I’m on the road this week in NYC and Austin, so today’s post will be a short one and won’t be my usual column-style essay. Instead, it’s a compilation of some recent media appearances elsewhere and a succinct (and hopefully sufficient) analysis of China’s Third Plenum Decision document’s position on generative AI.
WSJ and Odd Lots Podcast
In the Wall Street Journal’s most recent deep dive on AI, I provided some comments on the bull and bear case for generative AI adoption in China. Please check it out here!
I also recorded an episode with Bloomberg’s Odd Lots podcast with Jordan Schneider of ChinaTalk on AI, US-China technology competition, and Trump 2.0. (Note: the recording happened before the assassination attempt on Trump, the Republican National Convention, the Third Plenum, and Biden dropping out of the 2024 election.) Odd Lots is one of my favorite business and finance podcasts, the co-hosts Joe and Tracy are quite a dynamic duo, and I learn a lot from their shows, so I hope you subscribe to their podcast (even if you hate my episode).
Third Plenum on Generative AI
Last week, China held its all-important Third Plenum. Over the weekend, the Decision document was published. You can read the original Chinese version here and the official English translation here, courtesy of Zichen Wang of Pekingnology.
Predictably, the term “artificial intelligence” appeared in multiple sections related to “new quality productive forces” and technology advancement in general. However, the term “generative artificial intelligence” (生成式人工智能) only appeared once in the entire 22,000-characters-long document. And it was shoved into Section 10: "Deepening Reform in the Cultural Sector”, inside subsection 40. Here is the full text of the section in both languages:
(40) Improving the system for comprehensive cyberspace governance
We will deepen reform of the internet management system, combine the functions of online content development and management, and promote integrated management of media communication and online public opinion. We will improve the mechanisms for developing and managing generative artificial intelligence. We will step up the law-based governance of cyberspace, improve the long-term governance mechanisms for the online environment, and refine the system for protecting minors in cyberspace.
（40）健全网络综合治理体系。
深化网络管理体制改革，整合网络内容建设和管理职能，推进新闻宣传和网络舆论一体化管理。完善生成式人工智能发展和管理机制。加强网络空间法治建设，健全网络生态治理长效机制，健全未成年人网络保护工作体系。
This lack of importance and priority placed on generative AI clearly indicates that the Chinese government sees it as a risk and threat to be managed and governed, not a magic pill of mass productivity that should be unleashed everywhere. This ambivalent attitude isn’t terribly surprising for a government that prizes 100% control on public discourse. All LLMs and multi-modal models, by their technical nature, are probabilistic, not deterministic. Thus, you can never predict, test, and prove 100% of the time what a GenAI chatbot would say before it says it.
I made this point in January, when I published the post “Does China Want Generative AI?”. The core argument has held up quite well. Back then, I suspected that the government likely saw GenAI as more “soft tech”, similar to gaming, social media, and EdTech, except even more difficult to control, while the benefits are unclear. Thus, its posture towards this technology is “a giant shrug emoji at best.” Looks like this lukewarm attitude has been enshrined during the Third Plenum.
However, this does not mean the GenAI industry will be regulated away or wiped out like what happened to EdTech back in 2021. If these hard-to-predict foundation models can accelerate progress in the more favored “hard tech” realm – advanced manufacturing, robotics, supply chain management, aviation, any other high-tech sector that leaders can tour and touch – then they are still desirable technologies that will be supported. That is, of course, a big if that many companies and labs are working on trying to answer as we speak. Until we have that answer, what the Third Plenum reveals is a rather nuanced, pragmatic, “wait and see” policy stance towards generative AI.
