Good afternoon everyone. It’s so great to see you all come to my talk today on a Saturday afternoon. I heard people who can not get into IIT go to MIT. I know why today because you are so eager to learn new things.
Today I’m going to talk about innovation in the age of AI because everyone knows that AI is the major wave these days.
Before I talk about AI, I would like to go over with you what’s happened to the internet after smart phones. Or, how has mobile changed the internet? Of course, this is pretty much a China perspective because I guess most of you are more familiar with the US landscape, but China is slightly different because we have a relatively independent ecosystem in mobile.
The first change is apps becoming isolated islands. What that means is that there are quite a number of large apps that are wrapped around isolated (islands) - the contents, the services are not so easily accessible by search engines or third-party programs. We see that as a trend that more and more apps are doing things independently instead of relying on a search engine.
And the second is that content is linked to an author. What that means is in the PC era, we pretty much interacted with web sites or web pages. We know there’s a webmaster behind the website but we probably never think about directly communicating with the webmaster. But in the age of mobile, content is closely linked to authors, especially on social media.
And even today, especially in China, news feeds, orcontent feeds, are very popular. When you search for things, not only (will) you find the relevant content, it’s easy for you to find the author behind the content. So today, when you find the relevant content you can ask questions and most likely that author will directly respond to your questions. This is increasingly the case for mobile internet.
And the third one is video. Video is becoming the main form of content. We used to see text, then more and more images became available on the internet, and today video has become the most important form of content on the internet. People’s mindsets are also changing toward video content.
Today, if you search for, let’s say, the general relativity theory, you probably would imagine that a Wikipedia entry would come up as the first result. In the case of Baidu, a Baidu encyclopedia entry would come up as the first result. But if you think about it, a video clip, video content, could be a better answer for this query because we can probably find a very good talker, very good teacher, to talk about the relativity theory in a good way, very easy-to-understand way.
You feel you’re connected to the teacher, to the person who created that content, instead of just hard text. This kind of theory is relatively hard to understand. And video provides a lower barrier to entry for this kind of knowledge and content.
So this is what we see during the mobile internet age.
And in the age of AI, search is evolving too. So, how is AI changing search?
We’re also seeing a number of trends. The first is that the first result is typically the right answer. Right now about 60% of queries are answered by the first result.
So, we are increasingly giving direct answers instead of a very large number of links for the users to find the right answer. And I believe this kind of scenario will become more and more popular, or, an increasing number of queries will be answered directly by the first result or by a paragraph of content.
So right now it's like 60%, it will go to 70%, 80%, or even 90%. So increasingly your query will be answered directly instead of going through a list of websites or links. Because, if you think about the search problem, it's essentially an AI problem.
Although, 20, 25 years ago, when search engines became popular, the technology behind it had nothing to do with AI. But search is essentially an AI problem because you basically, humans, express their request, their interest, in the form of queries or text, then we use computers to guess what that human or what user means, then come up with the relevant answer. And if you think about AI, that's pretty much the definition of AI, letting computers understand humans and serve humans.
So solving the search problem is pretty much like solving the general AI problem. It is a hard problem, but we are getting closer and closer.
Then second, content feed blends with search results. What that means - given that in a lot of cases, in most cases, the first result is the right answer, or we can directly answer your question without having you go through a large number of links, so the rest of the links becomes redundant.
We actually don't need to give you a lot of redundant content. So once your query is answered, what we would like to give you is knowledge related to that topic, but not directly on that topic.
For example, if you search for Van Gogh, and the first result is about the general introduction of Van Gogh, then the second one can be a general introduction about Monet. It doesn’t have to have the word Van Gogh in it. Once your question is answered, we can expand the content based on your interest, not necessarily related to your query, based on our understanding of your interests, of you as a user.
In the age of mobile, we actually know a lot more about our users than the PC era, so we can actually extend the user's interests a lot. We can give them more and let users spend more time.
In China, on average every user spends about five hours on the mobile phone (per day) and that's still increasing. People spend more and more time, and for search, we can directly answer users' queries in one shot, so we are giving more and more relevant content to our users.
Then the third, I think many of you already have this kind of experience, the camera and microphone become the new keyboard. You don't have to express your interest in text only, you can express your interest in speech, in images, or in video. If you are interested in a certain plant and wonder what the name is, you can just use your camera and point to that flower and it will tell you. This has increasingly become accurate because of AI.
So if we have to look back for the past 10 years, as we just entered 2020, I think if we need to put a label on the economy, I would call it the internet economy, because internet changed our lives, changed a lot of things over the past 10 years.
It changed payment, food delivery, retail, ride-hailing. And more importantly, I think entertainment.
Internet changed entertainment. Ten years ago most of us spent a lot of time watching TV. Today, I was at a forum a couple of weeks ago, it's about this size, about 400 people. And I asked, who of you watched TV last night, and none of them raised their hand.
Today they spend, you know, five hours playing games or watching short videos just using their mobile phone. They don't watch TV anymore. So the internet fundamentally changed the way people entertain themselves. But, going forward, I think we are entering a new age, the age of AI. So, the characteristics of the economy will also change.
So in the coming decade, I would label it as “intelligent economy”. What does that mean, is that if we can see that internet changed the way we consume, or internet changed the way we entertain ourselves, the intelligent economy will change the way we produce. It will significantly improve productivity for humans.
There are also three layers I'd like to go through. The first one is the new mode of human-machine interaction, the second one is how AI transforms industry after industry, and the last one I'd like to talk about is the infrastructure for AI.
The new human-computer interaction. I think many of you already have this kind of experience. Today, new cars sold on the market are all connected cars, meaning that they are connected to the internet. When you get into a car, you have a screen (that is) bigger and better than your mobile phone screen. You have more expensive microphones, you have cameras, all kinds of sensors in the car, so essentially when you get into the car, you don't need to use your mobile phone anymore.
So you can see that it's pretty much all voice controlled. It connects with all kinds of car services, content, and it responds on a continuous basis. You don't have to use wake words every time.And this is an experience that's already on the market today.
And at home, you will also have an experience that is very different from today's mobile internet.
So when you have a smart display at home like this, chances are that you will use your mobile phone less. If you want to know the weather tomorrow, you ask this kind of smart display and it will answer you directly. But if you want to get the weather report from your mobile phone, you typically need to pull out your mobile phone from your pocket, unlock it, find the right app, and type in the destination. It requires a lot of steps.
But for a voice-first device like this, it's much more direct and more convenient. The barrier to entry is also lower. You don't even need to be literate. You use talk and it will get you the answer.
So because of this, for the past 10 years, we humans are increasingly dependent on mobile phones. I would say over the next 10 years we will be less dependent on the mobile phone, less and less, because wherever you go, there are surrounding sensors, there are infrastructure, that can answer your question, that can serve you. So you don't have to pull out your mobile phone every time. This is the power of AI.
In production, we also have this kind of new human-machine interaction. We call it "digital person". It's essentially a virtual assistant in the form of human, and doing things that complete your task, like this:
Why is this useful? In this case, we’re using it for bank services. A lot of banks can not afford to open all kinds of different branches in many cities. It's very expensive to rent that kind of real estate and hire lots of people. But we can establish this kind of virtual assistant, if you want to open a bank account or if you want to borrow money, or any kind of bank services that require human assistance, you can do that through this kind of virtual assistant.
And we found that people, users, feel more comfortable to deal with a virtual person than a real person. So not only does it save money, save space, it also becomes more user-friendly. You don't have any pressure. You can say whatever you want and do whatever you want.
So all of these are changing the way we interact with computers or machines. And AI is also transforming a lot of industries, in the sense of higher efficiency and lower risk. Let me go through a couple of them.
Customer service. You’ve seen the virtual assistant case for banking, but in many other industries, customer service can be transformed by AI. We've been working with a number of telecomoperators to assist their customer service using virtual assistants. You know in China, I think in India too, a typical telecom operator has like 100 different plans. When a customer calls in, the customer service people can typically recommend a plan that is suitable for that person.
But how do you figure out what's the best plan for that user in one or two minutes? It's very challenging for a real person. But for a virtual assistant, it's actually very easy and quick, and we can use this kind of virtual assistant to do a much more efficient customer service. That's for the telecom industry, and for many other industries we can also find similar cases.
For education, it’s a similar thing. We can come up with a personal tutor, personal assistant, to help students to learn new things. When the student has any kind of questions or problems, we use this kind of virtual assistant to help walk through all kinds of knowledge points and help the students learn.
Also, for the pharmaceutical industry, AI will accelerate the pace of drug discovery. We see a lot ofstartups doing this. Using AI, you can come up with all kinds of different combinations of molecules as drug targets. So you can very quickly generate a lot of potential drug targets and let the biologists, the scientists, to sift through and validate those drug targets.
AI is transforming transportation. This is a very big deal in China because in China we have built a lot of transport-related infrastructure: highways, metros, overpasses. It costs a lot of money. But the software layer of the transportation has not been improved much. In the age of AI, we think that's going to change dramatically. This is a video showing you that.
This is the so-called V2X, vehicle to everything, especially V2I, or vehicle to infrastructure. The roadside units will communicate with cars to improve the efficiency of transportation, avoid blind points on the road, assist self-driving, manage parking.
Apollo is an open source platform for automated driving. But it's not just for driving, I think it's for the whole transportation system. It's going to take many more years for fully autonomous cars to be available everywhere. But before that we can already use AI to significantly improve transportation.
Today, every year, more than a million people get killed in car accidents. We think using AI we can significantly reduce the fatality rate for that. Using AI, if you take over the traffic lights you can in real time get a sense of how many cars are there, which direction are they driving, and at what speed, and you can intelligently remind cars that are at risk using the roadside sensors. You can also in real time adjust the traffic light time so that the whole city works in a harmonious way, that the delay will be significantly reduced.
In a Chinese city called Baoding we took over almost all the traffic lights in that city and we were able to reduce the wait time by 20% to 30% during peak hours, so reduce traffic delays by 20% to 30%.
Now let's talk about the infrastructure. We know that infrastructure is very important. Highways and high speed rail significantly propelled the growth of China's economy over the past few decades, but going forward, I think the infrastructure for AI will significantly propel the speed of innovation. That includes the app development platform,deep learning framework, general AI technology,and chips designed specifically for AI.
At Baidu, we have more than 2,000 engineers working on our AI platform. The goal is to let all the other developers, we have millions of developers, to develop all kinds of applications in a more convenient way, a faster way, and a lower cost way.
For conversational AI, we have DuerOS that's used for smart speakers, smart display, or any kind of IOT devices. For Baidu Cloud, it’s optimized for all kinds of AI applications. Apollo, I’ve talked about it, it's an open source platform for autonomous driving. We now have more than 175 eco partners, including all of the major OEMs, Mercedes, BMW, Toyota, Ford. And for Baidu Brain we provide all kinds of basic AI capabilities such as voice recognition, computer vision, natural language processing, and all kinds of recommendation platforms that we use for mobile content. And PaddlePaddle is the deep learningframework originated from China, like Tens or Flow or PyTorch.
So AI is a big wave, but not every company, not everyone has the power to develop a full-fledged cutting edge AI technology. That's why AI platform is very important and that's why we've devoted a lot of resources to this kind of open source, open platform so that everyone can take advantage of that.
We also use AI for public welfare.
We use AI to help find missing people. In China, we've already found more than 9,000 missing people using AI technology, pretty much facial recognition technology. Even if after aperson is missing for more than 20-years, we had a case, a boy, he was lost at age four and at age 25 he was identified as that missing person.
And we use AI to help the visually impaired people. We've installed the Baidu Xiaodu smart speaker in a lot of the blind massage parlors. Those massage therapists who are visually impaired can use voice to control air conditioning, control the curtains, control a lot of IoT devices, which makes their life much easier.
So AI can be used in a lot of these public welfare cases.
I also have a claim, AI will make you immortal. What does that mean? It means that machines can become smarter and smarter, can learn from humans. And today, storage has become cheaper and cheaper, and we can afford to store a lot of personal information.
For example, I make a speech here and it is video-taped, it can be stored for a long time. And your voice can be stored, your video can be stored, your text, your articles, everything about yourself can be digitized.
And later on, based on this kind of digital information or content, computers can learn how you think. So after a while, it's not hard to imagine when Tim Cook wants to evaluate whether Apple should work on an autonomous driving project, he can actually ask Steve Jobs, the digital copy of Steve Jobs, if that’s a good idea. Because there is a lot of information about Steve Jobs stored on the internet, and computers can learn the way Jobs thinks. So this makes Jobs immortal. But it's not just Jobs, anyone, anyone’s information can be stored, can be learned, and made available when necessary. So in a sense, AI will make you immortal.
That's how fascinating innovation is, that’s how fascinating AI is. India is one of the fastest-growing smart phone markets in the world, and India is also a very large developing country right next to China. We’ve seen fast growth for both countries over the past few decades. And I think for next decade, there will be more opportunities for us. So we at Baidu are very much looking forward to working with Indian institutions to make a better world through innovation. Thank you all.
在谈 AI 之前，我想和大家回顾一下智能手机诞生之后互联网的演变，或者说移动时代是如何改变互联网的。当然，在座的各位可能对美国的情况比较熟悉，我将主要从中国的角度进行回顾。中国的情况与美国有些不同，因为我们拥有相对独立的移动互联网生态系统。
首先，应用程序正在变成一座座孤岛。也就是说，很多大型 APP 已经成为相互孤立的状态，它们的内容和服务无法通过搜索引擎或第三方程序方便获取。我们认为这是一种趋势。越来越多的应用程序开始独立运行，不再依赖搜索引擎。
进入 AI 时代，搜索也在不断发展变化。那么，人工智能究竟如何改变搜索呢？
这就是所谓的 V2X，vehicle to everything，尤其是 V2I，vehicle to infrastructure，车路协同。路侧设备将与车辆进行通讯，以提高交通效率、避免道路盲点、协助自动驾驶、管理泊车等等。
在对话式人工智能系统方面，我们拥有用于智能音箱、智能屏或任何物联网设备的 DuerOS。我们的百度智能云也针对各种人工智能应用进行了优化。就像我刚刚提到过的，Apollo 是一个自动驾驶开源平台。现在，我们已经拥有超过175个合作伙伴，其中包括各大汽车主机厂商（OEM），比如梅赛德斯、宝马、丰田、福特等。通过百度大脑，我们提供多种基础人工智能能力，例如语音识别、计算机视觉、自然语言处理以及用于移动端内容的各种推荐平台。飞桨（PaddlePaddle）是源于中国的深度学习框架，类似于 TensorFlow 或 PyTorch。
以后，根据这些数字信息或内容，计算机可以模拟人类的思维方式。因此，不难想象，再过一段时间，如果蒂姆·库克（Tim Cook）想评估苹果是否应该开展自动驾驶项目，他就可以询问史蒂夫·乔布斯（Steve Jobs），或者说经过数字化的史蒂夫·乔布斯。因为互联网上存储了大量有关史蒂夫·乔布斯的信息，所以计算机可以模拟出乔布斯的思考方式。通过这种方式，可以让乔布斯永远地活下去。不仅是乔布斯，任何人的信息都可以被存储，被计算机学习和模拟，并且在需要时进行信息输出。因此从某种意义上说，人工智能可以让人永生。