AAAI President Subbarao: The learning process of machines is different from that of humans | CCF - GAIR 2017

Leifeng.com reported that on July 7, the three-day "CCF - GAIR" conference opened at the Shenzhen Grand China Sheraton Hotel. Today's third speaker is Professor Subbarao Kambhampati, who is currently the president of AAAI (American Association for Artificial Intelligence) and a professor at Arizona State University. His main research is in automated decision-making mechanisms, especially in the field of artificial intelligence for artificial perception.

He is also a very well-known professor. He spent a lot of time and especially mentioned the impact of artificial intelligence on human perception and society. He is also a member of AAAI and serves in many international organizations. He provides consulting services for the 2016 conference project and the 2016-2025 AAAI project. Subbarao told us about the rise of artificial intelligence, the differences between artificial intelligence and human learning models, some challenges of artificial intelligence, some thoughts on the ethics of artificial intelligence, and finally introduced his research. Leifeng.com has compiled the content of his speech.

Machine learning processes are different from humans

Artificial intelligence was not taken seriously in the past, but now, more and more media are paying attention to it. Many countries also hope that AI can bring about changes. But the learning process of machines is different from that of humans. Humans observe things in the world, and then they begin to have emotional intelligence, then they have social intelligence, and they also have perceptual intelligence and reasoning intelligence. The process of machines is the opposite. The reason for the difference is that the solution that computers use to solve problems is different from the solution that humans use to solve problems.

Talk about the world pattern

AI can help our children accumulate basic knowledge. In the future, AI reasoning skills will be greatly improved. However, the current AI technology craze may blind us and prevent us from seeing its prospects. Our current AI is not good enough, and we can also have super intelligence. For example, it can be applied to predicting criminal behavior.

How to use artificial intelligence with caution

Systems like criminal behavior prediction are very controversial and must be justified by reasonable, fair, and ethical reasons. AI is now a powerful technology. Many companies have commercialized it, so we need to promote the development of prudent and ethical AI. At the same time, we still have many challenges in the field of artificial intelligence, such as how to make machines understand the meaning behind language, how to make artificial intelligence have a complete understanding of the world, and how to prevent adverse events in human-machine interaction.

How should artificial intelligence and humans interact?

Subbarao also told us what we should pay attention to when it comes to AI and human interaction:

1. We can see positive reports about artificial intelligence, but we can also see negative reports. We must pay attention to negative reports.

2. Conscious artificial intelligence: how humans and machines can collaborate effectively

3. Humans want robots that can perceive and talk to them

He believes that humans have emotional responses to the environment and things around them. This is our emotional intelligence and social intelligence. We need social intelligence and emotional intelligence to make up for these two skills that machines do not have.

In addition, Subbarao also introduced his research, the main purpose of which is to be able to identify human emotions and infer the intentions and expectations of human actions. Finally, he also gave a brief introduction to AAAI (American Association for Artificial Intelligence).

The following is a transcript of Subbarao's speech compiled by Leifeng.com:

I am from Arizona State University. Today I will discuss with you the rise of AI and the development of artificial intelligence. I am very grateful to the organizers for inviting me to China. This is my third time in China.

As you mentioned earlier, I am also the president of AAAI. First of all, I would like to briefly introduce AAAI to you. AAAI is mainly a research and development institution focusing on artificial intelligence. It is also the oldest scientific research institution in the field of artificial intelligence. It was established in 1979. It now has a large number of member institutions. At the same time, AAAI also holds the top artificial intelligence forums and related activities. I would like to mention in particular that China has become a very important member of AAAI. There are many scholars from China. Especially after China joined, the number of AAAI papers submitted and the number of scientific research submissions have far exceeded that of the United States. Last year, the AAAI conference was planned to be held in New Orleans, USA. Because it was during the Spring Festival, many Chinese developers did not attend. Therefore, we modified the time this year, and Professor Yang Qiang, a current member of the executive committee, made a lot of efforts in it. (Leifeng.com Note: In addition, Professor Yang Qiang is also the program chair of the CCF-GAIR 2017 conference, the director of the Department of Computer Science and Engineering at the Hong Kong University of Science and Technology, and the chairman of the international top academic conferences such as KDD and IJCAI. Professor Yang Qiang is the first Chinese AAAI Fellow. His election changed the situation that AAAI was limited to the European and American circles in the past. Professor Yang Qiang has been committed to promoting the internationalization of AAAI, including recommending and helping to invite Rao to give a speech for this GAIR conference. He was a guest at the Leifeng.com Salon and interpreted the historical changes of AAAI .)

The AAAI AI Annual Conference has received strong support from the Chinese AI academic and research community. The number of participants and Chinese members of our AI Annual Conference is relatively small, although their paper submission volume is relatively high, so I also hope that more and more people will participate in the AI ​​Annual Conference in the future, making our annual conference more and more international, so that we can listen to the voices of AI experts from China. At the same time, we also hope that everyone can register as a member of AAAI. The annual membership fee for Chinese members is US$20, and the student fee is US$18, so for everyone, if you join this AI Annual Conference, it is a very simple thing, and the cost is not high.

When you join AAAI, an artificial intelligence organization, you will see artificial intelligence experts from all over the world. You will see some artificial intelligence experts from India. They are also in China. I will also introduce members to each other.

Since I am the president of AAAI, I will briefly introduce AAAI to you. The most important message is that you are welcome to become a member of AAAI.

Today I will mainly talk about artificial intelligence. There is a very interesting topic in artificial intelligence now, which is conscious artificial intelligence. In addition to its own characteristics, what other challenges does conscious artificial intelligence have? In addition, I will also talk about artificial intelligence with artificial consciousness, because this subdivision of artificial intelligence may receive a lot of attention from the academic community in the future, and the research on this issue is also full of challenges.

People are generally saying that AI can't do much, but when I was studying at the University of Maryland, I read a book. After reading this book, I found that AI can do a lot. So later we saw Google. Every day I subscribe to Google's information, and they send me all the information about artificial intelligence. It can be said that among all the articles I receive every day, there is at least one about artificial intelligence. Now we see more and more articles and news, as well as media paying attention to artificial intelligence, and these articles are getting more and more outrageous. This morning I received an article saying whether artificial intelligence is too hot. You can see that there are many interesting situations in the field of artificial intelligence. There are also many companies that are using AI technology to engage in certain businesses, such as shareholder meetings. In 2018, more and more people may talk about AI again, but no one said so before. In 2017, we have begun to hear or see that shareholders' meetings are using AI technology. Of course, governments of various countries are also paying attention to AI. During Obama's administration in the United States, the US government had a future-oriented plan, which also focused on artificial intelligence. Obama said that he hoped that the entire United States could be transformed by AI technology. Of course, we also see the same national rulers, in China, the same, hoping to vigorously develop AI technology and transform China's productivity.

Of course, what is the message brought by the AI ​​craze? It is that a geek like me can have the opportunity to appear on TV and talk about AI technology. In the past, academics like us who do research did not have the opportunity to appear on TV. Now we are invited to talk about artificial intelligence and robots on TV.

What do I want to talk about today? Considering the big background I just talked about, some of the audience here may be AI professionals, and some may be beginners. I have briefly sorted out for you to help you understand why AI has become so popular now. Let's think about human intelligence first. Let's look at babies. When human babies are born, they begin to have some characteristics and behaviors. In the early days, children have the ability to perceive and imitate. They will observe the world, and then they will imitate our behavior. This is a picture made by my wife's student. He used this picture to see his child doing some simple behaviors while eating. In addition, as children grow up, they will gradually look at things in the world, and then they will begin to have emotional intelligence, and then they will have social intelligence. In addition, they will also have perceptual intelligence and reasoning intelligence. For example, when everyone goes to school, they cultivate their own perceptual and reasoning abilities. Of course, in college, middle school and elementary school, we have to take various exams to cultivate and exercise our perceptual abilities. This is how humans develop their intelligence.

Let's look at AI again. In the 1990s, we had an expert system, which was a rule-based system. At that time, many companies began to adopt this rule-based system. At that time, all mainstream companies would use this rule-based system. In the 1990s, there was the Deep Blue computer, which competed with Russian chess players and won. So at that time, computers began to have reasoning skills. In 2000, we saw the upgraded version of Deep Blue. Deep Blue was able to perceive and choose between a cake and a chess piece. At that time, the Deep Blue computer itself did not have the ability to perceive. After 2000, we saw some of the latest AI technologies that gave our AI the ability to perceive. In addition, they began to perform speech recognition and image recognition, and I found that if speech and image recognition are combined with perception, it will be a very hot topic in the future. From the past to the present, humans have gone from perception to intelligence, while machines have taken the opposite path, so machine learning is different from human learning.

Why is the learning of machines or AI opposite to that of humans? Why don't they perceive first and then observe? We must understand that the solution of computers to problems is different from that of humans. When a computer sees something, it will have its own thinking, while if a human sees something, he will describe it and also argue and discuss with others. Therefore, for humans, the human learning model is more suitable for the human brain, while the machine learning model is suitable for machines. Therefore, reasoning is very difficult for machines.

Let's look at the world today. If our machines' perception ability gets better and better, today's children spend six or seven years accumulating, understanding, and learning the world's knowledge. If machines can have perception ability in the future, they can help our children accumulate basic knowledge. In addition, I am not saying that artificial intelligence is not that important compared to humans. In the future, artificial intelligence's reasoning skills will be greatly improved. And not only learning, machines will get better and better at extracting visual information and images. In addition, in the future, we will also train machines to speak, understand, and explain in a chain.

Why is AI technology so popular now? This is because we have found that with modern AI technology, we can use artificial intelligence technology to let our mobile phones see the world we see, and the mobile phone can also feedback the information it sees to us. If we can achieve such human-computer interaction, everyone can use artificial intelligence. Everyone knows that in the latest Go game, if we can use this kind of human-computer consensus or work together, AI technology will be more widely used in the future. At the same time, this phenomenon also has some side effects. We see the use of AI technology in all aspects of production and life. In the cartoon, you can see that when the Deep Blue computer played against human chess players, it said, "This is 1970, don't talk to me about AI, AI technology is not so smart at that time." Now it has come to the other extreme. Every industry is talking about AI, and Deep Blue and AlphaGo, these machines are compared and measured with human intelligence every day. So now everyone has to talk about AI when they see anything, because this is what people are willing to hear, which is also a negative effect brought about by the popularity of AI. I think the popularity of AI technology may blind us and prevent us from seeing its future prospects.

Now of course we can capture images and do speech recognition, but is that enough?

This is Archimedes, a Greek philosopher who said, "Give me a fulcrum, and I can move the earth." Now people say, if you give me a big enough GPU and big enough data, I will create super intelligence for everyone. This super intelligence may look funny, but in fact it shows us a new prospect for artificial intelligence in the future, that is, super intelligence.

Now we must learn from the detours we took in the early 1980s and 1990s in deep learning and big data, and see how to move forward. This brings us to some AI booms. We all know that AI has developed very quickly so far, but some people think that AI has entered a freezing period. Of course, it is very interesting that last year a company said that we can use facial recognition to tell you whether this person is a criminal or a terrorist, so that we can build a machine learning system that can eventually help us predict who may become a terrorist. This is an Israeli company, but I am not sure whether this is a so-called artificial intelligence system. In China, people's IQ is also very high. Last year we saw data from China. How can we use facial recognition methods to help us perceive people's faces and include inference? If you go to the police station, you may see a large number of photos of people, and then look for these data in the cloud, and use deep learning to infer whether these people will commit any crimes. These photos show Chinese people, not that they are all criminals, we just took their photos.

Through machine learning, the accuracy of predicting criminal behavior can reach 95%. It can identify who is a criminal and who is not. It can help people in society to determine whether they are criminals. However, such a system has also caused a lot of controversy. People are discussing whether this is a good application of artificial intelligence and whether such a method can be used in the intelligent prediction of criminals. This article wants us to imagine why we need to talk about ethics in artificial intelligence. I specifically talk about this point. For any complex and powerful technology, we must think about how we should use it and how to use it in a cautious way. When AI is developing, many people report on AI technology. We only talk about the technology itself. Now we also need to explore some ethical methods in artificial intelligence, including ethical principles in the use of technology.

This reminds me of another organization, the Partnership for Artificial Intelligence, which is an alliance initiated by several well-known companies in the world, such as Google, Microsoft, Facebook, IBM, etc. I also joined this alliance as a director of the AI ​​department. The most important purpose of our organization is to help us understand how to use artificial intelligence technology in a socially responsible way through this organization. We have indeed made a lot of technological changes and technological developments, but we do not want to use this technology to determine who is a criminal and who is not, because such an approach is really crazy. At the same time, we can also explore the use of other AI technologies, which may cause the leakage of people's privacy or affect the rights of citizens. From this point of view, this is the reason why we established this alliance, and I am also a director of this alliance. From my personal point of view, we mentioned the importance of safety-related artificial intelligence, as well as a fair, transparent, trustworthy, and accountable AI, and the cooperation between humans and artificial intelligence systems.

We specifically mentioned this point here. If we want AI to decide whether a person is a criminal or not, we cannot find some ridiculous reasons. We must be able to explain to people why it is used in a reasonable, fair and ethical way. This is what I want to mention in my speech today. AI has a very big impact on our culture, and AI has also brought a lot of benefits. But now AI is already a very powerful technology. In the process of development, many companies have commercialized it and carried out a lot of transformation. But we should think about the use of these technologies and how to develop them in a prudent and ethical way. I believe that everyone can better understand us AAAI and our alliance. In fact, we are constantly advocating the ethical and moral use of AI technology.

There is a video like this on YouTube, in which I specifically mentioned the impact of artificial intelligence on social ethics and issues regarding its use. If you are interested, you can take a look at it.

Next, I would like to discuss with you the development of our intelligence, cognition, and sensory technologies, especially the changes in perception and cognitive technologies. Another point is our control and manipulation of robots. From the perspective of manipulation, we can also rely on big data to achieve this. But I would like to ask, have we finished all the research? This is not the case. We still have many challenges. One point here is that most of our current learning and cases are based on machine-based and knowledge-based learning, but the number of cases is still very small. Human learning is generally that you give me a fragment, or I simulate and use it based on my own knowledge. But it is not possible in the artificial intelligence industry at present. Another point is common sense, which I will discuss with you later. Another point is the incompleteness of the world. We can see that all our current understanding of the world is still incomplete. In the process of doing AI, do we need a completely complete knowledge? But for humans, we have no way to have a clear understanding of the world. How can we ensure that AI can have a complete understanding of the world? Another point is interaction, the interaction between machines, artificial intelligence and humans. We hope that no adverse events will occur during the interaction with people. We hope that this artificial intelligence can have close, effective and positive contact with people without causing too many adverse events. I will discuss the research challenges in this area with you later.

First, let's look at common sense. Common sense is something we often mention. What is common sense? My favorite is the case of Magellan. Many people have this common sense. We all know the answer to this question, but for Magellan, on which voyage did he die? I believe that this is the same in both Chinese and English. If you know the story of Magellan's voyage, you can understand this very clearly. There is another point. We can see that from a semantic point of view, it is exactly the same on the surface, but can we find out what the semantic difference is between the same two sentences? Since 2016, we have begun to understand semantics. We started a relatively large competition last year. The final result of this competition was actually not very effective. We hope to really find the semantic changes behind the same language, so that our artificial intelligence can really understand this, and at the same time help us better understand the semantic changes of human natural language that the machine cannot understand. In fact, there are some other changes, and I would like to analyze them with you, just like the question about Magellan we mentioned.

I just mentioned incompleteness, which is also a very big challenge. Because now more and more people are beginning to discuss super intelligence, or the development of supreme intelligence, and the complete intelligence of machines, which will replace the whole world, and humans will become extinct, or some people say that humans will become slaves of machines. I don’t think you have to worry about these parts at all, because I believe that if we have such a killing machine, if it uses an incomplete understanding of the world, it will indeed bring a lot of negative effects. But for us, such a result is very unpredictable. How can we train machines in an incomplete mode? This is very important, and I will talk about this point in the next part.

The last part I want to discuss with you is a very big challenge, that is, how can we ensure that artificial intelligence and humans can interact with each other. This is also what I will share with you in the next ten minutes.

What I want to point out here is that in artificial intelligence, the relationship between artificial intelligence as a whole and humans is very interesting. They hope to help humans, but they don’t really want to see a lot of problems in human-machine collaboration. Think about it, the next time you mention the development of artificial intelligence, we may have spent a lot of time watching our artificial intelligence machines defeat very famous chess players. It helps people while defeating people. Many people mentioned the deterioration of the relationship between artificial intelligence and humans. At the same time, there is another point that in previous articles, many people mentioned that artificial intelligence will replace humans and humans will become slaves of artificial intelligence. But for us, we should pay attention to this, especially for the negative reports about artificial intelligence, we must pay attention.

Last year, I was responsible for hosting the international conference of our Global Joint Conference on Artificial Intelligence. At that time, we established a special team. We mentioned that it should be conscious artificial intelligence. We especially mentioned how people and machines can be effectively combined and work together. This is the most important thing. At that meeting, we mentioned why we should design such a future subjectively, why we should design a future that makes us worry about ourselves, and at the same time, if you think that machines can defeat humans, what else should people do? We should be able to let machines assist people and enhance human capabilities. This requires cooperation from all walks of life. Many people mentioned conscious artificial intelligence. Artificial intelligence or conscious artificial intelligence can really save us from the traditional robot-dominated trend. We believe that negative reports on robots will continue to decrease.

This is the JASON report, which is a well-known think tank and consulting group that provides services to different departments in the United States. I have attended a JASON meeting before, and someone talked about the three major issues we must decide now. JAZON is now conducting a lot of research on artificial intelligence, and at the same time he showed us how we can make efforts in various aspects of artificial intelligence. He specifically mentioned that artificial intelligence developed to enhance human productivity will eventually move towards the development of conscious artificial intelligence. At the same time, we can see that when we predict the future development of artificial intelligence, there is also a White House report that specifically mentions that we must find new algorithms for artificial intelligence with human consciousness.

Now we have entered the era of artificial intelligence 2.0, and we will move towards an era of more human-machine collaboration. Why do we have to make people aware of this? Why do we mention this? Why are we doing this work? I am here to discuss with you the current sociological issues in AI. Whenever human-machine collaboration is mentioned, people feel that it is a process of expectation. Everyone thinks that only people solve problems, while machines stand by and can't do anything. For many young people here, or engineers and technicians, we all know that this is our real machine dialogue. This machine can have very good intelligence, but there is a person behind this machine. He is explaining this robot, but there is a real person hidden inside. This is a lie. Such a lie is actually that people hope that artificial intelligence with human perception will develop.

Are there any other challenges for artificial intelligence? If machines do more and more things, will human capabilities be replaced by machines? Everyone believes that if the capabilities of machines become higher and higher, we will no longer need humans. Let's take a look at human-machine integration or human-machine cooperation. Human-machine cooperation can broaden the depth and breadth of AI technology, and now human-machine integration can also dispel people's concerns. For example, some people worry that robots will replace all humans to do work. If human-machine collaboration is carried out, this doubt will be dispelled. At the same time, human-machine cooperation or conscious artificial intelligence is also a new research direction, which may be considered a very cool scientific research direction in the future. There are many new scientific research challenges in human-machine integration, which cannot be seen from the top or government level. We say that there are many problems in the operation of human-machine integration that need to be solved.

Some say that people need our own brains, not just to watch tigers and lions, but to observe each other. Most of the time, what the human brain actually does is not natural observation, but social behavior, such as listening, communicating and interacting. This is where the human brain and the machine brain can cooperate with each other. Because the human brain has the function of social communication, it can fill the gap in this regard.

(See PPT) This slide is what I showed you earlier. I didn't talk about emotional and social intelligence just now, but emotional and social intelligence is exactly what the human brain has and is good at. We use these two aspects of intelligence to teach children to be disciplined in school and we tell them how to control their emotions. At the same time, we also have social intelligence, which allows us to interact with others better. For example, if we don't have this kind of emotional and social intelligence, during the interview process, there may not be any interaction between the interviewer and the interviewee. If a robot interviews, it will lose the job because it can't interact with its interviewee. In addition, emotional and social intelligence will help artificial intelligence simulate the human brain and gradually have some emotional and social intelligence. Now we already know that if people start not to use AI systems, it is because AI systems do not have social and emotional intelligence. For some Microsoft experts present here, it may not sound so happy, because artificial intelligence does not have social and emotional intelligence, and we know that in Microsoft's office, they will have a robot that will help their employees to do their work. This robot embodies all the latest robotics and artificial intelligence technologies that Microsoft has achieved. However, when Bill Gates announced that he was going to retire, all the employees present had emotions. Some were happy, some were sad, and some might even cry. However, they found that the robots they were working with had no emotional expression at all. This means that if we perform poorly in our daily work and need emotional comfort from our colleagues, the robots that work with us cannot do it because they themselves have no emotions. Therefore, we should now start to consider whether this kind of machine-assisted artificial intelligence can be introduced into the workplace, because they themselves do not have emotional and social intelligence.

Now let's look at the structure of an intelligent subject. We have an intelligent subject. First of all, we need to see that within this intelligent subject, it has brakes and sensors. We can see the world and observe the information of this world. As for the human brain, the human brain is also an intelligent subject, but the way it observes information is completely different. As we all know, we can also simulate the human brain. We will find that if we let AI machines simulate the human brain, we will see that the structure of an intelligent subject in this table becomes very complex and rich. For example, if there is a representation situation, they can not only see what it is now, but also think about what the people around me are thinking at this moment, and whether there is any relationship between what the people around me are thinking and me. The machine is not only interested in the surrounding information, but also begins to associate and associate with the surrounding things because of the new intelligent subject of human-machine fusion. Just like the human brain, our machine and human brain are integrated, and we will pay attention to more and more information about the surrounding environment, including what this person is thinking, what this person will think in the next moment, and whether what he thinks is related to the current environment. In addition, we must continue to trace this information. Therefore, the information captured by the human brain is becoming more and more complex and rich. If we integrate the human brain, we can let the machine perceive and solve real-life problems. If we stay alone behind closed doors, the world around us is very simple, but in social life, we are never alone, so we need social intelligence and emotional intelligence to make up for these two skills that machines do not have.

Let's look at this picture again. In the next few minutes, I will talk about what our team has done in the field of artificial intelligence of human consciousness. We mainly study the different decision-making and learning processes of humans and machines, and then simulate the integration of humans and machines. In addition, we also need humans and machines to work together. Here you can see that we let a person talk to a machine, and let the machine try to understand the intention of the person opposite it, and make corresponding actions based on the understood intention. For example, what does it mean when he opens his arms, and what does it mean when he puts his hands on the table. Of course, we have also invested a lot of money in this project. So far, our technology can enable the machine to understand what its human collaborator wants it to do at a simple level, and understand the human intention, and see whether the person is happy or unhappy through the human expression. In addition, humans will send it different signals, some of which represent happiness, and some of which represent unhappiness. The machine must make corresponding judgments based on the received signals, and it must not only do the actions that humans expect it to do, but also the actions that humans hope it to continue to do in the next link. Because in general, it may be easy for us to understand what we want to do and what others want us to do, but why others want us to do so is a more complicated proposition. Therefore, in the process of human-machine training, we train the machine to listen to our instructions, not only to understand our intentions, we not only use human language to communicate with the machine, but also add gestures and expressions. If gestures and expressions are not enough, we also provide additional information to let the machine understand humans, and let it know what the intention and expectation of doing a certain action is.

For example, we did a similar experiment in the Netherlands, where people saw a ball in front of them and then passed the signal to the machine, which then sensed whether it should grab the ball. Now let's see the machine and the person trying to grab the blocks on the table, and the robot needs to understand what humans want it to do when they start to grab the blocks. So through human-machine collaboration, human emotions and social intelligence make up for the emotions and social intelligence that machines don't have.

Let's look at the challenges. First, we must have a good system for human-machine interaction. As I just said, the most complex problem in the world is that we have to consider what the person next to us is thinking. This is the most difficult proposition in the world. So we have to consider that human thinking is actually very difficult to solve. We have to train the machine to understand what the person next to it is thinking, which is an even more difficult problem. In addition, we must not only ensure that the machine understands what the person next to it is thinking, but also be able to make a clear behavior based on its judgment. Another point is that everyone knows that I am an AI expert, but even as an expert like me, I have no way to develop a robot that can fully understand human intentions and interact with humans. Of course, we all know that if such a robot really exists, it may get along better with humans because it can understand what humans want it to do in this closed social environment, but it is actually very difficult to make such a robot. In addition, we also need the approval of governments of various countries that humans and machines can interact and collaborate at a certain level. For example, is it acceptable to let robots interact with humans in certain special circumstances? In addition, in some sub-industries, such as defense, can humans and robots cooperate. In addition, as you can see in this picture, robots have their own perception of the world, while humans have their own preconceptions about what machines can do. Are these two preconceptions exactly the same? Not necessarily. We must let humans understand that what they imagine machines can do is not what machines actually think in their brains. We must enable robots to communicate better with humans, such as how to get from one destination to another, what is the shortest path in the process of moving forward, and what is the best plan. Humans and machines still have a lot of things to collaborate on in such simple interactions. In addition, there is how to make plans and how to explain the plans in clear language. All these aspects of communication will affect the future prospects of human-machine integration.

I have almost finished talking here. I will show you a short video at the end. Let me briefly review everything I just said. First, I gave you a brief introduction to AAAI. AAAI is the American Association for Artificial Intelligence. We accept papers on artificial intelligence every year. Today, research on artificial intelligence has become a very important part of our social responsibility, and we can also see that China's contribution to artificial intelligence in the world is increasing, because one out of every two artificial intelligence papers we receive is from Chinese scholars. Secondly, I also talked about the differences in learning between human babies and machines. In addition, I also talked about some professional thresholds of current AI technology. Finally, I talked about how to collaborate with humans and machines and cultivate artificial intelligence with human intelligence. Of course, this is also a very difficult proposition for us in the future, and it is also a problem that scientists like me may solve in the future. Because everyone knows that the ultimate goal is not to let machines replace humans, but to let humans collaborate better with machines.