Lawyer's opinion: Is the "feeding" AI painter a creation or a theft?

**Original:**Xiao Sa lawyer

Core Tips

1. According to Article 20 of the "Administrative Measures for Generative Artificial Intelligence Services (Draft for Comment)", AIGC service providers "will be investigated for criminal responsibility in accordance with the law if a crime is constituted."

2. "Feeding" AI paintings by AI painting service providers does not infringe the right of reproduction and the right of information network dissemination, and will not constitute a crime of copyright infringement.

3. If the copyright owner takes technical measures, and the AI painting service provider avoids the technical measures to obtain pictures through "crawlers" and "feeds them", it may constitute a crime of copyright infringement.

Although the service provider of AIGC only outputs according to the user's requirements, and the user has a strong freedom to decide how to use it, as a platform, it also needs to bear certain social responsibilities. Article 20 of the "Generative Artificial Intelligence Service Management Measures (Draft for Comment)" issued by the Cyberspace Administration of China on April 11, 2023 mentions that aigc service providers "will be investigated for criminal responsibility in accordance with the law if a crime is constituted." Sister Sa's team has been in compliance before | "Feeding" AI paintings, or is it an infringing work? ! In the article, we have studied the legal nature of "feeding". From the perspective of AI creators, we believe that AI creation does not constitute infringement under normal circumstances, but it still needs to be judged on a case-by-case basis. Today we will talk about the criminal legal issues involved from the perspective of AIGC service providers.

First of all, let's review the principle of AI painting. Specifically, AI painting has a three-layer mechanism. The first layer is the basic logic layer. AI painting can complete the transfer of style. The second layer is the database. On the basis of the first layer of basic logic, AI needs a large number of "feeding pictures" to learn and summarize, and obtain different image parameter examples. This step is the so-called deep learning process. The third layer is creative output. The power of AI painting is that the pictures it generates not only conform to the text description, but more importantly, it can create images that conform to aesthetic logic. On the basis of the previous step, AI needs to tell it which result is beautiful through human engineers, and adjust the proportion of increasing such output results. This step is example learning.

Through long-term reciprocating deep learning and example learning, AI has mastered some general drawing rules, and corrected the model by summarizing the rules. Therefore, AI painting can be roughly divided into three stages of work, data collection, data processing, and image production. "Feeding the map" is the core of the second step of "deep learning", and it is also a controversial behavior. An excellent AI painting model must be supported by a huge database, so many service providers will choose to use "crawlers" to obtain large amounts of data. The space is limited. In the following, we mainly analyze whether the behavior of “feeding” AI paintings after obtaining pictures through “crawlers” constitutes a risk of copyright infringement for AI painting service providers.

The "Copyright Law of the People's Republic of China" (hereinafter referred to as the "Copyright Law") adopts an enumerated approach to the protection of the rights of copyright owners, and in Chapter V "Protection of Copyright and Rights Related to Copyright" lists many copyright infringements in detail. the behavior of. However, only six acts are listed in Article 217 of the "Criminal Law" in the crime of copyright infringement, and this is still in the latest amendment to the Criminal Law - "The Amendment to the Criminal Law of the People's Republic of China (Eleven)" (hereinafter referred to as "" "Revision Eleven")—increased from four to six. The "Revision Eleven" made major changes to this article, not only increasing the upper limit of the sentence, but also making certain changes to the constituent elements. On the one hand, in addition to the behavior of "copying and distribution" that constitutes the elements of copyright infringement, the behavior of "dissemination through information networks" also constitutes the elements of copyright infringement; sabotage of technical measures” imposed restrictions.

With regard to data collection, we will mainly discuss whether the first and sixth items of this article will be violated below.

The first of which mainly involves the issue of copy distribution and information network dissemination.

The provisions on reproduction and distribution protect the reproduction right of the copyright owner. According to Item 5 of Article 10 of the "Copyright Law", the right of reproduction refers to: "the right to make one or more copies of a work by means of printing, photocopying, rubbing, recording, video recording, dubbing, remake, digitization, etc." Reproduction is to fix a work on a tangible material medium in a known or unknown way, so that the work can be perceived, disseminated, and copied by others. Therefore, we believe that it generally constitutes a copying act in copyright law, which should reproduce the work on a tangible material carrier. Only through a certain material form can the work gain fixity, so that the original and the reproduction have a clear contrast. When it is stored as data, and its image parameter example is obtained, it is difficult to determine that it violates the author's right of reproduction when it has not been output.

The regulations on information network dissemination protect the copyright owner's right of information network dissemination. According to Item 12 of Article 10 of the Copyright Law, the right of information network dissemination refers to the right to provide works to the public in wired or wireless ways, so that the public can obtain the works at the time and place they personally choose. Generally, we believe that the database of AI painting service providers is non-public, and the public has no chance to directly obtain the works. On the other hand, can the public obtain the original works indirectly through the use of AI painting services? We also think it is unlikely. The core technical capability of this type of AI generative model is to represent the content created by humans with a certain high-dimensional "vector". If this content-to-vector "translation" is reasonable enough and can represent the characteristics of the content, then all the creative content of human beings can be converted into vectors in this space, then we may restore the original vector by giving all the vectors Content. But at present, our "translation" obviously does not have this ability, that is, a large amount of real-world content cannot be summarized by the "vector" of the AI system. Therefore, we believe that even if you want to trace the source through keywords, the public still cannot obtain the original work, and the AI painting service will not infringe the copyright owner's right to information network dissemination.

其中第六项主要涉及“反向工程”的认定问题。《著作权法》第四十九条赋予了著作权人采取技术措施的权利，现实中随着人们权利意识的增强也往往开始采取设置浏览权限等方式保护自己的作品，此时“爬虫”无法直接获取作品，但通过一定的技术手段也可避开技术措施获得作品。目前我国还没有关于因通过“爬虫”绕过或破坏技术措施获取数据而认定为侵犯著作权罪的裁判案例，但存在因为未经著作权人许可，复制游戏数据后修改其作品采取的保护著作权或者与著作权有关的权利的技术措施，而被认定为侵犯著作权罪的案例（案号：（2022）沪0107刑初81号）；民事案件中，也存在将通过“爬虫”避开技术措施认定为侵犯信息网络传播权的裁判（(2016) Beijing 73 Minzhong No. 143). To sum up, we do not rule out the possibility that obtaining data through "crawlers" bypassing or destroying technical measures may be deemed a crime of copyright infringement.

At the same time, according to Article 50 of the "Copyright Law", there are exceptions for circumvention of technical measures, that is, circumvention of technical measures is allowed if the conditions are met.

But obviously after searching, we can't classify the behavior of "crawler" into the above scope. The amendment to the sixth item of copyright infringement in the "Revision 11" discussed above is a product of the revision of the Copyright Law in 2020, and it has inherited the attitude of opposing both direct and indirect circumvention behaviors. However, most of the opinions in the academic circle believe that we should not prohibit direct circumvention behavior, that is, direct circumvention behavior should be regulated by checking whether its subsequent use behavior violates copyright and constitutes direct infringement. Following this line of thought, there is room for reviewing whether it meets the relatively broad "fair use", that is, "copyright restrictions" in my country's "Copyright Law". However, taking a step back, even if the above legislative viewpoints are adopted, "feeding pictures" can hardly be included in the category of "fair use".

Based on the above, we believe that the act of “feeding pictures” generally does not have the risk of constituting a crime of copyright infringement, but if “feeding” AI paintings through “crawlers” and avoiding corresponding technical measures, it may constitute a crime of copyright infringement .

write at the end

"Copyright" formally appeared and developed in the 18th century, and was established in 1709 by the enactment of "Queen Anna's Act" in England, more than 300 years ago. The essence of copyright or authorship is to provide legal protection to the fruits of intellectual labor by endowing it with a certain monopoly, thereby encouraging intellectual labor. With the changes of the times and the advancement of science and technology, people's intellectual achievements are showing a trend of diversification. Copyright protection through positive enumeration and identification will inevitably lag behind the development of the times. On the one hand, we should encourage a future of sharing and openness. In the information age, data sharing and use will be the wave of the times; ways to protect copyright. Regardless of the current nature of the act of providing AIGC services, its development in the field of painting has had a huge impact on the industry. It is an indisputable fact that a large number of original painters are unemployed, but the profession of "AI painter" is rising. While we are considering whether AIGC-related creations violate the relevant laws and regulations on intellectual property rights, should we give more humanistic care to those people who are unwilling to accept the AIGC era?