ChinAI Newsletter #25: Tencent - A Complete Takedown and Rethinking of China's so-called AI giant
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. *Subscribers are welcome to share excerpts from these translations as long as my original translation is cited.
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.
Tencent - A Total Rethinking
To celebrate the half-year birthday of ChinAI, we’ve got one of my favorite translations we’ve ever featured: a 10,000+ word essay on Tencent, the one Chinese tech company with that one Chinese app that probably every citizen-of-the-world, Economist-reading Westerner can name, but one that very few of us (including me) actually know much about. Thanks for sticking with ChinAI, and if you’ve found it useful, could you please take 5 minutes this week and share the subscription link with 5 friends?
Three years ago, in August of 2015, Connie Chan (now a general partner at Andreesen Horowitz) wrote a piece “When One App Rules Them All” that explained how Tencent’s WeChat actually works. Now, in August of 2018, an elder in China’s investment community， Li Guofei (李国飞) has written a piece titled “A Total Rethinking of Tencent’s Strategy” [全面反思腾讯的战略] that explains how Tencent itself works and doesn’t work through the perspective of the inner workings of its data and algorithms, in comparison to some of its main competitors in China. In contrast to my usual approach of just giving quick takeaways from the translation, because this piece goes through Tencent’s development history and references a lot of apps and competitors that readers may not be the most familiar with, what follows is a lengthier section with more background context and synthesis of the translation:
Background and Context:
Who is Li Guofei (李国飞) and why do his thoughts matter: Li Guofei has been an investor in China’s investment community for more than 20 years. He previously worked as a fund manager at Penghua Fund, a privately-owned asset management company – one of China’s main asset managers. He left Penghua Fund in 2002. Aside from this, I haven’t found much about him. Apparently he pops up every now and then to give talks at workshops and universities (he gave a presentation at Peking University in 2010 on investing that was very well received), and many of his writings on his WeChat Accountget a lot of coverage. Per that account, this article was written over the course of over three months, with the help of an assistant named Shen Wenfeng. Influence: it seems like each one of his articles make a huge splash and this one on Tencent has spread like wildfire. For the xinzhiyuan (AI Era) article which gives a little context before publishing Li’s full essay, it’s been viewed 34,000+ times in the past week, and this is one of many portals that have shared the essay in full to get page views. A Google search of the exact match for the essays’ title (“全面反思腾讯的战略”) gives 17,800 results. This essay was published August 13. Also, as you’ll see in the full translation, he will casually mention friends at Tencent who give him inside details, so this is a level of access that few get.Comes out right as Tencent is going through a Tough Spell: Multiple indicators (second-quarter profit decline, slowest year-on-year profit growth since 2012, market value loss of $150b since January); Decline in revenues from games, which contribute to nearly 40% of Tencent’s total revenue, can be attributed to a sector-wide freeze in new game licenses as well as Tencent not getting regulatory approval to charge for its PlayerUnknowns’ Battlegrounds (PUBG) game, which has >400 million players worldwide.
More so than other translations from past weeks, this essay is chock-full of allusions to past events and apps/companies, here’s a few this about Tencent you should be familiar with before reading:“AI in All”: Tencent’s strategy to invest in AI, which can easily be put up against Baidu’s “All in AI”: key elements are its establishment of an AI Lab, the Youtu Lab which has gotten some good results in facial recognition, a Go playing-machine called “FineArt,” and an AI medical imaging product called Miying3Q War: This is a reference to a dispute between Qihoo 360 and Tencent that occurred from 2010-2014: the 3Qs reference an accusation by Qihoo against Tencent's QQ chat app over the latter scanning users' computers and collecting information; Tencent fought back by not allowing users to log in to QQ if they had Qihoo's web security app installed. Qihoo sued Tencent and the case made it up to the Supreme Court, which ruled in favor of Tencent, on the grounds that it company did not have a monopoly on the market. The war was a loss-loss for both sides though and Tencent suffered in the public eye over perceptions it was abusing its power.Key Aspects of the Tencent Empire:WeChat (the app to rule them all); WeChat Pay (mobile payment service); WeChat Applets (mini-programs that sit within WeChat so you don’t have to install, exit, and enter different apps, e.g. Google’s Caihua Xiaoge mentioned in issue #21)Tencent Video (video streaming); Weishi (short video/memes app); QQ Music (music streaming service)King of Glory (China’s most popular game) and many other gamesSummary and Contextualization of Li’s Main Points in three chapters:
Chapter 1: Let’s start with a case: Tencent v. Alibaba in the new retail field as an example of why all companies are only as valuable as their Data+Algorithms
First, Alibaba’s approach is the exemplar: In 2008, Alibaba made a very intentional choice to transform from an e-commerce company to a data company, and Jack Ma has repeatedly emphasized that Alibaba is a data company. This has been key to their remarkable success with a new business called Hema, a supermarket chain that emphasizes freshness (in-store shoppers can select live seafood and eat restaurants inside the supermarket; online shoppers receive free delivery within 30 minutes). The essay walks us through how Hema succeeded due to data+algorithms:Site selection for Hema stores: Tmall (Alibaba’s business-to-consumer e-commerce platform) has a lot of address data for purchases and deliveriesWhich goods to put in the store: Purchase data from Tmall and Taobao (Alibaba’s online shopping platform)Hema also contains various food and beverage outlets, which ones should it include? Alibaba bought Ele.me, a food delivery startup so it has food delivery data to help answer this question.How do you maximize storage and logistics efficiency? Alibaba’s Cainiao is a logistics data platform.How in the world can you guarantee half-hour delivery: Ele.me data can helpPush content in the Hema app: depends on consumption record in Taobao, Tmall, and previous consumption in HemaTencent has tried to empower (key word for this translation) companies to compete with Alibaba in new retail, including Carrefour and Yonghui Superstores, but Tencent supports only some of the technical aspects, such as facial recognition, cloud services, and small applets, but it can’t provide the same level of big data and corresponding algorithms that Alibaba can.
Chapter 2: Why can’t it provide this same capacity in terms of big data and algorithms? We need to look at issues of 1) Strategy and 2) Organizational Structure
First, Tencent has not upgraded its algorithms to process data (note: the piece primarily focuses on two types of algorithms – 1) user profile algorithms which analyze user data and categorize users under tags such as “Manchester United fan”, 2) content distribution algorithms which match these user profiles with characteristics of the content)The running case is Tencent vs. Bytedance, which owns Toutiao(newsfeed app) and Douyin or also known as TikTok (Chiense music video platform and social network), here’s how the two stack up:Bytedance companies exceeded 700 million monthly active users (MAU) in March 2018; its advertising revenue in 2017 was 15 billion RMB and its target for 2018 is 45 billion RMB; it has 1500 engineers (800 of which are algorithm engineers)Even though Tencent’s user data is “far and away” larger than Toutiao’s, Tecent’s algorithms “still give a very imprecise profile of users” because “Tencent’s customer data is scattered in various departments and has become the ‘private property’ of departments” (e.g. WeChat’s advertising algorithms are not under the purview of the WeChat department but are actually under another department which does not have access to the data of the WeChat team)Tencent does not have a strong algorithm research departmentas each unit has its own algorithm engineers so there’s a lot of low-level, redundant development of algorithms --- compared to Bytedance, the number of engineers who are solely dedicated to doing work on improving algorithms is “pitifully few” according to Li’s analyst friend; also, Tencent hasn’t had a CTO since Zhang Zhidong retiredBytedance’s advantage in algorithms is challenging Tencent’s advantage in overall traffic: in June of this year, the monthly usage time proportions of Tencent’s mobile apps dropped by 6.6 percentage points from 54.3% a year ago to 47.7%, while the monthly usage time proportion of Bytedance’s mobile apps jumped 6.2 percentage points from 3.9% to 10.1%.
Second, Tencent has failed to aggressively acquire more data and add more dimensions to its existing data
Four main issues:It has limited its willingness to fight and compete due to concerns it was too strong after the 3Q war: “Tencent has now ‘put the weapons back in the arsenal and let the war horses graze on the hillside’”-- Its decision to focus on “two and a half” domains (social, content, and Internet+) has allowed competitors to challenge it from those other areas: Douyin (AKA Tik Tok) emerged from short videos but is trying to become a social app, challenging an area of Tencent’s core strengthIts data is fragmented, and it has not spent the time and money to integrating the company’s internal data, according to Li’s good friend at TencentTencent has not added data in other dimensions such as e-commerce, search, and offline data: Tencent only gets data on the total amount of transactions made through WeChat pay, but it cannot get the deeper and more valuable data on product names, unit price, etc.Tencent invests like it is a VC or PE company rather than like it was a tech giant like Amazon, Google, or Alibaba: the latter grouping focus on investments through which they can wholly-own or own controlling shares of companies so they can get the core data from these companies. Tencent is okay with being a small shareholder, which doesn’t get it the core data from the companies it invests in.Chapter 3: A Long-term, Grand Strategy for Tencent: What can Tencent do to regain its competitive strength in data and algorithms?
In data: thoroughly open up channels to share the internal data of the company and establish a hierarchical data authorization management mechanism
For algorithms:Set up high-standard research institutions, recruit world-class algorithm maestros to fully tackle the problemEnter into academic cooperation with some famous research institutions in order to find breakthroughsAcquire a large number of excellent algorithm companies at home and abroad on a large scale, and to solve the problem by purchasing technology. Now this is what is called a multi-path, multi-echelon, and full-capacity offensive.Look, Tencent is not doomed, and it still has the strongest asset in the game in Wechat and its portfolio of apps still gets nearly half of all China’s monthly usage. There's some favoritism shown toward Alibaba in this piece in my opinion, but overall I think it's a relatively balanced take and the third chapter emphasizes the esteem the author has for Tencent's leadership. But in the transition from traffic 1.0 (users go and find information) to traffic 2.0 (information goes and finds users), data+algorithms will only become more and more important and Tencent has serious problems in those two aspects.
Take a good half hour and read the whole thing, I promise it’s worth it:
A Total Rethinking of Tencent's Strategy
This Week's ChinAI Links
Chinese Idiom of the Week: Adding a new section for those Chinese language buffs out there. I’ve been having a lot of fun learning new idioms and phrases with each of these translations. This week’s article ended with a new favorite: 抛砖引玉(pao1zhuan1yin3yu4) – a self-deprecating idiom which has a literal translation of “casting a brick to attract jade” and figuratively means to offer a few ordinary remarks to draw others into making valuable suggestions and comments. Li Guofei ends his pieces using this phrase.
We at FHI were glad to host Graham Webster and Scarlet Kim earlier this year in May for a roundtable on the implications of China’s AI development on privacy. Their piece in Foreign Policy makes a good case that U.S.-China competition over tech shouldn’t be used to water down privacy protections.
Christina Larson’s piece on China’s governance through data, AI, and internet surveillance is well worth a read. Though, I think this week’s translation provides a rejoinder of sorts. If Tencent (with all their talent, data, algorithmic tools, access to hardware) can’t come up with a somewhat integrated dataset on their users that is usable, can you imagine the hurdles and barriers up ahead for the Chinese government in terms of actualizing their visions of digitally-backed governance?
Folks at Nesta have consistently been doing some of the best work on China’s innovation. This analysis of AI/ML research trends using Arxiv data finds that China has more than doubled its publication activity in deep learning-intensive domains since 2012.
Correction to last week’s issue – in this section I mentioned that Qi Lu had left Baidu: while he has left his role as COO, he’s still listed as the Vice Chairman of Baidu’s Board & Management. H/t to friend of the newsletter Danit Gal for pointing this out.
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 jeffrey.ding@magd.ox.ac.uk or on Twitter at @jjding99