Intellectual Property Analysis of Digital Contents and Works in the Background of AIGC (II)

View: 42     Date: 2024-03-29 08:46

Author: Wenwen MA

In the last article, we discussed the infringement risks faced by AIGC, and today we will further discuss whether AI-generated content can be protected by copyright and ownership issues.

China's copyright law clearly states that a work should be an intellectual achievement that is original in the fields of literature, art and science and can be expressed in a certain form. As for the content generated by AI, it does conform to the categories of literature, art and science, and also has a certain form of expression. However, there is some debate about whether it has the attributes of originality and intellectual achievement. Originality requires the work to be independently done by the author, but the current definition of author in most countries' copyright laws does not include AI and is usually limited to natural or legal persons. This has led some to argue that AI-generated content is the product of an outcome based on algorithms, models and big data, rather than real intellectual work. But it's important to note that the AIGC process is not entirely algorithm-driven. When using AI tools, users also need to participate in the process, making certain choices and adjustments, which is also a process of intellectual application.

As can be seen from two cases in China's judicial practice, China has acknowledged the copyright of AI-generated content to a certain extent. The first case, which took place in 2019, involved Tencent's Dreamwriter intelligent writing assistance system. The court held that articles generated by the system meet the form requirements of written works in terms of performance, and their specific forms of expression were derived from the creator's personalized selection and arrangement, with certain originality, and their copyright should belong to the developer of the AI tool. The second case was recently heard by the Beijing Internet Court, about the copyright ownership of the image generated by Stable Diffusion. The court found that due to the user's intellectual input in the process of generating the image, the image had the elements of intellectual achievement, and there were identifiable differences between the image and the previous work: First, the prompt word for drawing elements and parameters for drawing layout settings reflected the user's choice and arrangement. Second, the final image was obtained through the adjustment and correction of prompt word and parameters, reflecting the user's aesthetic and personality judgment and expression, so it had originality, and the copyright should belong to the user of AI tool.

These two judgments show that China's judicial level has an open attitude on the issue of copyright of AI-generated content, but the specific ownership needs to be judged according to the actual situation. Considering the huge investment AI developers have made in product development, model building and training, it is reasonable for the copyright to belong to AI developers if users use AI-generated content in a simple and uncreative way. However, many current AI platforms, as commercial products, give up copyright in user agreements in order to attract users and require users to use AI-generated content again for model training. This practice has, to some extent, affected the recognition and attribution of copyright in AI-generated content. Therefore, in actual operation, we need to comprehensively consider the interests of all parties to ensure the reasonable distribution and protection of copyright.

If AI is only used as an auxiliary tool, and the content ultimately generated depends mainly on the user's originality and intellectual input, then copyright should indeed belong to the user. In practice, however, it is difficult to determine whether the instructions entered by the user are original. Typically, this requires an analysis of the complexity, novelty, and substantial impact of the instructions on the generated content. At the same time, the intelligence and originality ratio of users is also a difficult issue to quantify precisely, and often needs to be assessed by considering factors such as the user's overall creative process, the use of AI tools, and the uniqueness of the resulting content. In addition, even if the same instructions are used, the results generated each time may be different due to the randomness and unpredictability of the content generated by the AI. This leads to another question: Should all these different contents have copyright? This, too, is a topic worth exploring in depth.

Looking back at the Beijing Internet Court's decision, it is not hard to find that there is still several aspects to discuss. It also reflects the complexity and diversity of the current copyright issue of AI-generated content. In other countries, there are also different approaches and views on the copyright attribution of AI-generated content. For example, there are cases in India that recognize an AI painting program as a co-author of a work of art, while most countries have yet to grant AI civil subject status. Some countries, such as the US and South Korea, have made it clear that AI-generated content is not protected by copyright, but US law also leaves some room for the copyright of AI-generated content. If a human artist selects or arranges AI-generated material in a creative enough way and modifies it to meet copyright protection standards, the work can still be copyrighted.

Taking a specific case in the United States as an example, although the award-winning work of Space Opera House was generated by AI, author Allen had deeply processed and created the AI-generated content through many revisions and text prompts. Allen considered his creative input to Midjourney to include input prompts, adjusting scenes and choosing the tone of images, among others, these efforts that were similar to the creative process of other artists. Although the U.S. Copyright Office found that the images generated by Midjourney were not copyrighted original works themselves, the portions of Allen's manual edits using Adobe Photoshop were deemed creative enough to be considered for copyright.

This case shows that the US Copyright Office, in determining the copyright ownership of AI-generated content, focuses on analyzing the respective roles and creative contributions of AI and people in the content generation process. This detailed analysis helps us to better understand the nature and attribution of copyright in AI-generated content.

Regarding the intellectual property issue of AIGC, although there are still many gaps in the legal level, the tide of AIGC has been unstoppable.

For traditional creators, they should act quickly to protect their works. In addition to registering with the copyright office, advanced technologies such as trusted time stamps and blockchain storage can also be used to ensure that the rights and interests of their works are effectively protected. If the creators do not want their work to be used for AI model training, they must make a clear statement to protect their rights and interests.

AI developers must strictly ensure the legitimacy of the data when training their models. If intellectual property data is involved, it is important to obtain the relevant authorization to avoid getting into legal disputes. For the outsourcing model, it is more necessary to strengthen the management and control of intellectual property risks in the supply chain and fulfill reasonable duty of care. At the same time, the ownership and other matters should be clearly stipulated in the user agreement to help avoid potential risks.

As for AI users, they also need to be wary of generated content that may infringe others' copyrights. When using AI to generate content, it is important to carefully review it to ensure that it does not infringe on the rights and interests of others. In addition, understanding and abiding by relevant laws and regulations is also an important part of protecting our own rights and interests.