Dialogue motivation The emergence of artificial intelligence and autonomous systems has not only challenged the human understanding of nature, but also challenged human understanding of itself, including the legal risk of unthinkable reality. How far is artificial intelligence from our lives? In the age of artificial intelligence, what risks will we face? How to prevent it? Around these topics, the "Legal Daily" reporter initiated a dialogue with industry researchers. Dialogue person Tencent Researcher Cao Jianfeng Legal Network reporter Zhang Wei What is artificial intelligence? Reporter: For artificial intelligence, many people still know about sci-fi movies and alpha dogs. Conceptually, what kind of existence is artificial intelligence? Cao Jianfeng: In the popular understanding, artificial intelligence is the task that a machine can accomplish when humans need intelligence. In the past, we believe that only humans on the earth have intelligence. Human intelligence includes rational thinking, language ability, logical reasoning, and emotions. It can solve problems, have their own goals, and can perform actions. Of course, the boundary of the word "smart" is still controversial and changing. At present, there is no consensus in the scientific community on the concept of artificial intelligence in artificial intelligence (AI) and autonomous systems (AS). The American "artificial intelligence textbook" is based on human beings. It has a general definition of artificial intelligence, that is, thinking like a person, acting like a person, thinking rationally and acting rationally. There are some other definitions of artificial intelligence, such as setting goals for yourself, and then developing a series of plans to implement them. The intrinsic technology or core of artificial intelligence uses some deep algorithms, such as neural network algorithms, to simulate the human brain structure, thereby simulating the process of human thinking and decision making. There are also deep learning, genetic algorithms, evolutionary theory, etc. Deep algorithm technology supports the robot to make certain decisions. The difference between artificial intelligence and previous computer programs is that it has autonomy. It only needs to give it some of the most primitive data. It can itself come up with a conclusion and make its own decisions and judgments. It has the ability to learn. This is autonomy. Sex. Reporter: We have contacted more early artificial intelligence, should be regarded as chat robot software. It will conduct language learning through the use of all users, "evolve" itself, and gradually conduct more extensive dialogue. Of course, the chat bot software also learns swear words, and the user tells it in the conversation that "What you said is swearing," and it corrects itself. In addition to this kind of chat robot software, what artificial intelligence applications do we encounter in current daily life? Cao Jianfeng: Foreign big software companies are all launching their own chat bots. At present, this kind of application is still relatively preliminary. In addition, such as a self-driving car or assisted driving, the driver does not need to press the steering wheel, and can autonomously complete parking, change lanes, and turn. Artificial intelligence applications include hotel service robots, rescue robots, personalized recommendations, and more. What are the legal risks? Reporter: In the film and television dramas, we will see a lot of questions about artificial intelligence and human ethics and law. With the development of artificial intelligence technology, will we encounter these problems in real life? Cao Jianfeng: The AI ​​system has become more widely used and has entered many aspects of social life. The first is the issue of security. From a legal point of view, it can be transformed into a responsibility issue. For example, will drone damage our property and personal safety? The driver does not need to drive by himself and is more like a passenger. If the driverless vehicle malfunctions and causes damage to the driver, can the driver be compensated? In the hybrid driving mode, the driver and the car need to cooperate and a car accident occurs. Does the manufacturer need to take responsibility? The distribution of responsibility between drivers and manufacturers is still relatively vague. For another example, how to ensure that artificial intelligence complies with social norms and social values, and that it will behave like a human when it comes to risk. For example, the rescue robot, how to choose between the old and the weak, to save who to rescue who, these need to be considered in advance. There are privacy issues. Artificial intelligence develops rapidly, and it makes decisions that require large amounts of personal data to exist in artificial intelligence systems. The artificial intelligence system performed "portraits" for us to form a "sense of impressions." We may not even know that we have been pictured by these software or systems, and we have no way of knowing whether these pictures are true or not. Or rather, the system gives us some labels. We may not know what kind of labels it is giving us. Are these labels wrong or have some problems? When these labels are transferred to other occasions, they will have a Impact, for example, may affect our application for loans, job hunting, etc. We are unaware of this issue because the information it uses is not open and opaque. Reporter: In other words, the artificial intelligence system may collect one-sided information, which will affect people's rights in the future. Cao Jianfeng: What kind of information the artificial intelligence system collects and how this decision is made is beyond our control and more often than not. For example, some companies in foreign countries use AI to conduct personality tests when recruiting. If the system thinks that candidates will suffer from a certain disease or pregnancy in the future, it may cause the applicant to be rejected and violate the law prohibiting discrimination. Another problem is algorithmic discrimination. This is a more pressing issue. In the past, a lot of work was done by humans and now replaced by artificial intelligence systems, such as personalized recommendations. How to control the risk Reporter: Should we deal with the legal issues that may arise with artificial intelligence? Cao Jianfeng: It may be too early to discuss some issues, such as whether the intelligence of artificial intelligence will surpass humans and get out of human control. At present, it is still relatively far. However, some of the more specific issues still require early attention. For example, driverless cars, hybrid cars, will soon enter our lives, and the legal issues involved need to be considered in advance. Reporter: Some of the issues mentioned just now include the threats and liabilities of property and security, the observance of human laws, and algorithmic discrimination. Should these all become legal concerns? Cao Jianfeng: Yes, many aspects are legal issues that can be solved by law. Ex-ante supervision is mainly about R&D. If human society's ethics and value norms are not taken into consideration, artificial intelligence applications may make decisions that are contrary to human expectations in the face of dangers and accidents after they are put into the market. The industries in the world are concerned about the research and development of artificial intelligence. That is to say, whether the design development process should have some ethical codes, including the AI ​​system that complies with human ethics and values, ensures safety, and the public trusts in R&D. First, a written law must be followed, and writing a law into the code requires some methodologies. When faced with a crisis, what kinds of good and evil, and right and wrong, should be observed in artificial intelligence, and these common ethical rules, social values, and values ​​shared by some groups must be coded. In the R&D process, programmers may not be technically flawed, but there may be loopholes in the social field. Once the application may result in discrimination or security risks, it is necessary to establish a review mechanism of the ethics committee, from the social norms and values. From the perspective of reviewing whether the algorithm and logic are consistent with human norms and values, and giving advice from a professional perspective. Reporter: It can be seen from the dialogue just now that the principle of openness and transparency is very important, but how can this be resolved? Cao Jianfeng: There are several solutions to the problem of information transparency. The first is the artificial intelligence open source movement. Open source software, as the name suggests, allows everyone to share source code, review software, and find problems in a timely manner. Alternatively, the artificial intelligence system may not need to be disclosed to everyone, but the source code and other materials need to be disclosed to the approved regulatory authorities. Does the vehicle consciously “law-abiding†if it assists driving automatically? Narrator: Zhu Hong, Beijing Citizen On the afternoon of July 15, 2016, I drove the car to Beijing Fourth Ring and started the automatic driving assistance function. In the automatic assisted driving mode, the steering wheel of the vehicle will automatically rotate according to the direction of the road to drive the vehicle to turn. If the line is to be synchronised, the owner will automatically turn on the line when the distance is allowed after the driver has activated the lamp. When parking, after an empty parking space, stop and hang reverse gear, the screen will prompt to see the parking space, click on the "auto parking" button, the vehicle automatically goes back into the parking spaces. In the automatic assisted driving mode, the interaction between the vehicle and the driver is mainly presented on the dashboard directly in front of the driver's line of sight. After the car has recognized the lane line, or recognizes that the preceding car is following, the driver can open the “speed limitâ€. Speed ​​limit is the number of speed limit signs on the side of the road detected by the camera. The car will immediately increase its speed so that it can travel as fast as possible without hitting the speed of the vehicle. After that, the driver can start the automatic driving assistance mode. When entering the automatic assisted driving mode, the screen will prompt "Please keep your hands from leaving the steering wheel". About 70 kilometers around the Fourth Ring, I divided into two sections. In the first paragraph, I started from Siyuan Bridge and went to an interactive design company near Sihui. In the second paragraph, I set off from Sihui to the south and walked around the Fourth Ring back to Jiuxianqiao. Let me talk about the road conditions. I encountered obvious congestion at Sihui, Xueyuan Bridge and Wangjing Bridge. The West Fourth Ring Road was open. This time around the Fourth Ring took a total of more than one hour, it was relatively smooth. In the process of assisting driving around the ring, I was scared out twice. The first time there was a malicious line of uncivilized driving of the vehicle. At that time, the assistant driver noticed that the car next to him was suddenly jammed in. The car had already begun to slow down. I personally “playedâ€. The second time I started from the Sihui Bridge to the Jingtong Expressway from the Fourth Ring Road, where several roads converged, there were ten small plastic railings with a height of half a meter. The cars in the automatic driving assistance did not seem to detect these plastics. Small railings, open directly to the small railings, I immediately felt the brakes. The fourth ring was "opened". In order to safely follow the traffic rules, I forcibly interrupted the automatic assisted driving for nearly 10 times. When I encountered other vehicles that were uncivilized, I forced the brakes more than 20 times. In a dangerous situation, I took the initiative to interrupt the automatic driving assistance nearly 10 times. In the meantime, the first time I took a careful look at the cars around me, I discovered that there were so many cars that violated the traffic rules. Crimping lines accounted for two lanes. In the automatic assisted driving mode, the vehicle analyzes the white line of the road; when there is a preceding car that can follow, priority is given to the previous car, and the white line is used as a reference. When encountering the front car casually pressing line, the vehicle in the automatic assisted driving mode will also follow the arbitrary pressure line, closely following the illegal pace of the traffic in the preceding car. The car follows the lane when it is assisted driving. The automatic assist driving mode is ideal for clear lanes and is not very reliable on the roads and construction sections. In peacetime, I don't need to rest while driving around the ring. Open the automatic driving assistance mode, but because of the concentration of energy, I need to ease the spirit. I parked on the roadside 3 times. Regarding automatic driving assistance, my feeling is that this feature is available, but the driver cannot distract. Auto-assisted driving also requires a process of mutual trust between the driver and the machine.