Revenir au site

Baidu’s COO, Qi Lu Discusses AI with YC's Daniel Gross

· Founders and VCs,Baidu,Qi Lu,YCombinator,AI
broken image

Daniel Gross interviews Qi Lu, covering general developments in AI, China's approach to AI and self-driving cars, China vs. US tech environments, differences across large tech companies, and more. Qi is the COO of Baidu, and Daniel is a partner at YC.

[Editor's note: this interview has been edited and condensed for clarity]


Daniel Gross [00:00:34] – Cool. I guess, first question that is on my mind is, and I think many others, help us understand why you left. Previous to Baidu, you were very senior at Microsoft and a lot of us are wondering why you decided to leave to Baidu.

Qi Lu [00:00:52] – Two things. One is, I left Microsoft purely for personal reasons because I had an injury. I broke my left hip, so I needed a second surgery and needed to take some time off. Because my job at Microsoft was very critical to the company, I felt it was in the best interest of the company for me to move on. A great thing is that I had a good successor who is extremely capable. I’m super happy that he is taking over and leading the company’s productivity business moving forward. In particular, also, I was able to have a very good relationship with Microsoft. I continued to serve as the personal adviser to the CEO, Satya Nadella, and also to Bill Gates. When I go back to Seattle, I often go see them.

Daniel Gross [00:01:50] – Then, how did you decide to go to Baidu as opposed to any other place?

Qi Lu [00:01:55] – That’s for a simple reason, which is AI… Most people in our field would agree, AI is the next big wave. I think AI is particularly meaningful, because in my views, China has a structural advantage in terms of AI technological development and the commercializations. In that context, Baidu offers a very unique opportunity for me. First of all, Baidu, in many ways, is the Google of China. Its heritage was a search engine, and as a result, from an engineering capability perspective and a cultural perspective, it’s uniquely positioned to seize the AI opportunity. Also, I happened to be, a friend known Robin Lee, the founder and CEO for almost 20 years, so there’s a lot of long term relationship and trust, so that was just a good opportunity for me to take.

Daniel Gross [00:03:06] – In what ways is China’s approach to AI different from America’s?

Qi Lu [00:03:16] – First of all, I think it’s environmentally different. Approach wise, I’ll come back to the approach aspect from my vantage point. From the environmental perspective, I think China has a unique structural advantage for AI technological development and commercialization of AI technologies for a simple reason, if I may just explain my thinking on why this is so. In this wave of technology development, there’s one aspect that is fundamentally different from the previous generation of the big technology wave, which is data plays an essential role. I’ll offer you this simple example. You can have 10,000 engineers, great engineers, or you can have a million great engineers. You will not be able to build a system that understands human conversations. You will not be able to build a system that will recognize objects or scenes of images because you need to have data. A simple analogy is very much like humans. When you and I grew up, it’s not like our parents or God is writing coding to our brains. Our builtin neuro-engines have the ability to learn, so our sensory systems, essentially, our perceptive systems, whether it’s visual systems or whether it’s auditory systems that we are able to observe the world. Our observations, those sensors, these are data. This data carries knowledge, and we are able to learn from our interaction with the world. As we grew up, we acquired knowledge. The same thing happens for AI technology. It’s not about writing code this time. It’s about writing code that implements algorithms

Qi Lu [00:05:14] – with both soft and hard wares that are able to learn, and learn knowledge from the data. If you take that perspective, data in my view is for the AI era. It will become a primary means of production. By definition, means of production is a form of capital. We look at, historically, in our human history, let’s say, in the agricultural era, land is the primary means of production. You can see everything is organized around the land. All the wars are competing for land. In the industrial eras, the means of productions are primary labor equipment, different type of equipment. And certainly, financial capitals, human talent. But in the AI era, my view is that data will become a primary means of production. Harnessing data becomes key. And that comes back to China, because China has a different socio-economical policy around it. For certain segments, not on everything. For certain segments, it’s much easier to acquire and harness it. With that, it creates an environment for developing AI technologies, and then commercializing those technologies towards market oriented applications or social applications. It is in that context that China has a structural advantage. In terms of approach, there would be cultural differences, even in the entrepreneur world. The startups in the China environment, they tend to work in their ways. That, I will say, Silicon Valley and China, there’s common attitudes, there’s some different approaches, but that’s not the bigger factor. In my view, it’s the environment that’s the more

Qi Lu [00:07:03] – determinant factor making China to be, relatively compared to other marketplaces or other regions, a better place for AI development, because of data.

Daniel Gross [00:07:15] – Interesting. I guess, one question I’m wondering in particular is, in the US, there’s this belief that one of the ways China is somehow doing better when it comes to technology is that the government is much more integrated with companies and their initiatives. Is that something that you see at Baidu? As you guys focus on your different AI initiatives, are you able to work very closely with the government?

Qi Lu [00:07:40] – In general, the Chinese government, at this stage, has a lot more willingness to invest in infrastructures, in talent, and they in particular see AI as an opportunity for China, in many ways, to ride that big wave, to elevate its innovation capacity. There was, about somewhere between one to two months ago, there was a white paper published by the Chinese government that actually spills details about, by 2030, how the Chinese government plan to systematically invest in infrastructure, talent and technologies to enable China to lead in AI technologies in many different dimensions. In general, the government indeed has a lot more willingness and commitment to invest. With regard to private company, a particular company like Baidu, which is more a view that’s culture and practice-wise closer to an American company. Listed in NASDAQ. In turn, the working culture is very entrepreneurially closer to Silicon Valley style. We do, in many ways, operate independently. We view, essentially, market opportunities as the primary objective to pursue those opportunities. And we enter a win-win environment with the government initiative. We welcome that. For example, Baidu is the host of national labs and Baidu is also working with various different government entities when they have expressed willingness to support certain areas of AI technology. For example, let’s say for self driving cars. We will work with those government entities to discuss opportunities that are mutually beneficial. But as a company, our primary means is market success. We don’t have any other agendas,

Qi Lu [00:09:51] – because we are an independent company. We want to build products that service our users. When there’s synergetic opportunities with government support, we will collaborate with the government when there is mutual benefit, mutual win-wins.

Daniel Gross [00:10:06] – Do you think that China will beat the United States to having mass adoption of self-driving cars?

Qi Lu [00:10:15] – My belief is, the opportunity to commercialize and deploy autonomous driving technologies in various forms, China will have opportunities to get ahead of the United States over the next three to five years. Primary, I will say, a few areas. One is, different regions, whether it’s municipality or provincial government or central government, they see this as an opportunity for China’s auto industry. Right now, the China auto industry, there’s no real strong technology. Heavy fragmentation with over 250 OEMs. The Chinese government would very much like to take the autonomous driving dimension of innovation to enable the Chinese auto industry to leapfrog, to be the world’s best and lead the world. The government is a factor. For example, there’s five municipal governments right now. Members or partners of Baidu’s open autonomous driving ecosystem, an open platform called Apollo, they work with us on a variety of initiatives. For example, a new kind of driving schools that will certify autonomous vehicles for different levels of maneuverability. Just like a driving school today, they will certify human drivers if you pass a certain test. We’re working on that. We’re working with a new city that’s being kind of build ground up. By the Chinese government’s plan, it will be bigger than Shenzhen. It will be bigger than Dubai in five to 10 years. It’s called Xiongan. It’s a massive new city that’s being built from pretty much zero. So, we’re working with them, designing new infrastructures,

Qi Lu [00:12:15] – a new segment of the city that makes it much easier for autonomous vehicles to be deployed. As an example, let’s say, today’s cities, you have street lights, and the street lights, in many ways, are a sensor device. It enables the sensors of a vehicle to be able to better see the road. It just happens to be the one area that does the sensing are humans, and humans use eyeballs. When it’s dark, you won’t be able to see the road, see the separation of the roads, and you have street lights. But imagine in the future when the sensor is not done by the human eyeballs, but a different sensor, whether it’s lidar, radar, or cameras, whatever the sensor technology used. The future city infrastructures, those street lights will be intended for non-human sensor capabilities to see the road and to be able to navigate the road. We’re actively designing those new type of infrastructures and having ongoing discussions with these municipal governments to lay out plans to build those infrastructures, with the intent to have commercial deployment of autonomous driving in various forms. If you combine all those efforts together, I very much believe in the next three to five years we’ll see autonomous driving in China to get deployed in more variety, in larger scales than other markets.

Daniel Gross [00:13:50] – Fascinating. Going back a little bit to, kind of, more broadly, China and the United States. You were managing very large software engineering teams here in the United States, and now, you’re doing the equivalent in China. What are some cultural differences you’ve noticed in terms of how people work, how you have to manage, in between those two countries?

Qi Lu [00:14:15] – First of all, Baidu’s engineering cultures, product cultures, it’s very similar to Microsoft. Very similar to what I know of Google, even though I haven’t worked at Google, but I have enough interaction with friends who’ve worked at Google. Essentially, very heavy in technology. Very heavy in algorithms. Very heavy in large scale computing. Very weak in product design. Very weak in understanding user needs, human needs. As a result, the technology is good. The product, generally, isn’t great. I’m not critiquing or criticizing my former colleagues, but Microsoft as a company, in many ways, lacked behind companies such as Apple and Facebook in building truly mobile, particularly mobile consumer products that stressed the emotional connections with users. Whether it’s applications or services or devices, the fit and the finish, the experience design is very much more than appealed to a young demographic, the young generation. Microsoft, as a company, struggled on that. I see similar things, from what I can see. Google as a company, the products that I use. Baidu is at the same way. That’s one aspect. I always tried to change the engineering culture at Microsoft. Actually, that was the reason why I broke my leg. It’s a different story. You need to unlearn and learn a new way of doing things.

Daniel Gross [00:15:49] – Can you just tell us bout the bicycle you rode, which is ow you got that injury?

Qi Lu [00:15:54] – Yeah, there’s something called a backwards brain bike. If you search on YouTube how to ride a bicycle, there’s plenty of videos. Essentially, the bike goes the other way. If you turn the handle this way, the wheel actually goes the other way. There’s some profoundly important reasons, because first of all, we humans learn, there’s three primary ways that we learn. This is called experiential learning, and the bicycle riding, often said, is the best example because you cannot learn how to ride a bicycle by watching other people riding a bicycle, by reading about it, by people telling about it. You have to ride a bicycle yourself, and often, bumping, bruising, hurting, but guess what? There’s one thing. Once you’ve learnt, you never forget. It’s in the muscle memory, you don’t think about it. And that’s the problem for large organizations, for cultures, because the reason those big companies, they couldn’t survive when we’ve come, that’s based on Professor Rebecca Henderson’s study at the Harvard Business School. Those mature organizations, their muscle memory, the way they talk to customers, the way they do research, the way their design experiences was built, like, 30 years ago. They try to think, but their muscle memory doesn’t think. They will just do things that way. If you ask me why Microsoft couldn’t get mobile at all, it isn’t that we’re not working hard. We’re working super hard. It isn’t that people are not smart. We tried everything, we bought Nokia, we built Cortana. You name it, we tried everything.

Qi Lu [00:17:25] – But the product, honestly, sucks. It’s just because of the muscle memory. I was searching for an answer. Rebecca Henderson was the one who convinced me that this is the real problem. A Microsoft colleague of mine, his name is Bill Buxton. He’s one of a kind of people. He said, “Hey, Qi, you should try this bicycle thing.” It was really interesting. We built the bike, Bill Buxton and another one. The three of us tried to practice, because this bike, for a normal adult who knows how to bike, takes you about eight months turning every day. And once you learn how to ride that bike, you won’t be able to ride the normal bike anymore because you need to rewire your brain. I think for large organizations, culture change is that difficult because it’s your muscle memory. The way you do things, it just becomes habit. You don’t even think about it. Even though the CEO says, “You guys have to figure out mobile.” They tried, tried, tried. The mobile product just looked like a PC product in smaller form, right? Because that’s how they do it. Coming back on culture, I see Baidu has very similar traits of Microsoft that I work with. What I’m working on today at Baidu is really to change that engineering culture to be a lot more product centric, to be a lot more understanding of user needs, particularly for mobile products, for AI products. Then, brief answer, the engineering culture between companies that are in China versus companies in the United States, there are very various different aspects of it. The biggest thing, and I need to perhaps think more

Qi Lu [00:19:09] – about summarizing in my head what I observed so far, some of the key differences in terms of product engineering culture, the one thing I will say that stands out for me, I learned a lot in my eight months plus living and working in China. The product people in China are a lot more philosophical. They are a lot more reflective. They think a lot deeper than what you would typically observe from product people when they describe their product. Also, the Chinese R&D product leaders emphasize a lot more self reflection. They use the word cognition, but it means a person’s ability to understand, to make judgment, make decisions. Essentially, they emphasize a lot more self improvement for product people in particular.

Daniel Gross [00:20:06] – Interesting.

Qi Lu [00:20:07] – How you elevate your cognitive capacity. If you ask me, the one thing that stood out for me is, I used to believe the product people in the United States companies were better. Now, I kind of have it the other way around. I see better product people more often in Baidu and other Chinese companies that I interact with than perhaps, I will say, on average, the percentage.

Daniel Gross [00:20:32] – On that point, there’s another belief, in the West that California and Silicon Valley are very creative environments and they really allow ideas to come up and bubble up from any person in an organization, versus China where the image, I guess, that we think to ourselves is a very structured society that is very good at implementing something, but maybe not as good at creative, free thought. Would you agree with this sentiment at all, and if so, how do you think that plays out for, say, doing core research that involves a lot of creativity?

Qi Lu [00:21:14] – Yeah, great question, that’s a good one. I will say there are different degrees of truths towards the top-down nature for Chinese companies. Baidu, even though, among the Chinese tech companies, Baidu is the closest in terms of culture to Silicon Valley. A lot of people, their pedigrees are Google, worked at Google, worked at Microsoft. Mainly English, it’s also kind of a common working language. You won’t have any problem if you just speak English or write an email in English. Even that, the top-down phenomena happens. My hypothesis was, this is perhaps due to 2,000 plus years of Confucianism, you know? Confucius is essentially harmony through hierarchy, right? That’s the central idea of Confucianism. Having said that, the companies that I work with, including Baidu, all realize driver innovations is a lot more about empowering teams, empowering capable leaders to experiment, to try new ideas at a fast velocity. Baidu does a lot of those, and in the startups that I interact with, they emphasize that aspect a lot. There’s no difference in terms of belief and practices in Silicon Valley startups that I see. The large company, one company I’ll probably point to I believe, overall, does a good job is Tencent. Tencent, they have this challenging culture. Any ideas, they encourage and challenge the more authoritative or senior people. And also, for any major initiatives or any areas of new innovation, they tend to have two, three teams working on the same thing. There’s a lot more internal competitive dynamics that are going on. One last thing.

Qi Lu [00:23:17] – In Baidu, we have this quarterly meeting. We have all our company directors. We have about 200 directors. Once a quarter, we invite speakers, and the past few speakers, they all emphasized the aspects of building a learning organization so that a truly thriving organization, each cell, each team, they are able to be nimble, adapt, quickly learn. Even though there are these couple thousand years of Confucianism, I think it’s still somewhat there. It reflects to different degrees in different companies, but by and large, driving innovation, empowering teams, empowering leaders are the common understanding. Everyone in the organization is striving to do more and do better in that regard. There’s no fundamental difference than in Silicon Valley, I would say.

Daniel Gross [00:24:09] – Interesting. Do you think that Baidu and Tencent then are kind of the exception to the rule? Do you guys feel somewhat alien compared to other Chinese companies which may be more structured?

Qi Lu [00:24:23] – Yeah, I will say, among the internet or technology, IT technology related companies, even though I haven’t talked to a whole lot of them yet, they’re based on, what I have seen so far, largely in the mode that I just described. But when you go out of that range, you go to much more traditional companies. Let’s say, steel industries, or traditional retailers, then you will see more of the Confucianism, hierarchical styles in management. Again, I haven’t done studies. It’s just my perception, I will say. This is how I perceive it.

Daniel Gross [00:25:06] – Today, it feels like, in particular when it comes to AI research, most of the great research is still being done here in the United States. Do you think that will change over time? Will we start to see, 20 to 30 percent of the papers suddenly be published from China? Or will America kind of always be the hub of AI innovation?

Qi Lu [00:25:29] – This is one topic I have a somewhat ongoing discussion with many of my colleagues in China. Our current view is that the very top end of research that’s fundamentally paving new ground, I will say, the example being, let’s say, DeepMind and OpenAI, I think that won’t happen in the next few years, and that won’t happen in China. I wouldn’t say won’t. It’s unlikely to happen. The odds of that type of research happening in China, perhaps will take quite a few years. Right now, we see the research community, particularly the upper echelon, the gap is closing. The leading Chinese universities, the way the gap is being closed is, a lot of those researchers, their pedigree, they’re studying at top tier universities in the United States, whether it’s Stanford, Princeton, and they go back. The gap is closing, but the overall environment, the culture, context, it isn’t quite there yet, meaning that it’s completely driven by your imagination. The social economic surroundings is still not quite the same as the United States whereby you have truly world class people driven by purely the desire to seek knowledge, the desire to unleash imagination. Often these researchers are doing it in the context of personal fames, economic payback. Once you have those, you constrain yourself. You don’t see very far, you don’t pursue the bigger dreams. But our collective belief, it is myself, a bunch of colleagues, friends, we all believe, given enough time, let’s say, in the next five to 10 years, you will see top echelon research work happening in Chinese institutions,

Qi Lu [00:27:35] – and it’s certainly my hope that in the next, somewhere between five year to 10 year windows, we will have equivalent research organizations, let’s say, OpenAI, DeepMind type, that would be truly doing groundbreaking research towards AGI or different types of initiatives that would be at the very forefrontiers of extending the scope of humanness. It will take time, we believe it takes time. Not in the short future yet, but it will happen.

Daniel Gross [00:28:11] – How are you going to nurture that? Are you going to try to create a Baidu research lab that somehow has a different culture around it than what traditional Chinese academia has?

Qi Lu [00:28:24] – There are several organizations, but one is corporate research labs. Baidu is doing quite a bit and our peers, whether it’s Alibaba or Tencent, they are also investing quite heavily in corporate research labs. At the same time, the national labs or the top-tier universities, they are doing more and more. And in the private sectors, there is always ongoing discussion, a new type of research entities can be envisioned and they can be created. There is an ongoing set of ideas being explored. I think it likely will be a combination of corporate research lab, university and some new generation. Let’s say OpenAI of research of the issues will be established over time that will be capable of carrying top-tier research work that’s based in China.

This article was originally published by Daniel Gross via Y Combinator on November 15, 2017