ChinAI #248: XiaoIce, where do we go from here?
Part 2 of the history behind the world's most popular social chatbot
Greetings from a world where…
losers are always in the wrong [成王败寇]
…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: XiaoIce Peerless (part 2)
Context: Last week, our deep history on Xiaoice, China’s most popular chatbot, ended with a cliffhanger: how will it survive in a ChatGPT world of large language models? Let’s find out by translating the second half of Caixian Chen and Zibo Dong’s excellent piece on Xiaoice, published in AItechtalk[AI科技评论], a portal under Leiphone.
Key Takeaways: In July 2020, shortly after OpenAI releases GPT-3, Microsoft spins off Xiaoice as an independent company. Li Di, who incubated Xiaoice while he was at Microsoft’s Bing division, will be the CEO. It’s a rocky transition for a variety of reasons, but the most interesting one is on the theme of identity.
Before the spin-off, Xiaoice was special: unique among Microsoft’s global units, it had independent product release rights and its own marketing team. Because it symbolized the rapid rise of Chinese technology, “it had been protected by many Chinese executives at Microsoft.”
The article continues: “But in a situation where it is difficult to distinguish between the main and the external, Xiaoice, which was born in Microsoft China, has always been enveloped in the shadow of Microsoft America. Born in Microsoft and raised in China, Xiaoice’s genes have always been limited by the dilemma of its identity.”
Spinning off Xiaoice may have lessened thorny identity-based issues: “For example, on the one hand, when Xiaoice emerged from Bing, Bing was struggling in the Chinese market and had no advertising revenue, so Li Di was called a ‘patriot’ by Microsoft in the United States; on the other hand, the development of Bing required compliance with both the requirements of China and the United States. Just to persuade Microsoft headquarters to allow Bing to accept supervision from the Chinese government, Microsoft China sent many senior executives to Microsoft headquarters to lobby. Xiaoice was no exception.”
At the end of the day, we can’t ignore the fact that Xiaoice was late to embrace the trend of large models. Chen and Dong report: “Many Chinese scientists who had formerly worked at Microsoft and watched XiaoIce ‘grow up’ lamented to Leiphone, ‘It would have been better if XiaoIce had taken the large model route from the beginning.’”
The article attributes much of this to Li Di’s stubbornness. He has made many “anti-large model” public remarks this year, including the following: “out of 10 Chinese large model companies, 9 will die, 1 will live, or 8 will die, 2 will live.” When Xiaoice was spun-off, its three principal scientists left for other labs, and all of them have made contributions to large models in those settings.
Another notable factor that I had not considered. “After being split from Microsoft, XiaoIce became ‘China's XiaoIce’ in name, but in fact it still maintains an inextricable relationship with Microsoft. Microsoft is Xiaoice’s largest shareholder, and some of Xiaoice’s orders come from Microsoft’s TTS (text-to-speech services). The relationship between XiaoIce and Microsoft is that Microsoft is Xiaoice’s protective shell and also its ‘controlling spell.’ If Xiaoice researches large-scale models, it will compete head-on with its old club.”
What’s next for Xiaoice?
The company has leaned into digital humans and avatars. It leads the “virtual boyfriend and girlfriend” market with 8 million users. As part of this stream, Xiaoice has an “X Eva” platform which hosts digital clones of Internet celebrities to provide chat and companionship services.
XiaoIce has also achieved some impressive results with overseas expansion. For example: “XiaoIce's Japanese version – Rinna – has been ranked among the top ten AIGC (AI-generated content) products in the Japanese App Store for many years.”
At the end of the day, in my humble opinion, not all language AI companies need to be optimizing for productivity-enhancing AI assistants, and there’s some beauty in Xiaoice holding true to its roots of ordinary users building emotional bonds with chatbots.
A lot more I couldn’t cover in the FULL TRANSLATION: XiaoIce Peerless
ChinAI Links (Four to Forward)
Should-read: Chinese University of Hong Kong — Cornerstone of China's Computer Vision Scene (ChinAI #201)
Last Friday, Tang Xiaoou, founder of SenseTime, one of China’s AI giants, passed away due to an unspecified illness. If you want to learn more about his work and life, revisit this ~7,000 word English translation about how he built the Chinese university of Hong Kong into the cornerstone of China’s computer vision scene.
Should-read: CSIS’s Interpret: China Project
I’m very late to discover this great initiative by CSIS to translate Chinese-language articles, speeches, policy documents, and other materials. This link goes to all the translations that come up when you search for “technology” in their library. A lot of good stuff.
Should-read: AI is like… A literature review of AI metaphors and why they matter for policy
Matthijs Maas, for Legal Priorities Project, has written a report that I’ve always wanted to read: This report reviews why and how metaphors matter to both the study and practice of AI governance...The report then provides a non-exhaustive survey of 55 analogies that have been given for AI technology, and some of their policy implications. Finally, it discusses the risks of utilizing unreflexive analogies in AI law and regulation.
Should-read: Best China Books of 2023
The editors of China Books Review has put together an excellent round-up of recommended titles. Mu Ming’s (慕明) The Serpintine Band looks especially intriguing to me.
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 an Assistant Professor of Political Science at George Washington University.
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).
Also! Listen to narrations of the ChinAI Newsletter in podcast format here.
Any suggestions or feedback? Let me know at chinainewsletter@gmail.com or on Twitter at @jjding99