Blogs

AsiaIndustrial NetNews: Hong Xiaowen, Dean of Microsoft Research Asia, was interviewed by The Paper during the Boao Forum for Asia.

With the rapid development of artificial intelligence (AI) technology, human concerns about artificial intelligence are also intensifying: Will artificial intelligence replace human work? Highly intelligent artificial intelligencerobotWill crime become a factor of social instability? Will the future world conflict break out between humans and artificial intelligence?

Hong Xiaowen, Dean of Microsoft Research Asia, is optimistic, “If human beings need to worry about being wiped out one day, it must not be artificial intelligence (doing it), but only aliens or higher species. Just like human beings dominate. The earth has wiped out many other species.” Hong Xiaowen said in his concluding speech at a roundtable at the 2017 Boao Forum for Asia Annual Conference, “Don’t think of people and AI as opposites, the relationship between people and AI: one is People, one is a thing”, “If we want to create an artificial intelligence with life, consciousness, and emotions, it is better to have children.”

At this year’s Boao Forum for Asia, as an expert in the field of artificial intelligence speech recognition, Hong Xiaowen was invited to attend a number of discussions on artificial intelligence (AI)-related topics. During the event, Hong Xiaowen accepted an exclusive interview with The Paper.

In the interview, although Hong Xiaowen was tired, he still spoke at a faster rate when talking about topics of interest, which could hardly hide his excitement, and the original 15-minute interview was extended to half an hour.

Speaking of AlphaGo’s defeat of South Korean Go player Li Shishi last year, Hong Xiaowen said that this kind of game is not fair. As long as there is a lot of calculation involved, people probably can’t compare with machines. If people must compare their intelligence with machines, they should compare algorithms, not algorithms and calculations.

“When a person competes with a computer, do you feel very depressed when you lose? You don’t feel that at all. It’s like when a person competes with a car. You don’t need to be depressed at all.” Hong Xiaowen said with a smile.

Hong Xiaowen said that artificial intelligence will indeed bring about changes in the workplace in the future, and they will replace manual labor to complete many difficult and repetitive tasks. But the key to AI’s inability to completely replace human intelligence is creativity. “Because the most amazing things about people are related to creativity, and in terms of creativity, the progress of AI has no shadow at all.”

However, Hong Xiaowen also said that because big data and machine learning are more and more widely used, it has also brought adverse effects on the richness and diversity of human life. He said that the data learning of artificial intelligence always tells us what is the most suitable and efficient, and the result is less diversity.

Hong Xiaowen also repeatedly emphasized in the interview that only the power of artificial intelligence can form the so-called “super intelligence”. The primary goal of human development of artificial intelligence is to be controllable and useful. “We and computers can do things that we couldn’t do before. I am 100% convinced that people and machines can create better and more beautiful things.”

Currently, Hong Xiaowen serves as the global senior vice president of Microsoft, the chairman of Microsoft Asia Pacific R&D Group and the dean of Microsoft Asia Research Institute.

Hong Xiaowen is also a distinguished chief scientist of Microsoft and an internationally recognized speech recognition expert. He has worked at Microsoft for 20 years. From 2005 to 2007, he founded and led the Microsoft Search Technology Center (STC), which is responsible for Microsoft search products (Bing). ) development work in Asia Pacific. Previously, Hong Xiaowen worked for Apple and led the team to develop the Apple Chinese Translator.

1 2 3 4 Next > page

The following are excerpts from the interview:

The Paper: Last year, when Lee Sedol and AlphaGo fought against each other, you may not have realized how powerful AI is in playing Go. But now, especially after a series of matches between Master and Go masters, many people believe that AI already has an absolute advantage in the field of Go. In what areas do you think humans are less able to challenge AI in the future?

Hong Xiaowen: To challenge this thing, I think in fact, as long as one competition requires a lot of calculations, it is unfair to compare people with AI. When you talk about Alpha Dog, the Alpha Dog game is actually not very fair. Because you think Li Shishi needs an algorithm, how to play this chess, and then he has to calculate. Algorithms and calculations must be done by Li Shishi himself, AlphaGo’s team, the algorithm is from a group of people, and tens of thousands of machines are used there to calculate. There is a time limit to playing chess, so this is not very fair.

Therefore, as long as a lot of calculation is involved, people probably cannot compare with machines, and people will definitely lose if they compare with calculation. It’s a bit like when you race a car and fly a plane, it doesn’t make sense. If it is said that people must compare their intelligence with machines, they should compare algorithms, not algorithms and calculations. So I think any game that can set (scope), including one called Jeopardy a few years ago (Editor’s note: Jeopardy is a CBS quiz game show), ask you a question, in There are several answers to choose from, this kind of competition that has standard answers and can be found by means of big data, people basically have no hope.

But don’t feel inferior. As I said just now, when people compare calculations with computers, will you feel very depressed when you lose? You don’t feel bad at all, just like when people race with cars, you don’t need to feel depressed at all. This also makes us think about the definition of intelligence by human beings, which in fact changes with time and place. I often give an example. When I was a child, if someone had a quick mind, they would think he was a child prodigy. But today no one thinks this is a child prodigy, because it’s okay to count on a computer. In fact, it will become like this after playing chess, such as Go and chess. Now few people think that playing chess is really great wisdom, because it requires a lot of calculations.

The Paper: I have seen a foreign prediction that by 2050, the ability of AI will cover all types of jobs for humans, and now it has covered 59 types of jobs. What do you think of the possibility of AI replacing all occupations of humans?

Hong Xiaowen: I don’t know how this is calculated, I think it means AI or the wholeautomation, will indeed change the workplace. So what kind of things in the future I think AI will do better than people, and people are reluctant to do it, I think there are two categories: the first category is this kind of perception, such as speech recognition, computer vision, this kind of People don’t have to think about it, for example, when I see you and know it’s you, I brush my face. If you take it a little more broadly, today, for example, when going through security inspections, look at what is in the bag and whether there are dangerous items. This machine can definitely do better than people. And if people have a choice, they don’t want to do this kind of work, it’s boring and hurts the eyes. Then this kind of work, I think AI will be a good choice.

Then the other is that you have to do one thing again and again, you can also call it an expert system, you can collect data and you can optimize it. Let’s take an example of what is called preventive maintenance in a factory. The production line of the factory is very complicated, and the machine is also very complicated. You have to wait until it breaks, and then find out the reason, and you have to repair the machine, which may delay the downtime. For example, when the elevator breaks down, it takes a few weeks to find someone to repair it, send someone to come every few days, and then figure out what is broken, buy parts, and spend several weeks. Now, using the internet of Things and AI, you can upload this data at any time, including the sound it makes, the degree of lubrication, and the speed. After the data is uploaded, a model can be built to know what kind of eigenvalues ​​it is about to break, so that preventive maintenance can be performed. This will be of great value to the industry. I think this kind of work can be done by AI plus IoT and increasing data.

If we think of this kind of thing on people, our own body is better than prevention, so why are there many wearable products now, if we can transmit this data to the cloud all the time, use AI to analyze it, and also That is to say, before you have any problems with your body organs, you can know this and go to the doctor as soon as possible.

So, for example, I think in the future it is very likely that you will know after the first cancer cell grows in you, and go to the hospital as soon as possible. Cancer is a good example. The earlier it is detected, the higher the cure rate, the lower the cure rate if it is delayed.

So I think AI can do these jobs. As for replacing people, there are still many things that cannot be replaced. Because the most amazing thing about people is related to creativity, the progress of AI in creativity has no shadow at all. People’s creativity is played in various places every day, and many creativity are not repeated again and again, and it is impossible to collect data. Like I need to make an insightful judgment, for example, as an investment, we at Microsoft were considering whether to buy LinkedIn last year (linkedIn), when to buy, how much to buy, these things cannot be done by AI. But AI can help you analyze a lot of data, and then you can make decisions based on this thing. Not to mention a lot of things related to humanities and dealing with people, because people are unpredictable, and it is difficult for you to collect a large amount of data or collect very reliable data. At this time, I think it is difficult for these jobs to be used by AI replace.

page

The Paper: Is there a possibility that AI will be creative in the future? Because I read a foreign report, there is now a kind of AI that can write AI programs and code its own deep learning. If such a situation occurs, do people need to be vigilant?

Hong Xiaowen: This report may be a bit exaggerated. Today, there is nothing you can write your own programs. And the so-called program can be written means, do you write a program under my advice, or invent a new algorithm yourself? Because you don’t know where the next idea (idea) is, I solved a problem today, I don’t know how to solve the next question. How can a computer know something that I don’t know myself, this is problematic from a philosophical point of view. Because the hardware of any AI machine today belongs to the hardware, and the software is designed by us humans. Even if you make a piece of software today and say you can make this machine write a program, it is you who make that software.

And what creativity is, no one can understand. Creativity does not depend on big data, but mostly depends on small data, or even no data. I take Einstein as an example. Last year, gravitational waves received a lot of reports. The concept of gravitational waves was proposed by Einstein more than 100 years ago. After more than 100 years, the progress of science and technology is so great that we can barely measure a little data of gravitational waves, then Did Einstein have data more than 100 years ago, I can almost guarantee that there is no data, this is creativity jumping out of a rock, what is that? If we don’t know ourselves, how can we write a program, and then this It is impossible for the program to figure out gravitational waves by itself, from a philosophical point of view. To make a conclusion: we don’t know what creativity is, and we don’t know where the next creativity is. So for this reason, for computers to be creative, I think it’s too early.

Surging News: Imagine another possibility. With the popularization of AI in the future, will we lose a variety and richness of human life. For example, driverless cars replace human driving, and we will lose this kind of The pleasure of driving. What do you think of views like this?

Hong Xiaowen: Now there are a lot of AI big data analysis and machine learning, for example, watching movies and TV, if you like this kind of thing, I will push this kind of thing to you, the news is the same, China is called today’s headlines, there are similar so-called abroad social application. The result of this is that in the end there is less diversity.

This is not only in terms of diversity, but also in terms of social development, which is actually a problem that needs to be worried about. Because we need everyone to listen to different voices in this world in order to make the society more harmonious. Today, if every one of these circles is incompatible with each other, it will cause social opposition. This is actually a society The thing to worry about is diversity.

Yesterday, we had a discussion on the forum. Someone said, let’s look at the relationship between the two. This constellation and blood type are suitable for that constellation and blood type. Finally, we found the right person according to the big data analysis, but we also lost the way to distinguish the difference. Attributes of the possibility of friends of the opposite sex.

Is it a good thing that this society has become more homogeneous? But I think this is not only a matter of technology, but we need to discuss with people who study humanities and psychology how to create a more harmonious society that makes everyone happier.

The Paper: Everyone is talking about AI now, but our own personal experience is that AI is still a little far from our lives. If we say that AI is deeply penetrated into everyone’s life, just like today’s smartphones, we still need to How long will it take? What do scientists and businesses need to do in the process?

Hong Xiaowen: It should be Bill Gates who said, “We often overestimate what can be done in 3 years and underestimate what can be done in 10 years.” Just like AI, it seems that everything can be done now. Absolutely overrated. Even if we’re talking about AI that isn’t very creative, mistakes are often made today. Whether it is speech recognition or computer vision, it still has an accuracy rate, even if it is 95%, there are still 5% wrong. Some things are not allowed to have a 5% error. For example, autonomous driving is a good example. Who would dare to take a 5% error?

In addition, I think that people overly deify AI and also demonize it, because they think that human and AI are in a state of opposition. I think that if man and machine are opposed, we would have to worry a long time ago, we can’t run with the car, we can’t run with the plane, but the relationship between us and the car and the plane has always been that we and the car can run together Farther, we can fly to the sky together with the plane. Likewise, we and computers together can do things we couldn’t do before. I am 100% convinced that man and machine can create something better and more beautiful.

I think everyone has heard, “Bold assumptions, careful verification”, bold assumptions are a kind of creativity, but not everyone is right, most assumptions are wrong, bold assumptions must be carefully verified, be careful when verifying Big data and AI can help you to verify, the verification is fast and accurate, and you can immediately change the bold hypothesis for you to try.

For example, before there was no AI, there was a well-known “four-color problem”, which was put forward more than 100 years ago: any map, which has different shapes between neighboring countries and neighboring countries, can use four colors? Color, to ensure that no two adjacent countries use the same color. It took mathematicians nearly a century to prove it. Thirty years ago, a mathematician and a computer proved it. This is a good example. Humans can’t prove it, let alone computers, but humans and computers can do it. This is the performance of creativity.

So I think everyone is willing to accept this concept. I think AI plus HI (human intelligence), artificial intelligence plus human intelligence, can produce super intelligence. From this point of view, I think there is a very harmonious relationship between humans and machines, and it is not only the case with today’s computers, cars and planes are not like this, so I don’t think people need to compare people with machines, if that’s the case , people want to be very depressed. How can we compare our muscles with a large man-made tractor? Since we can’t compare, why should we compare calculations?

The Paper: What is the relationship between people and AI? People and tools?

Hong Xiaowen: I think all man-made things are tools, and the purpose of these tools is controllable. From a philosophical point of view, from a usage point of view, why do we build something that is uncontrollable. If I ask the machine to get me coffee, I expect it to get it for me ten times. But if I ask my wife to get me coffee today, five times out of ten she’ll tell me why you don’t make it yourself, and I’ll say “yes, you’re right”. This is the relationship between machines and humans. But creatures are different. Dogs and cats have life. We are different from machines, and we will have feelings for them. Our child is a good example, we want him to be smarter than us, but we can’t control him. So half joking, if one day the robot is out of control, it has its own temper, and we also have thousands of years of experience with people, so we should be able to adapt.

The rest of the full text page

The Paper: The AI ​​field is now divided into many technical sub-categories. For example, Microsoft Research’s products “Xiaobing” and “Xiao Na” use speech recognition as the technology entrance. So among all the subdivisions, which one is the most basic?

Hong Xiaowen: I think it is very appropriate to answer this question with the cognitive services that we do in Microsoft Cloud. There are several pieces in it, the first one is called speech recognition, and there is another piece in speech recognition, which you generally don’t see, we call it natural language processing, because the speech is just the sound I heard, but it becomes a language, and this language represents What do you mean, what do you want me to do, to do natural language logic. This logic is very important, because the most thing we do every day is to speak, and we want to give an order to do something, and we express it in the way of speech language.

The third part is called computer vision. We need to understand everything. The eyes are too important. The eyes are the window to the soul. Today, we say that a person can’t see the eyes. We think it is a very big defect. To dangerous things, but also allows us to better grasp the space, so computer vision is also a very important piece.

There is another piece called knowledge. Knowledge is no longer just language. Today, when we talk about Beijing, we know that Beijing is the capital of China, and it is also an ancient capital. These things are no longer just language. This is called knowledge.

The last piece is called the search API, because search is actually a way to let us not need to write down everything in our heads, but to be able to take the knowledge of the world and organize it, so these are actually a very important axis for the future development of AI. .

The Paper: Besides AI, what other forward-looking technologies are Microsoft Research Asia focusing on?

Hong Xiaowen: AI is of course very important, but for many other things, you can at least say that it is the hero behind the scenes. In the subject of computer, it has always been defined in this way, such as system, and the so-called network. Don’t forget, even if you have AI, your computer, whether it is hardware or software, can’t keep up with the calculations, and the network can’t be connected smoothly, and it won’t work.

When it comes to systems, it also includes hardware, like chips, how can it become faster. There is also a computer theory, there are some basic algorithms, and many algorithms are increased in a proportional series. At this time, how can you come up with a new algorithm that can make it faster in the same time? So this is also a very important thing.

There are also AR (augmented reality) and VR (virtual reality), which are neither cognition nor perception, but imaging, so that we can see the combination of virtual and reality, and then we can do it naturally, even a little You can also do naked eyes. Multimedia, AR, VR, MR are also important technologies in the future. There are also things like secrecy, which are related to how to maintain the privacy of users. AI belongs to AI, and there must be something that people can trust. How to protect people from network attacks, how to help users have a more secure and more credible system.

AI is just an algorithm, the processing of big data, and the visualization of big data. Microsoft has a core product called power BI, large-scale data, we just talked about AI, HI, a very important thing is how to rely on the interface to make it easy for people to try his bold assumptions, and then from the data to be able to. Seeing what I want, I create a scenario today, and I want the analysis of the data to let him see the result. Visualization and big data processing, you can also count it as AI, or it can be said that it is not AI, this aspect is a very important research project. This thing is also researched a lot in our research institute.

The Paper: Not long ago, Baidu Chief Scientist Wu Enda left Baidu, and the next day Tencent announced that it had recruited an AI leader Zhang Tong from Baidu to take charge of Tencent’s AI laboratory. Will the poaching of talents in the AI ​​field become more and more intense in the future? As a place where scientists gather, is Microsoft Research worried about the brain drain?

Hong Xiaowen: Rather than poaching, it is better to say that good people have their own ideas and choose the path they want to take. In high-tech industries, whether in China or Silicon Valley in the United States, the turnover rate of talents is probably above 25%. So I think this is the norm, not the new normal. Because high-tech is the most prosperous industry in the past 10 or 20 years. Of course, everyone has the most intense requirements for talents. Of course, they also strive for good talents. On the contrary, as the tide rises, there are many opportunities for good talents.

So I think it is natural to look at this matter with a normal heart. When reported by the media, don’t forget that it is a collective achievement for a thing to be so prosperous. You will see the so-called leaders, including myself. There are more than 200 people in the research institute. I am the one who speaks the most time, but it does not mean that I do everything in the research institute. We Chinese people have long said that “one general will succeed and ten thousand bones will die”, and only these famous generals are remembered in history. But it is absolutely impossible for Microsoft Research Asia to rely on me alone, and the loss of talents is very normal.

Microsoft Research Asia has been in China for the longest time. You just mentioned the other two companies. We are often praised, saying that we are the “Whampoa Military Academy” in China’s IT industry. This matter was reported a few months ago, 1 year ago, 3 years ago, 5 years ago, even 10 or 15 years ago, because many of our early scholars and experts, including our founding dean, Mr. Kai-Fu Lee , Zhang Yaqin, etc., have left us for more than 10 or 15 years. The fact that people reported that we are “Whampoa Military Academy” is no longer in the past tense or the present perfect tense, but in the present tense. From this point of view, it also means that we are still very pleased that we can introduce new talents, and cultivate new talents, and new talents can also become leaders when someone leaves. Therefore, it is more important that each company still Think of it this way.