ChinAI Newsletter #13: AI安全 - the 安全 translation dilemma --- AI Security/Safety Issue

Welcome to the ChinAI Newsletter!

These are Jeff Ding's weekly translations of writings on AI policy and strategy from Chinese thinkers. I'll also include general links to all things at the intersection of China and AI. Please share the subscription link if you think this stuff is cool. Here's an archive of all past issues.

I'm a grad student at the University of Oxford where I'm the China lead for the Governance of AI Program, Future of Humanity Institute.

The AI 安全 (Security/Safety) Translation Dilemma

This week’s translations are on two articles that explore two different notions of the phrase "AI 安全," which could be translated as AI security in some contexts and AI safety in the context in which groups like the Future of Humanity Institute think of AI safety.
First, a fascinating long interview with the head of Baidu Security, in which Jie Ma discusses the "Long Front" of AI Security

- If you're into details about hackers turned white-hat workers, how Baidu Security is organized into different teams, and some interesting "gossip" about the close-knit networks among key people in China's AI security space, the Google doc is worth a read; otherwise, here are some top-level takeaways:

  • Baidu is prioritizing AI Security and is taking a relatively open, collaborative approach to developing a secure ecology: have hired some of the top talent in the field and their x-lab has acquitted themselves quite well in competitions like Defcon, and are collaborating with Xiaomi/Huawei/manufacturers/through open source methods

  • Baidu's system/ecology-based approach to "AI Security" started with the announcement of the OASES alliance, which is an security alliance for smart terminals, which was jointly initiated by Baidu, Huawei, and China Academy of Information Communications Technology

  • AI Security is a multi-layered, "long-front" problem: vulnerabilities at the bottom layer (e.g. chip vendors and value-added resellers), throughout the operating system (made by many different vendors), and the upper system level (connections to the network, issues with secure transmission and certificate authentication)


Article: Baidu’s “AI Security” Formations in 2018 | Interview with Jie Ma

Dangers of Strong AI and AI Safety - Thoughts from Zhihua Zhou, Nanjing University professor

AI安全 could also be translated as AI Safety, a growing subfield of research looking at safety risks from high-level machine intelligence (e.g. books like Superintelligence)

Professor Zhihua Zhou is a name readers may recognize, as he's the professor who joined a boycott against lethal autonomous weapons, which I covered in the 5th issue of the newsletter (link to archive of all past newsletters at the top).

In this article for the China Computer Federation, Professor Zhou argues that 1) we don't know a viable path to Strong AI, 2) the mainstream AI community has historically not devoted its efforts toward Strong AI, and 3) even if strong AI is possible, it is something that should not be touched  by researchers 

Article: On Strong Artificial Intelligence (Zhou Zhihua)

This Week's ChinAI Links

I hope as we progress through each weekly issue, some of these people, companies, and institutions wo; start to look familiar. For example, the OASES alliance mentioned in the first translation involves Baidu, Huawei, and the China Academy of Information and Communications Technology (co-author of the Tencent book) -- all three have been featured in past newsletters. Thanks for feedback each week and please keep sharing links and comments on the docs!

Our FHI interns doing cool things:
1. Paul's work on the effect of AI technologies on the behavior of authoritarian regimes towards their citizens was used by the US Congress Human Rights Commission for a hearing on Artificial Intelligence and Human Rights.
2. Brian did a Mandarin translation of the OpenAI Charter

NY Times investigation reveals that Facebook gave device partners immense amounts of user data. Check out the ChinAI issue on the Tencent v. Huawei clash over data for more on the emerging dynamics between Internet companies and hardware companies re: user data – this will be an important notion going forward.


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 or on Twitter at @jjding99