Thoughts on several current legal issues involving artificial intelligence

Author: Song Jianbao, Director of the Internet Judicial Research Center of China Institute of Applied Law Source: People's Court News

In today's era of rapid iteration and widespread application of digital technology, artificial intelligence, as a master of digital technology applications, has posed many challenges to the existing legal system. Judicial trials not only face technological developments that are difficult to understand, but also need to effectively match the legal system with a certain degree of lag with the evolving social reality. To this end, the author will briefly introduce the main types of artificial intelligence and its basic principles, and talk about his own views on several current legal issues involving artificial intelligence that are closely related to judicial trials, for reference in future case handling and research.

Main types of artificial intelligence and their basic principles

Due to different theories or methods, the semiotic school and the bionic school were formed in the development process of artificial intelligence, and correspondingly, symbolic artificial intelligence and neural network artificial intelligence were formed. Of course, a specific artificial intelligence may combine two artificial intelligence technologies, symbolic artificial intelligence and neural network artificial intelligence. In judicial trials, it is necessary to clarify the specific type of artificial intelligence involved in the case and its basic principles in order to understand the intelligence and corresponding operations of the artificial intelligence involved in the case.

1. Symbolic artificial intelligence

Symbolic artificial intelligence The theoretical basis of intelligence is theorem proof, and its implementation method is to simulate human intelligence based on logical reasoning. The semiotic school believes that the basic unit of human cognition and thinking is symbols, and intelligence is the representation and operation of symbols. Human beings give computers some of the most basic logic and rules, and computers can obtain intelligence through a series of calculations and reasoning. Therefore, human intelligence is transformed into various symbols, knowledge, rules and algorithms, and then computer technology is used to represent and operate these symbols, knowledge, rules and algorithms, etc., and finally the human intelligence simulation of the computer system can be realized.

The implementation program of symbolic artificial intelligence relies on logical decision trees. A logical decision tree is a set of rules on how to process a given input. Programming based on data sets and rule sets, through a series of "yes" or "no" judgments, symbolic artificial intelligence finally outputs results in the form of reasoning. The decision-making process of the logical decision tree is deterministic, so in theory, every decision can be traced back to the decisions made in advance by the artificial intelligence R&D designers. At present, typical representatives of symbolic artificial intelligence include expert systems, knowledge graphs, knowledge engineering, and databases. Specific application areas include computational advertising in the Internet advertising industry, click-through rate estimation on search platforms, and risk control in the financial industry.

2. Neural network artificial intelligence

Neural network It is composed of a large number of artificial neurons, each of which imitates the neurons of the human brain to some extent. These neurons are arranged in layers and are connected to each other or at least to the neurons in the layer below them. Each neuron is capable of processing a given input and generating an output, and can transmit the output to the neuron of the next layer. The output of the neuron of the last layer is the system output and is not transmitted to other layers. Each neuron processes a given input based on internal instructions and the neuron's own weights and biases. Therefore, the input data at the top layer is processed and transferred to the next layer for processing according to the state of the neuron. Each neuron generates an output according to its weight and bias and transfers the output to the next layer until the processing result is at the bottom layer. output.

Neural networks learn how to process given inputs through training, which is designed to adjust the internal parameters of the neural network. The learning process will use a training data set, a validation data set, and iteratively feed the neural network the training data set to determine how to adjust the network parameters to reduce the error in its expected output. As part of training, the weights and biases of each neuron are also adjusted. Once the training process is complete, the AI ​​can be used to process real data, with its structure or topology frozen and no further adjustments or programming activities performed on the neural network or data. The data transmitted through the neural network is only subject to processing by each node. The state of a node, that is, how it processes input, generates output, and transmits output, is determined by the neural network itself and is learned through the aforementioned training, and is not determined in advance by the node programmer. The large language model is a typical neural network artificial intelligence.

Does artificial intelligence have legal subject qualifications

Legal subject Determined by current law and vary depending on the legal relationship. Whether artificial intelligence has legal subject qualifications should be examined and demonstrated from the perspective of legal relationships and the causes of legal subject types. The following takes civil relations as an example to examine the causes of the types of civil subjects and demonstrate whether artificial intelligence has legal subject qualifications.

1. How does a natural person obtain legal subject qualifications

Under patriarchal system, parents have the capacity to have rights, while family members and slaves have very little or no capacity to have rights. Under the manor system, farmers were appendages to the land and were subject to the rights of the lord. With the development of the exchange economy, family members, slaves, farmers, etc. gradually acquired rights and behavioral capabilities within the scope of the sales contract. With the development of the urban system and the boom of urban industry and commerce in the Middle Ages, urban practitioners became the subject of labor contracts through the "labor-wage" transaction, and also became the subject of consumer contracts through the purchase of daily necessities. It can be seen that changes in economic and social relations are the biggest motivation for natural persons to gradually obtain legal subject qualifications.

Under the principle of modern rule of law, all natural persons have equal civil subject qualifications, which affirms human value, dignity and subjectivity, and is the most ethical a basic legal principle. The subject qualifications granted to natural persons by the law consist of various abilities such as rights, conduct, responsibility, as well as party abilities and litigation abilities in procedural law. Although they have different social functions, they are all aimed at safeguarding the personal dignity and embodiment of natural persons. Human ethics.

2. Reasons for granting legal subject status to legal persons

The early legal system was centered on the state and individuals, and general groups did not have subject qualifications. With the development of society, the state is no longer competent to handle all production and life relationships, and individuals often feel weak in pursuing life interests and personal development. Since the Industrial Revolution, especially with the development of the capitalist economy from the 19th to the 20th centuries, capital groups of various sizes have continued to appear. Some business enterprises cannot be run by one person. Large-scale business enterprises need to be sustained for a long time. After all, the life of natural persons is limited. Partnerships and legal persons have become the main organizational forms of business enterprises. A partnership is established by a natural person through a contract, but it is not separated from the natural person's personality and property. A legal person can independently enjoy rights and assume obligations. Its property is not the property of its members, and its debts are not the debts of its members.

Why the law confers civil subject qualifications to legal persons so that they can independently enjoy rights and assume obligations has been an issue that legal scholars have debated endlessly since the 18th century, and has formed the basis for There are various theories such as the system theory, the purpose property theory, and the reality theory. But in essence, because legal persons can play social roles and have social value, it is indeed necessary to give them legal personality. Therefore, the law analogizes natural persons and gives them civil subject qualifications in order to meet the needs of human social life.

3. Does artificial intelligence have reasons to confer subject qualifications?

The current artificial intelligence is developed, designed and operated by humans and is not completely out of human control. It belongs to weak artificial intelligence. For weak artificial intelligence, the various operations of artificial intelligence are only an operation under the control of existing legal subjects, and are an expansion and extension of the existing legal subjects' own capabilities. In terms of legal technology, treating various operations of artificial intelligence as the behavior or joint behavior of existing legal subjects is sufficient to solve the subject qualification issues caused by artificial intelligence. Strong artificial intelligence and super artificial intelligence have cognitive and thinking abilities that are equivalent to or even superior to humans. Therefore, some people argue that these artificial intelligences can independently enjoy rights and assume obligations within a certain range.

Judging from the history of the above-mentioned natural persons obtaining subject qualifications, their acquisition of subject qualifications is not based on their intellectual development. It can be seen that the advanced level of artificial intelligence depends on whether it is granted There is no connection between their subject qualifications. Judging from the reasons for legal persons to obtain subject qualifications, the current social role and social value of artificial intelligence have not yet reached the necessary level to confer it with subject qualifications, and complex legal technologies and organizational forms are required to achieve this. Therefore, there is currently no sufficient reason to confer subject qualifications on artificial intelligence.

Legal liability issues involving artificial intelligence

As before As mentioned above, artificial intelligence does not have legal subject qualifications. Under the current legal system, artificial intelligence is just a product or a service. Therefore, the legal liability of artificial intelligence must be discussed under this premise.

1. Criminal liability involving artificial intelligence

Artificial intelligence Intelligence does not have subject qualifications, so artificial intelligence cannot be identified as a criminal subject and should not bear any criminal responsibility. No matter how advanced the artificial intelligence is, the harmful acts carried out by artificial intelligence should be regarded as harmful acts carried out by the natural persons or units that control or direct the artificial intelligence. The criminal subjects should be natural persons or units, and artificial intelligence is only the harmful acts committed by these criminal subjects. behavioral tools. This is not substantially different from instigating natural persons without the capacity for criminal responsibility to commit criminal acts or using animals to commit criminal acts.

The intelligence of artificial intelligence is often an important factor leading to the occurrence of harmful behaviors. In individual cases, the intelligence of artificial intelligence is an indispensable factor leading to the occurrence of harmful behaviors. . Therefore,when examining the subjective mentality of the perpetrator, we should focus on examining the source of the intelligence required by artificial intelligence to implement harmful behaviors. For example, whether the intelligence of artificial intelligence that implements harmful behaviors comes from the developer and designer of artificial intelligence, from the artificial intelligence service provider, or from the use of the user. In specific cases, whether the harmful behavior committed by artificial intelligence constitutes a crime and how to allocate criminal liability should be based on factors such as the criteria for the incrimination of the specific crime and the circumstances of the crime, the subjective state of mind of the perpetrator, and the criminal justice policy involving artificial intelligence. Specific judgment.

2. Liability for breach of contract and tort liability involving artificial intelligence

Smart contracts exist in large numbers in practice. The conclusion and performance of contracts may involve manual intervention such as review and intervention by the parties to the contract, or they may all be completed independently by artificial intelligence. Artificial intelligence does not have civil subject qualifications and cannot become a party to a contract. In the case of the intervention of the parties, the conclusion and performance of the contract by artificial intelligence can be directly recognized as the actions of the parties, and the resulting breach of contract and liability for breach of contract should be regarded as the breach of contract and liability for breach of contract by the parties. If the conclusion and performance of the contract are all completed independently by artificial intelligence, the artificial intelligence can be regarded as the agent of the party, and the relevant rules of the agency system can apply to the resulting breach of contract and liability for breach of contract.

Artificial intelligence is just a product or service. When it causes harm to others, the identification of infringement, the identification of the responsible party, and the allocation of civil liability should be Laws such as the Product Quality Act apply. The core of product liability is product defects rather than the fault of the perpetrator. However, the law stipulates various exemptions for defects that do not bear liability for compensation, including defects that did not exist when the product was put into circulation, and the level of science and technology when the product was put into circulation. Defects that cannot yet be discovered, etc. These exemptable product defects will be a common defense for those responsible for artificial intelligence infringement. The review and determination of product defects, the leniency and severity of the applicable defect exemption rules, etc. will greatly affect the determination and allocation of liability for infringement involving artificial intelligence. Therefore, considering artificial intelligence as a product, the existing legal system can solve the problem of tort liability involving artificial intelligence to a certain extent, but judicial practice also faces considerable challenges.

Whether the generated product constitutes the object of intellectual property rights

Artificial Intelligence does not have subject qualifications, so it is not a relevant subject under intellectual property law. For example, it is not an author or copyright holder under copyright law, nor is it an inventor or patentee under patent law. However, whether artificial intelligence products can constitute the object of intellectual property rights is currently a controversial issue, such as whether the generated products constitute works under copyright law or inventions under patent law. The author believes thatWhether an artificial intelligence product constitutes a copyrighted work or an invention under patent law ultimately depends on whether the artificial intelligence user has directly invested intellectual labor and intellectual contribution in the production of the product.

If the generated product is completely completed by artificial intelligence, or the user does not directly invest in any intellectual activities, the user’s intellectual contribution is zero , the generated product does not constitute a copyrighted work, nor does it constitute an invention or creation under the patent law, and the user of the generated product should not enjoy rights under the copyright law or patent law. If the generated product is completely completed by the user independently, and artificial intelligence only replaces the user or assists the user in completing some non-intellectual work, the generated product can constitute a copyrightable work or an invention and creation under the patent law, and the user can enjoy copyright on the generated product. legal rights or rights under patent law. The generated products in practice are often produced by artificial intelligence users who put in certain intellectual labor and make use of the intelligence of artificial intelligence. In short, artificial intelligence uses humans to have a certain degree of intellectual contribution to the generated objects. As for the extent of the intellectual contribution of the user of artificial intelligence, the generated product can constitute the object of intellectual property rights. This is a matter of judicial discretion in intellectual property rights.

The author believes that the intellectual contribution of the person using artificial intelligence must at least reach the intellectual contribution of the co-author of the work or the co-completer of the invention, otherwise it will be difficult to consider the use of artificially generated objects The production of intellectual property provides intellectual labor, and its enjoyment of relevant intellectual property rights is hardly legitimate. Whether the generated product constitutes a work under copyright law, the focus should be on whether the user directly invested in intellectual labor in the creation of the generated product, and whether the intellectual contribution made to the originality of the generated product reaches the level of intellectual contribution due to the co-author of the work. . If the user's labor input is similar to "organizing work for others' creations, providing consulting opinions, material conditions, or performing other auxiliary work," etc., the labor input should not be regarded as creation, and the resulting product does not constitute copyright. Legal works. Whether the resultant product constitutes an invention-creation under the patent law shall focus on examining whether the user has directly invested intellectual labor in the resultant product, and whether the creative contribution made to the substantive characteristics of the resultant product has reached the level of intellectual contribution due to the co-contributors of the invention-creation. If during the production process of the product, the user is similar to "a person who is only responsible for organizational work, a person who facilitates the utilization of material and technical conditions, or a person who engages in other auxiliary work", etc., he cannot be regarded as the inventor, and the product cannot be regarded as the inventor. constitute an invention-creation under patent law.

Copyright infringement issues in artificial intelligence training and generated products

After training with massive data, if artificial intelligence can output the expected products, artificial intelligence service providers can provide commercial services to users. This gives rise to two issues in copyright law. One is whether the use of other people's works without permission during the artificial intelligence training stage constitutes copyright infringement. The second is whether the artificial intelligence generated products will infringe the copyright of others.

1. Does using other people’s works without permission during the artificial intelligence training stage constitute copyright infringement?

After training with massive texts, pictures, audios, videos and other works, artificial intelligence can generate corresponding texts, pictures, audios, videos, etc. If other people’s works are used without permission during the artificial intelligence training stage, it should be a typical situation of using other people’s works without the permission of the copyright owner. According to the general principles for determining copyright infringement, such artificial intelligence data training activities constitute copyright infringement and constitute direct infringement. It should be pointed out in particular that although artificial intelligence data training is part of commercial activities, artificial intelligence has not yet been put into commercial operation. Therefore, there is still room for discussion in intellectual property judicial policies as to whether the unauthorized use of other people's works during the artificial intelligence training stage constitutes copyright infringement, especially whether it constitutes fair use under copyright law.

2. Whether the generated product will constitute copyright infringement

As mentioned above, when artificial intelligence completes training and can output the expected products, it can be put into commercial operation. Its structure or topology is frozen, the neural network parameters are no longer adjusted, and the neural network programming activities are no longer performed. In other words, the generated products are produced by AI R&D designers and provided to users by AI service providers.

If the generated product is identical or substantially similar in expression to the works of others, it constitutes substantial similarity for the purpose of determining copyright infringement and may constitute copyright infringement. infringement. According to the general rules for determining copyright infringement, even if the generated product is substantially similar to another person's work, there must be evidence to prove that the generated product is a copy of or derived from another person's work. If there is evidence that the generated product is indeed copied or derived from the work of others, it can be determined that the generated product constitutes copyright infringement. At this time, the R&D designer should be regarded as the producer of the accused infringing product, the service provider should be regarded as the provider of the accused infringing product, and the user should be the user of the accused infringing product, and the accused infringing product is artificial intelligence Made according to user's needs and instructions. Therefore, when the generated product constitutes copyright infringement, the R&D designers and service providers should be regarded as infringers. There is also room for discussion in judicial policy on whether the user's use behavior constitutes infringement.