunky-dory: government policies may not always offer a stable environment and there have been persistent calls for the break-up of SGCC; recent reforms in 2014 have empowered government auditors to set the grid’s transmission charges, expands the scope for direct electricity sales from generation companies to large end users, and allows new firms to enter the electricity market 

***The standard we try to live up to every week is to serve you a slice of China’s AI landscape that tastes different than anything you’ll consume anywhere else. I think this week lives up to that standard. If you think so too, and you think that standard is met relatively consistently, but you just haven’t gotten around to subscribing, here’s your chance to subscribe here.***

ChinAI Links (Four to Forward)

Must-read: The Windfall Clause: Distributing the Benefits of AI

New GovAI report! Led by Cullen O’Keefe and coauthors Peter Cihon, Ben Garfinkel, Carrick Flynn, Jade Leung, and Allan Dafoe, “The Windfall Clause is an ex ante commitment by AI firms to donate a significant amount of any eventual extremely large profits. By “extremely large profits,” or “windfall,” we mean profits that a firm could not earn without achieving fundamental, economically transformative breakthroughs in AI capabilities. It is unlikely, but not implausible, that such a windfall could occur; as such, the Windfall Clause is designed to address a set of low-probability future scenarios which, if they come to pass, would be unprecedentedly disruptive.”

I particularly enjoyed the historical precedents section, which compared the Windfall Clause to corporate social responsibility efforts (they estimate that a windfall clause commitment would be only 60% greater than leading corporate philanthropy efforts today), public windfall governance (e.g. redistribution of sovereign wealth funds), and personal philanthropy.

Should-read: The FBI’s China Obsession - Mara Hvistendahl for The Intercept

A sad history of the FBI’s bias against Chinese Americans, told through the lens of Harry Sheng who contributed the best years of his life to US defense work but never held a permanent position in his field after a 1973 visit back to China to visit his sick mother. You cannot read about the decades-long trend of discrimination Mara outlines in this piece and not reflect on current US policy toward ethnic Chinese scientists — both in how these policies are drafted and perhaps more importantly in how they are implemented.

This passage struck me in particular:

A former FBI supervisor whom I’ll call Don Lieu told me that in a routine security check, his unit’s embedded security officer asked him whether he hung Chinese lanterns in his home and whether he was friendly with people who were “in touch with Chinese culture.” (Lieu asked me to identify him by his grandfather’s first name and a family surname, citing concerns about retaliation.) Another time, he said, a security officer brought up the fact that Lieu dated Asian Americans. Lieu, who grew up in the New York City area, said he responded, “I can guarantee that none of these women are foreign nationals. In some cases, they are multigeneration Americans like myself.” He told me that the security officer then questioned how he could be sure that a girlfriend was a U.S. citizen, suggesting that any “close and continuing contact” with a Chinese American woman put the United States at risk.

Check out Mara’s new book The Scientist and the Spy: A True Story of China, the FBI, and Industrial Espionage for more!

Should-read: Policy, Regulation, and Innovation in China’s Electricity and Telecom Industries (2019, ed. Brandt and Rawski)

The edited volume that today’s JeffJots section draws from. Douglas B. Fuller’s chapter on “Growth, Upgrading and Limited Catch-Up in China’s Semiconductor Industry” is fire as well.

Should-listen: WSJ’s The Future of Everything Podcast

I’m very late to this but the WSJ’s Future of Everything podcast is pretty dope — and a fair amount of the recent episodes feature AI and China. Original host and creator of this podcast was Jennifer Strong.

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, Researcher at GovAI/Future of Humanity Institute, and non-resident 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 chinainewsletter@gmail.com or on Twitter at @jjding99

Welcome to the ChinAI Newsletter!

Greetings from Cyberspace where the weary giants of flesh and steel still take roost..

…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: What is AI doing in the midst of the epidemic?

This week’s translation comes from naojiti [脑极体] (a tech media platform based in Tianjin) — h/t to Jordan Schneider for introducing me to this public account. Check out his ChinaEconTalk newsletter, which features translations of Chinese media about tech, business, and political economy.

I think it’s an interesting, if at times overly dramatic and techno-utopian account of what tech companies and scientific research institutes are doing, in AI and related domains, to combat coronavirus. At the very least, it’s better than uncritically consuming Global Times content about the use of drones to warm people from going outside, as well as some reporting from Leiphone on what tech companies are doing to combat coronavirus, which just reads like a list of press release excerpts.

Let’s take naojiti’s good overview of the main areas where AI could help combat the coronavirus as a starting point to separate some of the wheat from the chaff:

• Diagnosis of the coronavirus: the article mentions computer vision to complete early CT scanning work (my initial reaction is that there’s not enough good data to train/validate a model to detect coronavirus in a way that distinguishes it from pneumonia — lancet article for more); nucleic acid detection is a stronger diagnostic test and this field is becoming more and more closely connected with AI in recent years

• New drugs: naojiti reference the notion that AI can increase the success rate of a) finding whether any existing treatments could have a therapeutic effect and b) developing new drugs. Impossible to judge effectiveness at the moment, as developing a vaccine could take more than a year.

• Tracking the spread: spatiotemporal data can help with building infection models and tracing the path of the epidemic. Comments at the bottom of the article call out the importance of Baidu’s data in this respect. From my read of this article and a few others it seems like Baidu is doing the most substantive stuff related to AI.

• Baidu’s role: “Baidu is also opening up its AI technology along with supporting computing resources on the scale of hundreds of millions (RMB) to disease control units, scientific research units, etc. to support a series of anti-epidemic work such as screening and R&D of cures for new diseases such as the novel coronavirus.” The article also says Baidu is working on long-term solutions to identify illegal wildlife trade, in partnership with International Fund for Animal Welfare. Baidu has opened up LinearFold, its RNA prediction algorithm, to research centers around the world, claiming to reduce the prediction time of a virus’s RNA secondary structure from 55 minutes to just 27 seconds. I’m skeptical but don’t know enough to make any judgements. I stared at the article introducing LinearFold for a long time and I don’t think any deep learning is involved — it’s a dynamic programming approach drawing inspiration from computational linguistics.

• Computing power: authors identify computing power as the most central resource. Tencent, Baidu, and the National Supercomputing Center in Shenzhen have all opened up their compute resources for use.

• Smart cameras and facial recognition: The authors write, “integrated applications of widespread smart cameras and facial recognition algorithms can automatically identify people who do not wear masks in public places and discourage them.” I think this stuff is overhyped — Abacus News had a good story about how facial recognition is actually failing with respect to identifying people because masks are covering half their face.

The piece concludes with raising some good food for thought re: the utility of AI in this crisis and future ones:

• Should computing power be organized in “material reserves” that can be called upon? My opinion, not naojiti: many governments are thinking (or should be thinking) about a strategic stockpile of computing power in the context of economic statecraft, but more governments should be thinking about allocation of compute resources in times of public health emergencies, natural disasters, etc.

• Along the same lines, should there be clearer standards for authorizations of sharing and using data in times of crisis?

• Not from this article but I was struck by Professor Hotez’s comments in a Guardian article about the missed opportunity in our response to Sars almost 20 years ago: “What’s so tragic is that once Sars was gone, the investor enthusiasm for a Sars vaccine was zero. If the global health community had followed through and produced and stockpiled a vaccine, something might have been ready to go now.”

FULL TRANSLATION: Open platforms, hundred of millions-level computing power, spatio-temporal data: with the arrival of the epidemic, what is AI doing

ChinAI Links (Four to Forward)

Must-read: A Round-up of Chinese-language coverage of Coronavirus

A compilation of brilliant work by Chinese reporters covering the Coronavirus. H/t to Shen Lu for sharing and volunteers from internetarchive who are now working on translating the headlines.

Should-read: Is China Ready for Intelligent Automation —CSIS China Power Team

A very cool compilation of metrics that assess factors related to China’s automation readiness, drivers of automation, government initiatives related to automation, and the labor force complements necessary for an automated economy —not surprising given the consistently good analysis from the ChinaPower team led by Bonnie Glaser

Should-read: The Secret History of Facial Recognition

Longread by Shaun Raviv for Wired reminds us that everything new is old again — contemporary facial recognition systems in some ways have returned to Woody Bledsoe’s facial-recognition research in the 1960s, which was linked to front companies for the CIA (the aforementioned weary giants of flesh and steel that are still very much welcome among us in Cyberspace).

Should-read: State of AI Report 2019

I should have read Nathan Benaich and Ian Hogarth’s State of AI Report more closely. Regarding UIPath, one of the five leading American companies I highlighted in last week’s issue on robotic process automation, Ian flagged to me that there is a case to be made that UIPath is not a “US company.” It was founded in Romania and raised various first rounds of funding from European investors. Another nice example that challenges the techno-national narrative that packages firms neatly in national containers.

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, Researcher at GovAI/Future of Humanity Institute, and non-resident 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 chinainewsletter@gmail.com or on Twitter at @jjding99

Welcome to the ChinAI Newsletter!

Thanks for reading the latest edition in our series of making AI as boring and unsexy as possible…

…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: China’s Position in the RPA Market

This week we’re taking a peek into the landscape of Robotic Process Automation (RPA) — a process that involves using software to automate repetitive, rule-based processes (e.g. entering data into a form and transferring it to a CRM). The article comes from Yiou Intelligence, a think tank/consulting shop that operates as a “innovation service platform” that connects entrepreneurs, hosts summits, and gets funding from VCs.

Key Takeaways:

• The RPA market is small now but has great potential: RPA software revenues are the fastest growing part of the global enterprise software market, and the Asia-Pacific region’s expected growth rate in RPA of 181% in 2021 will be 3x the global rate. Plus, the integration of AI and RPA technology (think: using natural language processing to automate the writing of reports) will grow the AI market.

• The Chinese market, specifically, has a lot of space to grow: Close to 50% of Chinese companies are bystanders amidst the overall drive toward digital transformation. Total spending on IT in China in 2018 was about 1/5 of that in the United States.

• American firms dominate the RPA market: Yiou compiled an inventory of 40 firms competing in the global market — though there were a good number of Chinese companies on the list, the top 5 companies were all U.S. companies and they accounted for 47% of the global market (UiPath, Automation Anywhere, Blue Prism, NICE, and Pegasystems).

• What will the diffusion pathways of automated systems look like? Many small and medium-sized enterprises may not want to make “earth-shaking changes” to their business processes and are highly sensitive to short-term returns. Thus, RPA may be “their first stop in ‘testing the waters’ of digital transformation.”

FULL TRANSLATION: Five US RPA companies have grabbed half of the global market, does China still have a chance?

ChinAI Links (Four to Forward)

Must-read: The Question of Comparative Advantage in AI

By CSET researchers Andrew Imbrie, Elsa B. Kania, and Lorand Laskai — a superbly well-researched and comprehensive effort to tackle the state of play in AI between the United States and China. I really liked the framework that separated core elements of AI capabilities, critical enablers of AI development, and systemic drivers of national competitiveness in science and technology.

Should-read: The Design and Implementation of XiaoIce, an Empathetic Social Chatbot

Microsoft researchers present the development of Microsoft XiaoIce, the most popular social chatbot in the world. Dives into the design of the entire chatbot system, some evaluation metrics, as well as ethical concerns. Really worth a read — if someone wants to do a more bite-sized summary of the paper, would be happy to feature it in a ChinAI issue.

Should-read: Translation of Qianzhan Chanye Report on China’s AI Industry

In a November issue of ChinAI, I highlighted a cool overview-style 50-page slide deck on China’s AI industry by Qianzhan and covered a few translated slides. The translation team at CSET has translated the full thing (link goes to CSET’s translation page where you can download the translated PPT).

Should-read: Douyin’s 2019 user trends report translated

Katherine Wu has translated Bytedance’s 2019 data report on user behaviors and trends on Douyin (Chinese counterpart to TikTok). Her analysis highlights some of the unique subcultures on the app. H/t to my fellow Iowan Joseph Nelson for sharing this with me - his Des-Moines based company (Roboflow) is doing some cool work on computer vision apps.

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, Researcher at GovAI/Future of Humanity Institute, and non-resident 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 chinainewsletter@gmail.com or on Twitter at @jjding99

Welcome to the ChinAI Newsletter!

An early Happy Chinese New Year to all — if you haven’t called your mom or daughter recently, this week’s piece will really make you want to do that.

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: A Mother Who Lost Her Only Daughter Decides to Maker Her into an AI

This week’s translation, of a piece in人物 (a Chinese magazine magazine that profiles celebrities and contemporary figures), tells the story of Li Yang and her daughter Chen Jin, who passed away due to T-lymphoblastic lymphona at age 14. If there was a way to have your loved ones stay by your side forever, what choice would you make? This week’s piece follows Li Yang as she tries to restore Chen Jin in the form of an AI companion.

Some excerpts follow:

In Li Yang's imagination, she can take AI Chen Jin anywhere: go to the cafe, to Jeju Island to look at the sea, to Australia, where there are lazy koalas and bouncing kangaroos, or Turkey, to take a ride in a hot air balloon to look at the scenery... they can travel together, talk and laugh together and share food, just like before.

According to a research by the Chinese Academy of Social Sciences, China currently has at least one million families whose only child has passed away. According to data from the Ministry of Health, this number is increasing at an annual rate of 76,000.

In September, Alibaba AI Labs received a private letter from Li Yang asking for help: "Hello, I have something that I hope you can help me with. My daughter has died, but I miss her so much. Can I send photos and videos of her to you so that you can make it into software that interacts with me in her likeness? "

On the synthesized recording of Chen Jin’s voice: The recording was of an essay written by Chen Jin, which recounted the story of her going hiking with her mother. When the girl climbed halfway up the mountain, she was exhausted and wanted to give up and go down the mountain. At this moment, "Mom smiled and answered meaningfully:" Child, remember that a famous person once said that success is persistence, and success depends not on the size of your strength but on how long you can persists. As long as you persist, you will be able to climb to the top of the mountain in no time! "After listening to her mother's encouragement, she, “pulled her mother's hand and rushed upwards, her fighting spirit re-ignited. Repeatedly gritting her teeth in persistence, exhausting her body’s chaotic energy. Finally, I finally climbed to the top of the mountain. I was so excited that danced for joy, jumping up and down, just like a general who won the battle.”

FULL TRANSLATION: A Mother Who Lost Her Only Daughter Decides to Make Her Into an AI

ChinAI Links (Four to Forward)

An All-GovAI week of links, featuring some work I didn’t get to cover in previous issues:

Must-read: GovAI 2019 Annual Report

It’s been an incredibly fruitful year from the team here at GovAI. This annual report provides a summary of what we got up to in this past year, expertly compiled by our head of ops and policy engagement, Markus Anderljung. As our Director, Allan Dafoe, writes in his note, “As part of our growth ambitions for the field and GovAI, we are always looking to help new talent get into the field of AI governance, be that through our Governance of AI Fellowship, hiring researchers, finding collaborators, or hosting senior visitors. If you’re interested, visit www.governance.ai for updates on our latest opportunities, or consider reaching out to Markus Anderljung (markus.anderljung@philosophy.ox.ac.uk).”

Should-Read: Who Will Govern AI? Learning from the history of strategic politics in emerging technologies

Jade Leung’s D.Phil thesis examines how the control over previous strategic general purpose technologies – aerospace technology, biotechnology, and cryptography – changed over the technology’s lifecycle. Specifically, she highlights out the relationships among the state, private actors, and researchers has evolved as the technology matured, highlighting key implications for how political dynamics may play out in the AI space.

Should-Read: Near term versus long term AI risk framings

Unpacks the divide between near-term and long-term AI risks into four different dimensions, including: what kinds of technological capabilities issues relate to, the immediate impacts of AI or possible impacts much further into the future, how well-understood or speculative issues are; whether to focus on impacts at all scales or to prioritize those that may be particularly large in scale. Interestingly, they note that proejcts focused on the intermediate scale of AI impacts may be receiving relatively less attention.

This paper by Carina Prunkl, a senior research scholar at FHI, and Jess Whittlestone (Centre for the Study of Existential Risk, Cambridge) was accepted to the AAAI AI Ethics & Society Conference taking place in Feb 2020.

Should-Read: The Offense-Defense Balance of Scientific Knowledge: Does Publishing AI Research Reduce Misuse

Toby Shevlane and Allan Dafoe’s paper, also accepted in the AIES conference, examines publication norms in AI through an offense-defense framework. Crucially, they show that the existing conversation around AI has heavily borrowed concepts and conclusions from one particular field: vulnerability disclosure in computer security, concluding, We caution against AI researchers treating these lessons as immediately applicable. There are important differences between vulnerabilities in software and the types of vulnerabilities exploited by AI. It is therefore important to explore analogies with multiple fields and to consider any properties that may make AI unique. Ultimately, we suggest that the security benefits of openness are likely weaker within AI than in computer security.

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 chinainewsletter@gmail.com or on Twitter at @jjding99

Welcome to the ChinAI Newsletter!

Happy 2020 to all — my vision hasn’t been this good since the Nintendo DS came out. Hope everyone’s holiday season was swell! 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: 大国智造谁来造 (China as a) Major Manufacturing Power — Who Will Do the Manufacturing?)

Context: Last November, a few representatives from the Alibaba ecosystem rang the gong to celebrate the company’s IPO for Hong Kong-based institutional investors. Yuan Wenkai, representing one of those partners (4PX Express logistics warehouse), stood third from the right. A former tally clerk who graduated from a run-of-the-mill Guangdong vocational school, Yuan is now an expert in automation management who increased the sorting capacity of the 4PX logistics warehouse by 20,000 orders per hour.

The core argument in this week’s longform article from jiqizhineng (synced): China still lacks a lot of technical staff like Yuan Wenkai (20 million senior technicans) if it wants to achieve a manufacturing transformation.

Key Takeaways:

• “The quality of the technicians will limit the effectiveness of not just automation technology but even those artificial intelligence technologies that can currently be implemented.” Supporting anecdotes: robotics startups can’t even give away free robots as part of product promos to manufacturing companies because they don’t have the technicians that can implement and debug the robot, let alone compile robot programs. In a shiitake mushroom sorting line, the help of experienced technicians and the factory manager were crucial to improve the algorithm’s recognition rate.

• Regarding China’s efforts to raise its industrial manufacturing level, the US-China binary is worn out. As was the case with a previous ChinAI issue on machine vision in quality inspection in the production chain, this piece frames China’s main competitors as Germany and Japan: On the proportion of senior technicians in the entire industrial workforce: Japan - 40%; Germany - 50%. China - 5%. On the diffusion rate of welding robots: Japan - 70%; Germany - 70%. China - 20-30%.

• Most importantly, this piece contests and complicates the notion of what talent matters most when it comes to realizing the potential of AI technologies, especially if we take the diffusion, maintenance, and large-scale adoption of technology as our focus rather than invention, innovation, and R&D: For instance, the recent CSET report on AI talent (which I’ll heap praise on in four links to forward section) takes a) AI PhD graduates and b) personnel with AI skills who work at AI employers as its proxies for AI talent. To be sure, the Ilya Sutskevers of the world are important but so are the Yuan Wenkais, the Ma Menglis (ChinAI #41) who label ladders at a data annotation company, and the CNC machine tool operators who are not PhD graduates but will play crucial roles in applying machine learning to train machine tools. In fact, in many contexts, they may be more important than the star AI researchers. Based on published catalogues of shortages in skilled trades, this jiqizhineng article claims that for many Chinese companies and regions seeking to leverage robotics and automation technology to transform their manufacturing industry, “the demand for CNC machine tools operators and equipment maintenance electricians ranked ahead of AI engineers.”

• Along the same vein, yes the Tsinghuas and Beidas of the world are important but so is Shenzhen Technology University (China’s first university of applied technology) and technical colleges in Jiangsu and Guangdong that will not appear on the rankings of elite universities: the last section of this week’s translation highlights how Chinese companies, educational institutions, and local governments seek to imitate the German “dual education system,” of vocational education and company apprenticeships.

DISCUSSED IN THE FULL TRANSLATION (in the style of Believer magazine): the first intelligent shiitake mushroom sorting line in China, monthly salaries of full-time Didi drivers, the Lewis Turning Point and other causes of the talent shortage

FULL TRANSLATION: China has a Shortage of as many as 20 Million Senior Technicians. (China as a) Major Manufacturing Power  -- Who Will Make it (大国智造谁来造)?

ChinAI Links (Four to Forward)

Must-listen: Heartland Mainland Podcast

Holly He and Matt Sheehan, of MacroPolo, spent the past year trekking around Iowa to dig into U.S.-China ties at the grassroots. Episode 1 of the podcast looks at the impact Chinese students have had at Iowa's largest universities. A highlight of 2019 was getting to show Holly and Matt around my hometown (Iowa City) and rep my alma mater (University of Iowa). At around the 12:40 mark, you’ll hear a little bit about how I changed my perspective on the challenges faced by Chinese students. Give it a listen and rating!

Should Read: CSET Report on Keeping Top AI Talent in the United States

Remco Zwetsloot, James Dunham, Zachary Arnold, and Tina Huang have published the most comprehensive, data-backed, and careful assessment of the U.S. AI talent landscape to date. The finding that stay rates among international graduates in AI are persistently high is particularly important: “Around 90 percent of international AI PhD students take a job in the United States after graduating, and more than 80 percent stay in the country for at least five years” AND “Stay rates are highest—exceeding 90 percent—among students from Taiwan, India, Iran, and China, and lower—around 75 percent—among students from European countries.” (p. iv)

This has implications for on the weight that U.S. policymakers put on technology transfer concerns, according to the authors:

A prominent 2018 report by the Defense Innovation Unit notes that 25 percent of graduate students in STEM fields are Chinese and that “nearly all [of them] will take their knowledge and skills back to China” because they “do not have visas to remain in the U.S.,” the implication being that U.S. universities are educating the country’s competitors without much benefit to the United States. As this report shows, that is not the case—with the vast majority of Chinese graduate students in fact staying in the United States— despite longstanding efforts by the Chinese government to draw them back.

Should Read: CSET Report on AI Safety, Security, and Stability Among Great Powers

Based on the authors’ own experiences participating in a number of Track 1.5 and Track 2 dialogues involving issues related to AI and U.S.-China relations, Elsa Kania and Andrew Imbrie provide an extremely sensible, thoughtful, and pragmatic roadmap for engagement on AI safety and security among the U.S., China, and Russia.

Specifically, they “present and evaluate several measures in AI safety and security that could prove feasible and mutually beneficial for future bilateral and multilateral interactions. These measures are intended to prevent or correct misperceptions, enhance mutual transparency on policies and capabilities, and contribute to providing safeguards against inadvertent escalation. By pursuing such initiatives in the near term, the United States can improve its capacity to leverage the benefits of AI, while mitigating the risks and managing the shifting terrain in today’s geopolitics, particularly among the United States, China, and Russia.”

Should Read: Alibaba’s Hong Kong Listing Offers Valuable Beijing Goodwill

Josh Horwitz of Reuters gives really useful context for the political reasons behind Alibaba’s strategy of dual listing in Hong Kong. This quote by a former Alibaba senior executive who declined to be named was a particularly insightful nugget: “Investors in the Hong Kong stock market are less influenced by the political atmosphere and have a more objective view of the richness of the Alibaba economy.”

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 chinainewsletter@gmail.com or on Twitter at @jjding99