Explore the protectability of AI-generated content

　　Recently, the widespread popularity of generative artificial intelligence has sparked many discussions related to copyright law. Among the massive amounts of content generated by generative AI, there is no shortage of original expressions. For example, users can enter a specific number of keywords and ask generative AI to use specific lexical elements to concate to form a complete novel. The results show that generative artificial intelligence not only does not simply stitch specific elements, but also forms a "work" with a certain intention and thematic literary color through a certain logical level. The high ability of generative artificial intelligence in "creation" has attracted the attention of the industry. For the discussion of copyright law on AI-generated content, the primary premise is to clarify whether the generated content has the possibility of falling within the scope of copyright law protection, which also returns to the basic topic of copyright law research on AI-generated content in recent years.

　　Define the subject of creation

　　The Mainland Copyright Law stipulates that a work refers to intellectual achievements that are original and can be expressed in a certain form in the fields of literature, art and science. According to the provisions on the constituent elements of a work, a work protected by the Copyright Law cannot be separated from the subject of creation, and the main factor of creation is manifested in the contribution to the originality of the work and the investment in intellectual achievements. Proceeding from the textual interpretation of the Copyright Law, the factor of intellectual achievements implies the natural person attribute of the subject of the creation of the work, that is, it is inseparable from the natural person's thoughts, emotions and other intellectual activities to invest in the creation of the work. At the international level, the Berne Convention for the Protection of Literary and Artistic Works takes natural persons as the subject of the creation of works, thus forming the basis of the Convention's rule system. For generative AI systems, if there is a complete lack of creative input generated by human activities in the generated content, it is difficult to define it as the subject matter protected by the Copyright Law.

　　Whether there is no human intellectual activity input at all in the operation of generative artificial intelligence is a question that needs to be further explored. The operation analysis of generative artificial intelligence models is obviously inseparable from the intellectual activity investment of natural people. From the perspective of the generation process, the generated content includes the continuous reinforcement learning and drill of the initial technical developer on the program design of the model and the simulation of the artificial neural network, as well as the questioning, invocation and guided correction of the corpus by the user using the subject. Through the joint efforts of multiple natural person subjects, generative artificial intelligence can achieve clear logic and unique results, and as the number of interactions increases, generative artificial intelligence can also continuously update and improve the database. It can be seen that the "creation" process of generative AI is not entirely dominated by the model. In fact, the development of current artificial intelligence models has not reached the stage of strong artificial intelligence with autonomous consciousness. From this point of view, it is possible that the generated content will become a work protected by copyright law.

　　With original content

　　For the definition of the authorability of generative AI-generated content, the core requirement is still the originality of the generated content. As an important constituent element of a work, the definition and criteria of originality have always been controversial. The essence of originality is analyzed through the theory of legal interpretation, which contains two major elements: "uniqueness" and "creation". The former emphasizes that the generated content needs to be independently created by the creative entity, excludes simple copying, plagiarism and plagiarism of the works created by others, and requires that the generated content is objectively different from the existing results, and cannot constitute substantial similarity; For the latter, regardless of the presence or absence of inventive step or the high or low standard, this element requires that the work protected by copyright law has at least an objective level of creative expression that is different from the existing expression.

　　Combined with the aforementioned requirements for creative subjects, the judgment on the authorability of generative AI-generated content can be further limited to the judgment of whether human intellectual activity investment produces originality. From the perspective of the programming of generative artificial intelligence, the creative expression generated by generative artificial intelligence is inseparable from the programming and arrangement of language models. The reason why generative AI can generate a large number of expressions is inseparable from the massive data input in the early stage, and the larger the amount of input data, the stronger the "intelligence" of generative AI. It can be seen from this that the design and development of programs mainly assists the generation of creative expressions from the level of data mobilization and application, and the direct source of creativity in the generated content is mainly the massive language resources learned from the public domain or even the field of copyright protection.

　　Increase intellectual investment

　　From the perspective of the use of generative artificial intelligence, the key lies in whether the current use behavior of the user can directly generate creative intellectual input. To summarize the current application practice of generative AI, the user's creative expression of generative AI mainly includes the following situations: First, directly put forward some kind of "creation" requirements to generative AI, such as writing a poem and writing a program; The second is to provide some elements to generative artificial intelligence, requiring it to complete the creation according to the given elements; The third is to adopt progressive guiding questions and optimize the logic and expression of the reply step by step. As far as the first and second use cases mentioned above are concerned, although the initiator of creative expression is a natural person subject, in fact, it is difficult for users to have a substantial impact on the creative expression of the final generated content, and the vocabulary, sentence structure and language style used in generating content are the result of the data resource call learned by generative artificial intelligence from the previous stage.

　　In judicial practice, some courts have held that users of generative AI programs only enter and retrieve keywords, and do not produce original expressions that condense their thoughts and feelings, and should not be recognized as works. For the third use case mentioned above, due to the increase in user interaction investment, in the process of repeated optimization learning and output of generative AI based on questions, the degree of influence of using the subject's original input in the generated content is increasing, and the proportion is gradually increasing, and the user may have a direct and substantial impact on the creativity of the generated content.

　　The definition of the authorability of generative AI-generated content needs to be traced back to the theoretical foundation of the copyright system and strictly based on the essential requirements of the constituent elements of the work. Both the theory of labor property and the theory of incentive fully affirm the humanistic nature of the creative subject. At present, the "works" generated by generative artificial intelligence are still essentially inseparable from the control of human thinking and the endowment of thoughts and emotions, and the prerequisite for the definition of copyright is the investment of human intellectual achievements, while generative artificial intelligence itself does not have the qualifications of the creative subject. On the basis of human intellectual activity, it is necessary to judge the originality of the specific generated content. The generated content of generative artificial intelligence mainly comes from its machine learning of massive language data resources in the early stage, and specific creative expressions are not directly related to program developers, and the user's usage behavior is mainly limited to the retrieval and application of its language database, and in most cases there is no direct creative investment. Therefore, most of the generated content of generative AI is not the direct result of human intellectual activity. However, the copyrightability of generative AI-generated content cannot be completely ruled out, and with the continuous annotation of human creative activities, various types of works that meet the requirements of the Copyright Law may also be produced.

　　(Feng Xiaoqing, Zhang Ziyue, School of Civil and Commercial Economic Law, China University of Political Science and Law)