Bytedance CEO Zhang Yiming at Tsinghua University: Part 2
Hey Tech Buzzers,
Welcome back to our series on Bytedance, maker of TikTok (and many other things!) and the most richly valued privately-held tech startup in the world.
The below is a translation of a 2-hour interview on March 20, 2018, with Zhang Yiming, Bytedance founder and global CEO. It is the second of four parts and was conducted by Professor Qian Yingyi, who was the Dean of the School of Economics and Management at Tsinghua University at the time. For greater context and understanding, we have annotated it with our comments in italics. Part 1 is here.
PS Mega thanks to Pandaily editor Lu Zhao for the translation! You can learn more about her work here.
Professor Qian Yingyi (L) and Zhang Yiming (R).
Interview Transcript
Companies Must Pay Attention to What’s Happening Outside
Qian Yingyi: It’s really important to keep up the interest and ambition to read persistently and systematically without the pressure of exams. When I visited Bytedance* in November last year (2017), I was very impressed, as was everyone else, by the rapid growth of a new-generation internet company that had only been in business for five years. Toutiao has created a new unicorn quickly in a fiercely competitive market. What is your core competitiveness?
Tech Buzz: I think it’s a good time to mention here that at this juncture -- March 2018 -- the company was still referred to by its hit product name, Jinri Toutiao* (typically abbreviated as “Toutiao”), even though yes, it had been registered under the name Bytedance from the beginning. But a business’ registered name is often different from the name under which it does business, so this is no surprise. Anyway, it wasn’t until a month after this interview in April that the company officially began (re)branding itself as Bytedance. Since we’re translating this interview in 2020, we have changed all references to Toutiao the company to Bytedance, since that is more precise, and left references to Toutiao the product as Toutiao.
Zhang Yiming: The first layer of our core competitiveness is our product, behind the product is our technical system, and behind that is our team and culture. An internet company’s technology is not going to be monopolistic nor some kind of top secret, it’s only the team and culture that can lead to continuous innovation and introspection. Our core competitiveness is team and culture.
Tech Buzz: It’s a bit odd to me that Business Schools haven’t focused on this aspect of the company; this case study, for example, seems to be concentrated on understanding the power of the algorithms. But as Zhang Yiming says here, Bytedance didn’t have some monopoly on “recommendation engines” as a technology. In fact, at the beginning, they didn’t even have anyone on staff who had experience writing one, and Zhang Yiming had to go to the internet to learn it for himself. Also, other Chinese companies soon had their own “AI-enhanced” apps as well, because Toutiao’s early success was no secret. But even though it has held onto its title as category leader, the app’s user growth has hit a ceiling in recent years, and it’s globalization efforts (called TopBuzz) were terminated for good last week. Of course, Zhang Yiming didn’t necessarily know that’s what the future held -- and that it was going to be TikTok that would propel him and Bytedance into the global spotlight. So it’s true: no single product insight will keep you ahead of the game, no matter how much “technology” it requires. But how much you choose to devote your time and resources to fine-tuning your company so as to be able to consistently generate such insights? That is key.
Qian Yingyi: Many companies have similar answers. Team, culture and tech talent are the most important. From your perspective, what are the special characteristics that differentiate your company from others?
Zhang Yiming: We do have some characteristics, but I am not sure if we are unique. For example, our human resource practices. We emphasize "Context, not Control", which means that among colleagues, we rely on providing context and information transparency to guide management and coordination.
Generally, companies tend to want to regulate processes and establish systems. We try not to establish too complicated or too many systems, which we believe will inhibit innovation. For example, the finance department can easily issue a rule, and the legal department can propose another rule. It’s easy for each department to establish two pages of regulations, but when all taken together, there will be many restrictions and lots of conflicts.
We pay more attention to context and hope to give employees more context instead of breaking down their tasks for them. I write down my goals and my staff write their own after seeing mine. I rarely change their goals; I may supplement them. After the company's big goal is set, everyone thinks of their own contributions to the goal rather than asking how others will contribute. This is our practice. It’s not necessarily suitable for all enterprises. It depends on your ideas about how to organize a corporation.
Tech Buzz: I’ve linked above to a full speech that Zhang Yiming did for Source Code Capital (a prominent VC and Bytedance investor) on what he means by this exactly, but it’s pretty self-explanatory. In addition, for the upcoming ebook, I’ve done hours upon hours of Bytedance current and ex-employee interviews, in addition to scouring the internet for everything written about and by the company, of course. What I’ve found is that while yes, there are significant differences depending on your small team and job function, for a Chinese company … Bytedance is actually pretty … decentralized. (The stereotype about Chinese companies, in case you didn’t know, is that they’re very hierarchical and bureaucratic, and I can’t say that is inaccurate based on my personal experience.) Of course, being flatter organizationally, which we’ve talked about in Tech Buzz Episode 28, and being less “process-driven,” also leads to plenty of other problems, such as just lack of clarity, weaker risk management and oversight, etc. There is no pro without a con, folks. Lastly, when he’s talking about goals, I believe what Zhang Yiming is referring to is the OKR system (Objectives and Key Results, famously adopted by Google), which Bytedance began utilizing in 2013, in its second year as a company. As he says here, indeed, his OKRs are fully transparent throughout the firm, by everyone. And so is everyone else’s.
Explanation of OKRs. Taken from weekdone.com.
Qian Yingyi: The following questions are related to the characteristics of your business. Unlike traditional news media, there are no reporters or editors at Toutiao. The core of the operation is a set of algorithms built from code. The algorithm records users’ behaviors in the app, so that in the sea of content, it can know the content you are interested in, and even the content you may be interested in, and promote them to you accurately. But this type of behavioral targeting and push model of information acquisition may also bring problems with it -- that is, how to prevent the public’s access to information from becoming narrower and narrower? This is a concern for many people.
Zhang Yiming: We noticed this problem very early. It is a topic of interest requiring expertise in the field of content recommendation, but it is not an unsolvable problem; it is an optimizable problem. We need to find a balance between exploration and convergence. Too much exploration decreases accuracy, while too much convergence leads to information being too narrow.
We do a few things. Firstly, we provide more discovery mechanisms in the product where you can explore more information by visiting different channels, following users, searching keywords, etc. Secondly, we have some public topics that everyone will see, which are selected by our colleagues. Thirdly, from an algorithmic point of view, personalized recommendation is not actually a recommendation for each individual, but essentially a collaborative recommendation. We recommend content to people who have similar characteristics as each other. There is a degree of generalization in the recommendation. Individuals are independent and unique, but groups consist of individuals with similar characteristics. For example, students at School of Economics and Management, Tsinghua University are similar individuals; Tsinghua students are similar individuals; all students are similar individuals; Chinese users are similar individuals.
The degree of generalization is controllable. We’re not only recommending to you what you are interested in; actually, it is more like everyone consumes but also everyone contributes. We can balance between the two. Convergence or generalization? We have tried different degrees on different people, and have made great improvements by now. Finally, let me emphasize, we recommend content based on interests, not opinions. We won’t only recommend one-sided opinions on the same issue, we will recommend all of them.
A super simple explanation of user-user collaborative filtering.
Tech Buzz: Here he’s giving a clear explanation of what sounds like user-user collaborative filtering. No doubt Toutiao is a lot more complex than just “finding users similar to you and serving you what they liked,” but then again, unlikely we’re going to get a high level of technical detail at a business school interview. I think what’s more important to note is that fear of “echo chambers” is everywhere, in China too. I’ve not seen a very satisfactory answer yet from a tech company on exactly how they are addressing this. On the one hand, Toutiao wants to show you only what you care about -- its first slogan can be translated as “only what you care about can be considered headlines” (你关心的，才是头条) and was not changed until after this interview -- but then it also claims that it’s going to burst your bubble and expose you to opposing viewpoints. I don’t know … that seems awfully difficult. Anyway, the point here is less that Bytedance has a solution, and more that it is of great concern in Chinese tech circles and society more generally. Personally, I would like to see it prioritized as a top concern for all social media and content companies … but like most of you, I’m not optimistic that will be the case.
Qian Yingyi: Regarding everyone's interests, inevitably, algorithms can't judge the quality of interests, so you can have some vulgar content. How do you deal with this?
Zhang Yiming: We take this issue very seriously. All social media platforms including Facebook and YouTube face the same problem. Being data-powered means that your content can be driven by headlines and negativity.
After discovering this problem, the first thing we did was to establish a content quality team, where we have thousands of people controlling for quality.
Secondly, on Toutiao we can see on the backend that users are mostly reading articles regarding daily life, education and work. Everyone sitting here may not see much information about agriculture, rural areas and farmers, but many agriculture practitioners do. We are constantly enriching practical content. I believe that the content on education and academia will continue to grow as well, with a higher proportion of practical information.
At the same time, we continue to improve our technology. Last year, we introduced artificial intelligence to identify copyrights, rumors, and vulgar content. The effect is very good and we have made great progress.
Tech Buzz: This interview pre-dates our very first episode of Tech Buzz by about a month, when Bytedance’s popular jokes app Neihanduanzi was shut down for good (they later revived it as Pipixia, which we cover in Ep. 67). At around the same time, Toutiao itself was taken off of appstores (along with some other news apps). Prior to that, Toutiao had already been in various skirmishes with the government over inappropriate content and other ills. It’s important to note that the problematic content was not political, but merely “vulgar” as deemed by the government (and many users, too). My personal opinion? Indeed, some of the content, which was all third-party contributed and probably designed to be clickbait, was just plain dumb. (You can find examples of them in Ep. 28.) Today, public perception has recovered somewhat, but at the time, it was really not necessarily a guarantee that Bytedance was going to survive the onslaught of regulatory scrutiny, incessant negative media coverage, and an unsympathetic public. And let’s not forget about the escalating war with Tencent, either (Tech Buzz Ep. 9). Anyway, I recently (re)installed Toutiao to make sure my understanding of the product was up to date. My opinion would be that Toutiao has indeed improved a bit … but there is still a long way to go. For one, I’m probably not going to be a frequent user. But the company does know that is an issue. In an effort to shed its “low”(brow) image, Toutiao actually signed a few hundred top influencers from Zhihu (China’s Quora, known for its intellectualism) in 2017. But as Zhang Yiming likes to say, people want to “vege out,” so let them. Not everyone wants to be intellectually stimulated all the time! I’d be curious … what do you think?
Qian Yingyi: There is another related problem. With the rapid development in mobile internet, we have seen many mobile products like short video apps rise and decline rapidly. This is an objective fact. You have many products, what do you think of this problem? What strategy do you have? How do you keep up with the times?
Zhang Yiming: When I was CEO at 99Fang, the most important thing I learned was to face uncertainty. Many people may worry about this issue: the market will change, can I keep up with the times? From my experience, we should confront the changes head-on and find the more fundamental things behind them. And the fundamental thing is the needs of users.
There are no products that don’t change because user habits are changing and user composition is also changing. Therefore, no matter what field you’re in, mobile internet or otherwise, there is no unchanging product. The core is to continuously evolve products based on real user needs, then you’ll be fine. Do not expect one solution to solve all the problems in perpetuity. To be able to continuously capture user needs, the team's culture needs to be sensitive to outside information.
A screenshot of two different Toutiao users’ personalized feeds. Taken from SCMP Abacus News, Dec. 4, 2018.
Platform companies must bear social responsibilities, just like infrastructure
Qian Yingyi: The information and social distribution business is a platform-type enterprise. As a platform company, there naturally exist responsibilities to society and the [problem] of how you shape values. What role do you hope to play in platform governance? And what do you think of your social responsibilities? We’re all very concerned about the responsibility of the platform.
Zhang Yiming: I have had more experience in the past two years. We usually say that an enterprise as a business organization has several forms. One is self-employed, you only need to manage yourself, you don’t affect others. One is a studio, where you participate in one part of a company. One is a typical enterprise, which has an upstream, downstream and peers. One is a platform-type enterprise, where you influence many people. There is also a global platform enterprise that is very influential in many countries.
In 2015 and 2016, we were moving towards becoming a platform company, and now we are moving towards becoming a global platform company. In recent years, technology has increased the penetration rate of mobile internet, and the influence of platform-type enterprises has become greater and greater. We meet our users and customers everywhere. After they are of a certain size, platform enterprises, especially technology platform enterprises, become social public infrastructure. At this time, the requirements for the platform are not just the legal responsibilities expected of commercial companies, but even higher.
As a technology company, first of all, technological innovation is our foundation. We use innovative methods to improve technology and create value. Secondly, as an infrastructure, providing information is our value. As a platform, we have a responsibility to promote the production of better content and provide better information services.
In the past two years, we made a plan to make sure there were at least 1,000 creators at Toutiao who earned no less than 10,000 yuan a month (~$1400). The goal was to promote the creation of original content. If we didn’t do it, the entire industry may have not improved fast enough. We also made a system to find missing persons, kind of like the AMBER Alert system in the US. We found the first missing person in February 2016. As of now, a maximum of 29 people were found in a day, and more than 5,000 people have been found in total. We wouldn’t lose any business commercially if we didn’t do this, but we believe that we’re also providing information services for society by finding missing people.
In the past, we’ve seen missing persons notices on newspapers or TV, but the efficiency was relatively low. Based on the location where the person went missing, we use geo-located pop-ups to find people, which is more accurate.
This has nothing to do with legal and commercial interests, but as a platform, if you don’t do it, maybe no one else either, so you should bear this responsibility.
Thirdly, apart from the responsibility of technological innovation and information provision, we, as a content platform, assume the responsibility of platform governance. Problems like clickbait, fake news, preventing internet addiction amongst minors, we as a platform have a responsibility to minimize those as much as possible. Perhaps legally, according to the Safe Harbor principles, companies are only required to handle complaints. But as a platform, we should take the initiative to do this and be responsible for the ecology of the platform.
We also see that globally Facebook has also been met with many challenges. They used to think they were just a technology company, and put more emphasis on their technology. But now, we emphasize more that we are a platform company, which uses technology to build the platform, and the platform itself spreads information.
Tech Buzz: I think I’ve said enough on Bytedance’s troubles at the time in the comments above, so I won’t say more here. A lot of those exact same discussions continue today, where people are asking, what really is the value of a Toutiao or a Douyin / TikTok to society? And I think there are many conflicting views. Furthermore, even if Bytedance believes in “social responsibility,” what does that really mean? Are we, as a society, in agreement on what that means? Complex issues, my friends, very complex.
Qian Yingyi: So from the perspective of your company, you’re conscious of all this and need to work on it. The supervision of platform technology companies is tightening globally, what impact do you think this will have on the development of platform companies? How will Bytedance deal with it?
Zhang Yiming: I think this is a problem that the entire technology industry must pay attention to. Technology has increased leverage and scale, and technology platforms have gradually become an infrastructure, like telecom operators and electricity. The internet is increasingly facing such problems. Platform enterprises should take the initiative to be more transparent, to communicate and supervise, and to explain to the public. This is a requirement for platform companies.
Qian Yingyi: You just mentioned public facilities. Do you think internet companies will have such attributes in the future?
Zhang Yiming: Like WeChat, its communication and payment services are already a part of life. So are Toutiao and search engines. In the original PC internet era, there were 200 million netizens and the penetration rate was very low, but in the mobile internet era, cheaper chips are installed in more and more devices and distributed to more and more people. Internet applications are indeed turning into an infrastructure.
Qian Yingyi: In addition to the infrastructure attributes, there is another very important attribute, that is, your core technology is artificial intelligence. I participated in your global “thought feast” held in December last year. You said in the opening speech, "Toutiao, as an artificial intelligence company, should be filled with responsibility and goodwill for the future of humanity as a whole."
You touched upon some important topics like artificial intelligence and its impact on society. It reminds me of Stephen Hawking who passed away last week. He was also concerned about artificial intelligence and the problems it brought. He and Musk co-wrote a letter saying that artificial intelligence should have such a sense of responsibility and goodwill. At Bytedance specifically how are you thinking about this?
Zhang Yiming: The development of artificial intelligence is growing quickly, bringing more and more breakthroughs. It is a very powerful ability. The greater the ability, the greater the responsibility. When using it, the main goal should be the interests of users, not the interests of business. We should use artificial intelligence technology to improve the experience and increase the value of the platform. This is true for artificial intelligence, as well as genetics and aerospace. They have great utility and can lead companies and industries both to another level, and also encounter many challenges.
Humans have encountered similar situations in the past. If you look back at history, you will find that when the chemical industry was first invented, it caused a lot of environmental pollution, but then the entire industry controlled itself and the problem of chemical pollution was solved. Now artificial intelligence has a greater impact than the chemical industry. This current generation of entrepreneurs should be conscious of this earlier; rather than waiting to solve the problem after it arises, they should think earlier.
Qian Yingyi: You just mentioned three areas, artificial intelligence, biotechnology and aerospace. Do you think that these three areas have changed a lot?
Zhang Yiming: They may bring very big changes.
Qian Yingyi: And also many challenges, technological ethics and moral challenges.
Zhang Yiming: Yes.
Tech Buzz: Of course, if you founded a company with AI at its core, you’re probably an AI-optimist. So these words do not surprise me. “We can figure it out.” “This generation of entrepreneurs” have a lot to do! I certainly don’t want to be living through the AI equivalent of acid rain.
Enjoyed this? Do subscribe to our mailing list and podcast to get more insights on not just Bytedance but China tech in general, and go to our website, Tech Buzz China, for more!
Unique insights and takeways on China tech that don’t always make it into English language coverage.