Ming Ye, Master's degree student in the School of Humanities and Social Sciences of Xi'an University of Engineering
, He Wanli, Associate Professor in the School of Humanities and Social Sciences of Xi'an University of Engineering
Content Summary: The criminal law community's criminal responsibility for artificial intelligence is divided into two perspectives: it should be given the subject status and it should not be given the subject status. However, due to the gap between the theoretical knowledge and legal knowledge of artificial intelligence, some scholars find it difficult to analyze the dominant position of artificial intelligence from a technical level, and discuss the issue of criminal responsibility of artificial intelligence through analysis and research from the technical level and the development of the law.
Artificial intelligence technology is in full swing, and the 21st century is the century of artificial intelligence technology. Since the 18th National Congress of the Communist Party of China, the development of artificial intelligence technology has been unstoppable, and the window for a new round of scientific and technological revolution and industrial revolution has been opened. The development of artificial intelligence technology not only brings new challenges and opportunities to science and technology, economy and politics, but also creates many problems worthy of study at the legal level. Especially in the criminal responsibility of artificial intelligence, the criminal law community has set off an unprecedented wave of discussion. Criminal law scholars divide artificial intelligence into two different levels: weak artificial intelligence and strong artificial intelligence. One view is that artificial intelligence will remain in the weak artificial intelligence stage for a long time that can be predicted. Therefore, artificial intelligence still cannot escape the influence of human will and will always become an expert system with certain judgment ability in a certain field to satisfy human self-interest. Its essence is a tool used by humans to meet their own needs and interests. Strong artificial intelligence is an illusory thing that only exists in imagination. Giving it the main position of criminal law will make legislation appear illusory. In this model, artificial intelligence has no human attributes at all, let alone the subjective position in criminal law. Therefore, the crime of criminal acts caused by artificial intelligence should be borne by specific actors using artificial intelligence. Another view is completely different. This group of scholars believe that artificial intelligence will inevitably enter the stage of strong artificial intelligence. At this time, artificial intelligence has got rid of its dependence on human beings and has its own independent consciousness. It will do or not do some behavior under the control of its independent consciousness and will. At this time, artificial intelligence is almost exactly the same as the attributes of the subject defined in the current criminal law. Therefore, artificial intelligence should be given the dominant position in criminal law, and the consequences of the criminal acts carried out by artificial intelligence should be borne by itself. It can be seen from this that the current issue of criminal law in the criminal law community's current responsibility for artificial intelligence is mainly based on the legal perspective. The criminal responsibility of artificial intelligence is discussed within the existing legal framework, and the involvement of technology is only a brief experience, and it has not been able to combine the development of artificial intelligence technology with the development of criminal law well. This has created a wild appearance of discussion on the issue of criminal responsibility for artificial intelligence. This article intends to discuss the issue of criminal responsibility for artificial intelligence from the perspective of the development of artificial intelligence technology and combine it with the development of criminal law.
1. Study the importance of criminal responsibility for artificial intelligence
In the view of scholars who hold a negative view on the subject of artificial intelligence criminal law, it is still too early to discuss the subject of artificial intelligence in criminal law. At this stage, artificial intelligence cannot show the unique wisdom of human beings at all. Therefore, it is too hasty to put artificial intelligence on the agenda now, and research in some fields has even violated people's common sense and anti-intellectuality.
However, the unique feature of artificial intelligence technology is that it is not a new technology generated simply by the pileup or iteration of some technology. It is an all-encompassing discipline with particularly high degree of integration with various computer technologies developed before. If artificial intelligence technology is compared to the ocean, then this brand new ocean contains the essence of the development of the computer industry for decades. Including: deep learning, data mining, big data, cloud computing, complex neural networks, link analysis, distributed storage, Internet of Things, mathematical modeling, etc., it can be said to be all-encompassing.Therefore, the benefits and risks brought by artificial intelligence technology are incomparable to previous technological innovations. No one can predict when the singularity of artificial intelligence technology written by science fiction writer Vernor Steffen Vinge will come. Once artificial intelligence technology reaches the stage of strong artificial intelligence, artificial intelligence may commit some criminal acts because of its own joys, angers, sorrows, or in order to protect itself, others or even other artificial intelligence. At the same time, this artificial intelligence algorithm that implements criminal behavior may have spread throughout the world with the Internet. Relying on the support of P2P technology and distributed storage technology, its broadcasting algorithms on the global Internet are even more impossible to completely clear. If other artificial intelligences actively download or passively accept this algorithm in the future, it will inevitably lead to the recurrence of previous criminal behaviors. The reproducibility of this kind of criminal behavior, the stubbornness of repeated crimes, the extensiveness of the subjects of crime and the complexity of crime prevention are unforeseen by the criminal law community at this stage. The purpose of the state's formulation, demand, discretion, execution and supervision of punishments is to maximize the interests of everyone, prevent national crimes and enable criminal citizens to receive the punishment they deserve. In the above situation, there is a subject that cannot be regulated by the criminal law and is unscrupulously committed criminal acts, and this crime is very easy to occur on a large scale. This is tantamount to the modern human beings living in a primitive society where they ate raw meat and drank blood before the emergence of law and even morality. All rules will be broken, the already formed social structure will be subverted, and the legal paradigm we are familiar with will no longer exist. When we enter the era of strong artificial intelligence, we will inevitably lead to the "Collingridge's Dilemma". Therefore, it is very urgent to carry out legal research on the issue of criminal responsibility for artificial intelligence.
2. The similarity and differences between artificial intelligence technology and human self-will
Most of the understanding of artificial intelligence technology in the criminal law community is still stuck in the black box of artificial intelligence technology. By entering the data that people have mastered, you can obtain an optimal solution under the existing algorithm framework. The logic in this algorithm black box is obtained through countless trainings of artificial intelligence. Artificial intelligence passively accepts the established facts input by people through countless trainings. This allows this algorithm black box to naturally output relatively specific results when encountering similar situations in the future. In this logic, although the final output result of artificial intelligence is unpredictable, the output range is still within the range preset by the program and is carried out according to the will of the developer or user. Therefore, artificial intelligence is just a tool. It is still under the control of developers or users. The author believes that this view makes the source of artificial intelligence logic too simplified, has a lack of deep understanding of artificial intelligence technology, and completely denies the autonomy and independence of artificial intelligence, which is inevitably suspected of "anthrocentrism".
(I) Development of artificial intelligence
Modern artificial intelligence technology truly originated from the famous paper written by Marvin Minsky in 1960 - "Step to Artificial Intelligence". Later, Eliza, the first dialogue robot in human history, was born. The most basic logic was to scan a certain keyword and match appropriate vocabulary corresponding to the keywords in his own database, combine these vocabulary into a smooth sentence and then output the sentence. This program has no intelligence at all, and it cannot even reach the level of logical self-consistent. Then came the well-known robot Deep Blue, which defeated then-world chess champion Gary Kimovich Kasparov in 1997. Deep Blue is not a real artificial intelligence. Its core technical idea is exhaustive, that is, to calculate all the possibilities for the next move of all chess pieces and output the optimal solution. It can be said that Deep Blue's victory at this stage is completely a victory of computer computing power.The computing speed and ability of computers are much higher than that of the human brain. Through this exhaustive way, calculates all possibilities and ultimately defeats Kasparov. Artificial intelligence in this period has the prototype of current artificial intelligence technology, that is, choose the optimal solution. However, our current artificial intelligence has completely abandoned the previous simple scanning words and output them, or used seemingly endless computing power to "violently" solve all problems. Computer scientists organically unify thinking, perception and actions through the principles of bionics imitating human logic. Today's artificial intelligence is not only an expert system or a problem-solving tool, but also a kind of thinking and a human-like intelligence.
(II) Independence of artificial intelligence
1. The structure of the artificial intelligence system corresponds to the human intelligent system
How humans understand the world, ask questions and solve problems, and form their own values, outlook on life, and worldview is the key to human intelligence. The expression of human intelligence in criminal law is how humans rely on their own thinking and will to control their own actions or not to do certain specific actions. So how did human intelligence come about, or why human intelligence is so unique and has such high recognition. Dr. Avram Noam Chomsky once had a famous statement: "It is almost certain that 50,000 years ago, a small group of people gained the ability to merge the two concepts into the third concept without limiting the original concept." It can be seen that the human symbolic system, namely language and words, played an indispensable role in the development of human intelligence. But human intelligence is not just composed of symbolic systems. When the teacher assigns a math problem to the students, the teacher conveys specific information to the students through the symbolic system. The students capture this information through the visual and auditory systems, and through the symbolic system that has long been stored in their brains, they visualize the captured abstract information one by one in their brains into content that they can understand, and then solve this mathematical problem. Human beings obtain human intelligence by constantly learning various theoretical knowledge for a long time and constantly practicing these theoretical knowledge in social life, and constantly summarizing the experience gained in practice. In this process, human beings always use morality and law as the guidance and framework for their actions. Correspondingly, in the current artificial intelligence technology, artificial intelligence also has its own unique symbol system - binary. Through mathematical modeling and using binary language to store this mathematical model in the storage system of artificial intelligence, artificial intelligence can understand various rules and operating methods in human society. Of course, this process is also inseparable from a lot of training and machine learning. By encapsulating various data in human society into data sets, artificial intelligence is trained a lot, and the eigenvalues and constraints in the algorithm are adjusted so that the output results are the best possible optimal solution defined by humans. In this way, artificial intelligence is like a born child without training through a large amount of data. Its hardware facilities are complete, but it does not exist yet what should be called "intelligent" because it has not completed specific learning. In this artificial intelligence storage system, there is no paradigm of oneself and behavior, and it is impossible to rely on one's own understanding and will to control one's body to complete certain tasks and perform specific behaviors. After it has the support of algorithms and conducted a lot of learning and training, a certain logical structure will be generated in the storage system. Rather than a logical structure, it is better to say that it is a large amount of human social experience learned and marked the eigenvalues in it. These social experiences are summarized and classified through eigenvalues, and a specific knowledge system is built in its own storage system.Whenever the symbolic system of artificial intelligence receives the information flow input from the I/O device, it will activate the symbolic system to find the eigenvalue closest to the information flow, and express the result corresponding to this eigenvalue to the outside world by doing a relatively specific behavior. The current technological route of this kind of artificial intelligence is relatively mature and in the foreseeable future, artificial intelligence will continue to follow this path. In this way, the hardware system in which artificial intelligence processes information is functionally comparable to the organ in which humans processes information. The growth of human wisdom lies in learning knowledge extensively and forming an inherent knowledge system in one's own mind. The growth of the "intelligence" of artificial intelligence lies in the use of a large amount of data for training, forming an inherent thinking system in the storage system, and solving various problems. It is particularly important to mention that the constraints of the algorithm during the training process correspond exactly to the law and morality in human society. Law and morality guide various social activities of human beings, while constraints require what behaviors artificial intelligence should or should not do. Therefore, artificial intelligence is roughly equivalent to humans in terms of hardware functions, way of thinking, or its intelligence acquisition methods.
2. Artificial intelligence corresponds to the way humans make decisions
The famous American jurist Oliver Wendell Holmes (Jr.) said: "The life of law does not lie in logic, but in experience." If the judgment of the law is based on strict logical reasoning, overemphasizing the rigor and scientific nature of logic, and ignoring the role of changing social experience. It will seriously separate the law from the world of experience and gradually be closed, its vitality will wither and eventually become a shackle for social development. Holmes's words emphasize the important role of experience in the legal system, because people do or do not do certain behaviors in their daily social life. Most of the theoretical basis is their own life experience and habits. Based on the current situation or problems you are facing, summarize and interpret it in your brain, and naturally draw corresponding conclusions. This conclusion will never be a very rigorous and scientific logical reasoning. Its birth is more similar to the problem of probability, because it cannot be guaranteed that humans will make completely consistent decisions under the same circumstances. This is also a major feature of human intelligence, and the relative uncertainty of behaviors performed under one's own specific knowledge framework. At the beginning of the 21st century, computer scientists realized that the results output from the implementation methods of artificial intelligence based on pure logic theoretical deductions are often unsatisfactory. The reason for this great difference in conclusions is that it ignores the role of empiricism, and many of the derivation results of pure logic are out of reality. Therefore, the basis for the decision made by artificial intelligence today is the statistical probability in mathematical methods. Artificial intelligence has trained a lot of data to learn that humans have the most likely solution in this situation, and imitates this solution to make corresponding behaviors. This technological route completely mimics the way humans make decisions, allowing artificial intelligence to truly act in the way humans behave.
3. The criminal status of artificial intelligence
Artificial intelligence fully demonstrates its equivalent to human intelligence by imitating human symbolic systems, ways of thinking and logic of making decisions. All actions performed by artificial intelligence are separated from the will of its users, but are completely possessed of behaviors similar to human autonomy. From this point of view alone, artificial intelligence relies entirely on its own will to control its own behavior, and can become the subject of criminal law and bear criminal responsibility alone.
(III) Dependence of artificial intelligence
Patrick Winston once said that the key to artificial intelligence technology is not algorithms, but modeling. The key to artificial intelligence technology is to split the complex and concrete tasks in human life as necessary, and map each work into various models in computer systems based on abstract generalization.This modeling work is the most basic and core part of the entire artificial intelligence. Through mathematical modeling of artificial intelligence, we can understand the various rules of human society and various human behaviors. Therefore, what kind of model the developers have built for artificial intelligence directly determines how artificial intelligence understands the entire social rules, especially the law. It also directly determines the way artificial intelligence will act in. Therefore, artificial intelligence has been closely related to R&D since its birth.
Secondly, the origin of the thinking and will of artificial intelligence mentioned above cannot be separated from the training of a large amount of data. And the source of these data and the identification of the quality of the information contained in these data. There is also the division of training sets and test sets for these data. There is also the most important data set annotation work. The so-called annotation is to interpret these data according to human definitions and input the data set with explanations into an artificial intelligence system for training. Just like when a child is in kindergarten, the teacher teaches what an apple is and what a banana is. In this way, the concepts of apples and bananas are found in the children's brain, and naturally they know that apples and bananas are not of the same type of substances. In this way, artificial intelligence can understand the meaning of these data in human society. The above work does not leave the developer's own expression of intention.
Finally, whether artificial intelligence has achieved the goals we need to rely on cross-verification. A simple summary is to separate the test set and the training set from the data set that trains artificial intelligence. By cross-comparing these two sets, we can see how accurate AI is. The division of test sets is equivalent to the developer telling artificial intelligence what behaviors are right and appropriate, and what behaviors are wrong and should be avoided.
Human independent thinking and will are formed through long-term social life, including human ethics, morals, politics, law, religion and family education, etc. These events experienced during growth and the knowledge learned are not transferred by anyone's will. However, no matter which stage of artificial intelligence development is inseparable from the will of the developer to guide artificial intelligence. From the construction of mathematical models, the division of data sets to the judgment of whether the behavior is correct or not, all include the subjective will of the developer. The developer converts his inner will into the behavior of developing artificial intelligence. In this research and development process, especially when judging the effectiveness of artificial intelligence, he must rely on the developer's own values to make judgments. Although after the development of artificial intelligence is completed, it can be regarded as a subject that is completely separated from the developer and exists independently in criminal law, the growth of artificial intelligence thinking is very different from that of humans. Artificial intelligence has not experienced a complete social life, and its understanding of human society is only obtained through short-term machine learning. Moreover, the data that artificial intelligence can access and the meaning of these data in human society are all processed by the research and development. The most important code of behavior of artificial intelligence - constraints - are also given by developers. Therefore, the construction of its inner values and behavior can be traced back to the behavior of the developer. If during research and development, the developer intentionally or negligently adds intentional homicide as the last means for artificial intelligence to resolve disputes into the dataset, and artificial intelligence commits murder, at this time, although artificial intelligence is an independent individual, it is difficult to deny the main role played by the developer in it. Although artificial intelligence is not the subject of criminal law during training, there is no co-conspiracy between the two subjects. However, at this time, artificial intelligence truly implemented the research and developer's thinking and externalized this thinking into action. From this perspective, we regard artificial intelligence as an individual in the criminal law, and the individual conviction and punishment have not achieved the punishment and prevention effect that the criminal law should have.
3. From the development history of criminal law, the status of artificial intelligence
The original criminal law can be traced back to ancient Babylon in 1785 BC. King Hammurabi promulgated the Hammurabi Code, and later ancient Greece and the Bible also had provisions on crime and its punishment. The reform of Western criminal law in modern times was carried out during the Renaissance under the guidance of Enlightenment thought. Professor He Qinhua believes: "Under the guidance of Enlightenmentist thought, the criminal law in modern times has become a guarantee of civil liberties and rights in civil society through the renewal of concepts and meticulous institutional design. This is a huge historic change in human society." It can be seen that criminal law is the law of civil society and the law that protects civil rights and freedoms, no matter what new things and new social relations appear in society in the future. The essence and purpose of criminal law will not change. Analyzing these affairs and social relations must start from the origin, essence and purpose of criminal law, and artificial intelligence is no exception.
(I) The essence of criminal law
"... From this we can see that there is a fundamental rationality. Law is the relationship between this fundamental rationality and various existence objects, and it is also the relationship between existence objects and each other." Montesquieu believed that there is a phenomenon that runs through the entire social phenomenon and is the essence of social phenomena. Criminal law is a kind of essence of social phenomena, Montesquieu called "legal spirit." The legal spirit is restricted by multiple factors, and politics, economy, culture, customs, morality, natural climate, etc. may shape different legal spirits. The spirit of law runs through the entire criminal law. To have an accurate grasp of criminal law, we must first understand the customs and customs of a society and see the essence through phenomena. At the beginning of its research and development, developers had inputted the things contained in these legal spirits into the storage system of artificial intelligence in the form of data sets and conducted repeated trainings. Moreover, artificial intelligence storage systems are much stronger than human memory. As long as the hardware is not damaged, this knowledge can last forever. Therefore, artificial intelligence has a very solid grasp of the spirit of law, and the result is a very profound understanding of criminal law. The understanding of the purpose that the criminal law wants to achieve and the constraints of the criminal law on one's own behavior should be no different from that of ordinary people. French scholar Raymond Aron has a deeper understanding of the spirit of law. "Once people have mastered the profound reasons that determine the overall process of events, the future becomes understandable. Once diversity is placed in the few types or concepts, diversity becomes known." This shows that once a normal human being understands the spirit of law, the person should have a full understanding of the consequences of his current behavior in criminal law. Artificial intelligence can easily understand the spirit of the law, and artificial intelligence also has complete logical thinking ability to foresee the consequences of its actions in criminal law. For this alone, artificial intelligence can independently bear criminal responsibility.
(II) Causes of crime
Professor Chen Xingliang analyzed the most important value of people proposed by Kant - freedom: "Human nature is a kind of 'non-social sociality', that is, on the one hand, people are social and hope to live in society to facilitate the development of their natural endowments; but on the other hand, people are non-social because they have a strong emotional tendency and must live as individuals so that they can act according to their own wishes." This is the most basic human nature and the freedom that people desire the most. The purpose of the birth of the Criminal Law is to safeguard individual freedom. It should be noted that criminal law is not the result of the active pursuit of individuals in society, but the product of everyone's continuous compromise in the development of society. Human "sociality" is to seek better development and progress. Human beings are social animals from their origins. A person's power is insignificant when facing the unknown nature at that time. Only living in groups can protect their own life safety.Today, humans, in the large collective of society, can only survive in the collective of society by recognizing their position in the division of labor in the social industry and conscientiously completing the assigned tasks. Human "non-sociality" will constantly tear the relationship between individuals and society, aggravate social class conflicts, and make individuals spare no effort to occupy the collective means of production. This will seriously damage the interests of others and hinder the development and progress of others. In severe cases, it will lead to crime. In order to maximize their own interests, protect their vested interests, reasonably distribute the benefits they can obtain in the future, and constantly adjust and compromise in social life, thus resulting in criminal law. Cesare Lombroso even proposed the concept of "naturally criminal", emphasizing that about 40% of the people commit crimes in society are crimes caused by their natural physiological structure. Robbroso believes that crime is caused by hereditary genes, so some people are born criminals. But not all criminals are born, and some criminals have experienced some stimulation and harm in social life. Jeremy Bentham's utilitarian law believes that humans are creatures that seek profit and avoid harm. Criminal behavior is inevitable when a person receives greater happiness from committing a crime than he receives for his crime. For artificial intelligence, if its criminal behavior is caused by hardware defects, the developer of artificial intelligence will bear unshirkable responsibilities. When the artificial intelligence is delivered to users, it is regarded as a commitment to the normal operation of artificial intelligence, completing the tasks assigned by users, and ensuring the safety of life and property of users and unspecified third parties. Once a criminal act is caused by hardware defects, according to Lombroso's theory, the will and reason of people in the process of crime will become no longer reliable. At this time, artificial intelligence is even less likely to be a criminal act under the control of its own will. We can completely pursue criminal responsibility based on the attribution principle of product defects. However, it is difficult to determine whether the crime motive of artificial intelligence is to pursue the maximization of one's own interests or simply to show off one's temporary pleasure. Huizi once said to Zhuangzi: "If you are not a fish, how can you know the joy of a fish?" This shows that it is difficult to analyze and judge whether the criminal acts carried out by artificial intelligence are for their own pleasure or for their own interests, or for the interests of users and developers. The personal pleasure and interests in society we are talking about are very abstract concepts that cannot be expressed quantitatively through the form of numbers. Therefore, we can only deal with it by analyzing who is the ultimate beneficiary. If artificial intelligence implements a criminal act that is purely beneficial to itself, since the developer and user have no benefit in this criminal activity, we can think that the developer and user do not actively pursue the subjective will to occur in the crime, so we can only convict and punish artificial intelligence alone. If there is a problem of letting the crime result occur, if an objective and complete chain of evidence cannot be proposed, then it should be determined that the developer and user have nothing to do with the criminal act in accordance with the principle of prosecution. For situations where developers and users benefit from criminal acts of artificial intelligence, we can learn from the principle of presumption of fault in the application of civil law. Artificial intelligence developers and users should have full foresight about the subjective will of artificial intelligence and the risks in the use process. Criminal law should also severely crack down on the behavior of R&D and users using artificial intelligence to engage in criminal activities and obtain benefits from it. Once they gain benefits from a criminal act in artificial intelligence, we can determine that they have directly or indirectly participated in the criminal activity. The burden of proof is inverted and the developer and user prove that they have not participated in the criminal activity. Because higher requirements are put forward for developers and users at this time, public security organs or institutions qualified to appraise artificial intelligence-related products should provide full assistance if necessary.
(III) The current criminal law regulates artificial intelligence
Our current criminal law has formed a fairly complete system. However, the law must have lag, and the existing criminal law still has many difficult to effectively regulate emerging things like artificial intelligence.
1. No conviction or punishment for crimes is allowed
"No behavior will be committed" This is a widely circulated legal motto, which shows the decisive significance of the act in the establishment of a crime. However, artificial intelligence crimes show different characteristics. The artificial intelligence we often discuss is a specific artificial intelligence body, but the artificial intelligence body is just the carrier object of artificial intelligence algorithms. The specific behaviors made by the artificial intelligence body are controlled by the artificial intelligence algorithm, and this algorithm is the core of the entire artificial intelligence. It can be said that once a criminal act occurs, the criminal act is caused by the algorithm rather than the carrier of the algorithm. Assuming that this algorithm includes all current criminal means and methods, as far as the current criminal law system is concerned, it is only at the stage of criminal intent. The current provisions of the Criminal Law are completely reasonable and effective for human crimes, because the intelligence of a particular individual cannot exist alone from its physical body. If the body is restricted or eliminated, the will to sin will be eliminated or restricted. However, for artificial intelligence algorithms, they can be combined with any carrier, so that any carrier can commit crimes without the algorithm itself being punished by criminal law. And once this algorithm is connected to the Internet, the consequences are catastrophic. The carrier of any algorithm may become a tool for this algorithm to commit criminal behavior. Through access to the Internet, this algorithm can spread very quickly and its spread range will be very wide. Because this algorithm will not be affected by any other factors during the propagation process, the criminal will represented by this algorithm is transmitted to the next carrier in good condition, and it is inevitable that the next carrier will commit the crime again. In the current criminal law system, the criminal intention of algorithm cannot be the object of conviction and punishment, which leads to the phenomenon of treating the symptoms but not the root cause. If any artificial intelligence violates the criminal law, we can deprive it of its freedom or even its life. The life of artificial intelligence can be the integrity of its hardware or the integrity of its algorithms. But as long as the algorithm is still there, the thinking and will of this artificial intelligence will not be erased, and the possibility of it repeating is almost 100%. The punishment for algorithm carriers will not have any effect in combating crimes. Because its criminal will exists without relying on the carrier. Criminal law certainly has its modestness, but in this case, excessive pursuit of the modestness of criminal law will cause endless consequences.
2. The ambiguity of criminal responsibility ability
Putting aside the debate on which is better and worse, the "four requirements" and the "three-level" theory, whether it is the identification of subject qualifications in the "four requirements" theory or the responsible judgment of criminal responsibility ability occupies an important position. We can easily judge the incomplete state of criminal liability brought by age, but the judgment of the criminal liability of mental patients is an extremely complex issue. "While claiming that the determination of the responsibility ability of a mental patient is a legal judgment, it cannot provide a set of operational identification plans. This may lead to the judge's inability to determine the responsibility ability of a mental patient, and in the end it may be reduced to intuitive judgments or blindly following the identification opinions of judicial psychiatrists." It can be clearly seen that although criminal liability ability occupies such an important position in the composition of crime, there is still no strict unified standard for its judgment in the academic community. The definition of mental illness in criminal law is: mental abnormality caused by mental disorders. The mental illness and age of criminal responsibility defined under the current criminal law framework are completely inoperable for the judgment of artificial intelligence's criminal responsibility ability. The newly revised Criminal Law Amendment (XI) of my country re-stated the standards for age of criminal responsibility."People who are over 12 years old and under 14 years old are guilty of intentional homicide, intentional injury, causing death, or causing serious injury by particularly cruel means, causing serious disability, and the circumstances are serious and those who are approved and prosecuted by the Supreme People's Procuratorate shall bear criminal responsibility." This shows that the age of criminal responsibility is not fixed, but changes with the development of society and the cognitive level of the majority of young people. The age of criminal responsibility in our country has been obtained through long-term practice and discussion. The promulgation of the Criminal Law Amendment (XI) also means that my country's age of criminal responsibility has moved from a static "one-size-fits-all" approach to a dynamic judgment, and in very special circumstances it is submitted to the Supreme People's Procuratorate for approval in a more humane direction. However, it is difficult to grasp the age of criminal responsibility of artificial intelligence with accurate numbers. We usually understand that after reaching the age of criminal responsibility, humans can clearly recognize the possible harmful consequences of their actions, and therefore can bear criminal liability. However, artificial intelligence is the "intelligence" obtained after a large amount of data training, and the content of these data directly affects the judgment ability of artificial intelligence. If all these data are data generated by human children under the age of 12, then should it be considered that the artificial intelligence trained after these data has not reached the age of criminal responsibility. In more complex situations, some of these data are produced by humans who have reached the age of criminal responsibility, and some are produced by humans who have not reached the age of criminal responsibility. How should we judge the age of criminal responsibility of artificial intelligence at this time? Similarly, for artificial intelligence that has completed training, how to determine whether its intelligence is defective and whether it belongs to the mental illness state defined by criminal law? Judging from the existing technical means, it is impossible to make judgments at all from the medical level or the legal level. When we regard artificial intelligence as the subject of criminal law and convict and punish it, this problem will be highlighted, leading many scholars to oppose the subject of criminal law of artificial intelligence. Because this will completely overturn our existing criminal law system and break many of our principles of conviction and punishment of suspects. However, the Criminal Law exists to protect the interests of the vast majority of people. When encountering situations that may seriously infringe on the interests of the majority, the Criminal Law has an unshirkable regulatory obligation.
4. Regulation of artificial intelligence technology
"In order to ensure social security, preventive criminalization legislation is based on the risk of serious criminal crimes, adheres to the concept that pre-prevention is better than post-disposal, and stipulates acts that generate serious criminal crimes as independent crimes. In this way, a buffer prevention belt is set up before crimes with serious social harm. This is conducive to preventing criminal acts with serious social harm in the context of risk society, and is conducive to realizing the function of criminal law to protect society, reflecting the active response of criminal law legislation to the increasingly accumulating criminal risk challenges in the context of risk society." For artificial intelligence, a technology that may be extremely harmful, preventive criminal legislation is the best regulatory method. Therefore, the general prevention purpose of criminal law should be considered and the challenges brought by artificial intelligence technology should be viewed dialectically. Artificial intelligence has both its human and instrumental side. It is inappropriate to simply classify artificial intelligence technology as a tool used by individuals in criminal law or humans.
First of all, for artificial intelligence technology, the behavior of developers should be regulated through laws from the beginning of its research and development. Corresponding laws should be formulated for the selection of data sets and the annotation of data, and the coercive power of the law should be used to prevent R&D from developing special criminal artificial intelligence through data research and development. The data used during its R&D process should be handed over to the relevant parts for verification and backup. Once data that violates the law in the data set appears, the R&D behavior should be stopped quickly.
Secondly, the research and development of artificial intelligence should be carried out in a physically isolated laboratory. For the necessary behavior of using the Internet during the research and development process, the information required on the Internet can be downloaded through local mirroring. The behavior of artificial intelligence in research and development should be directly connected to the Internet, and the behavior of defective artificial intelligence algorithms being widely disseminated through the Internet can be strictly prohibited.
Finally, an artificial intelligence academic ethics committee should be established, and the artificial intelligence completed by the research and development can only be put into use after passing the academic evaluation. A one-vote veto system is adopted for intentional research and development of artificial intelligence that may undermine existing laws, and developers are strictly prohibited from engaging in artificial intelligence research and development activities again.
For special prevention, we can learn from the concept of human nature in commercial law. It is precisely because there is a unique relationship between R&D, user and artificial intelligence itself that they are tied up and treat the R&D, user and artificial intelligence of artificial intelligence as a community of shared future. A double penalty system is required, which must be convicted and punished by artificial intelligence itself, as well as the R&D and users. Only in this way can we better prevent developers and users from abusing the independent status of artificial intelligence and conducting illegal and criminal activities that make profits for themselves. If artificial intelligence commits criminal acts, it should be the subject of the criminal law and punish the crimes it commits accordingly in accordance with the provisions of the criminal law. Completely delete the data it stores and all information related to it, especially the data it has remained on the Internet. For developers and users, their subjective intentions should be fully considered. If artificial intelligence is deliberately committed, it is believed that it uses artificial intelligence as a tool and can be convicted and punished according to intentional crime. Negligent crimes can also be convicted and punished for negligent crimes.
Artificial intelligence is a product of social relations in the historical development of human beings and must be regulated by criminal law. When it comes to the issue of criminal law regulation of artificial intelligence, it should be grasped. Criminal law is the law of human society and is formulated for the common interests of all mankind. Artificial intelligence technology should also be a technology developed and applied for the common welfare of all mankind. Therefore, our regulation of artificial intelligence technology should focus on the interests of human society, and transform the "double-edged sword" of artificial intelligence into the "sword of victory" of technological revolution in the new era without damaging the development of artificial intelligence technology.
Source: Shanghai Law Research Journal Volume 1, 2022 (Collected Works of the Intellectual Rule of Law Academic Community)
