Creative AI: when new isn’t original

By Frank Liu, Shanghai Pacific Legal
Copy link

Ever-evolving artificial intelligence (AI) tools are used by more and more people to improve efficiency in life and at work. As a result, the relevant IP issues have come into sharper focus.

At present, AI-related intellectual property rights mainly involve copyright. During the process of using big data to collect information for training purposes, AI may utilise the works of others, thereby raising copyright issues.

This article will discuss how AI-generated content constitutes copyrighted work and how to protect it, based on two domestic precedents, specifically:

  • Case 1: Li v Liu (2023)
  • Case 2: Copyright infringement of Ultraman’s image in 2024.

Copyrighted works

Frank Liu, Shanghai Pacific Legal
Frank Liu
Shanghai Pacific Legal

There have always been different views on whether AI-generated content qualifies for copyright protection. In case 1, the Beijing Internet Court deliberated on this very issue.

A copyrighted work consists of four elements, with originality being the most contentious. In case 1, the court held that “whether AI-produced images reflect the author’s personalised expression should be assessed on a case-by-case basis and cannot be generalised”.

The law requires that originality must reflect an author’s personalised expression. Mere mechanical combinations of preset software elements may result in IP that is lacking in originality.

Since the plaintiff selected, depicted and set the parameters of the model, subject, style and presentation mode in designing and selecting the content of the image in question, the court held that the process reflected the plaintiff’s aesthetic preferences and personalised judgement. In the absence of evidence to the contrary, it was affirmed that the image was independently created by the plaintiff and reflected his personalised expression.

Consequently, the court deemed the image to be a copyrighted work.

Copyright holder

The first question to address is whether AI itself can hold copyright. Currently, China’s Copyright Law and Civil Code only recognise natural persons, legal persons or unincorporated organisations as authors. Thus, AI falls outside this scope and does not qualify as a copyright holder under the law. In case 1, the court affirmed AI’s instrumental role, stating it was not a legal person, in discussing the relevant work.

Can the AI designer be considered the author of AI-generated content? According to case 1, if the AI-generated content is simply a mechanical combination of preset elements within a computer program, it may lack originality and fail to meet the criteria for copyright protection under the law.

Even if the content created by others using an AI tool is finally recognised as a work under the Copyright Law, the court in case 1 held that the designer of the AI neither intended to create the image in question nor preset the contents to be subsequently generated. Therefore, the AI designer was considered the producer of an authoring tool rather than the author of the image. Its intellectual inputs were reflected in the production of the authoring tool rather than the intellectual inputs into the image in question. As stated in the licensing agreement provided by the plaintiff for Stable Diffusion, the AI designer disclaimed any rights to the output content. Consequently, the court held that the AI designer in the case did not qualify as the author of the image in question.

In case 1, the plaintiff, as a user of AI tools, contributed intention, ideas, and intellectual input into the content production process, reflecting his or her personality. Therefore, the plaintiff was deemed the copyright holder of the image in question.

Tort liability

In case 2, the defendant AI company provided its members with an AI image generator, which would produce pictures after inputting simple commands such as keywords. However, most of the pictures generated by this tool were substantially similar to the works of the plaintiff’s copyrighted artworks.

Despite the unique nature of the generative AI platform, the Guangzhou Internet Court adhered to general tort principles in determining copyright infringement. The court adopted the conventional approach to copyright infringement of artworks, holding that the defendant AI company infringed on the plaintiff’s rights to copy and edit the works in question on account of substantial similarity.

In addition, the court cited article 4 of the Interim Administrative Measures for Generative Artificial Intelligence Services, stressing that AI service providers should exercise the duty of due care and respect IP rights. With regard to duty of care, the court maintained that the AI service provider should:

  • Establish a complaint reporting mechanism;
  • Warn users of potential IP infringement risks; and
  • Clearly label AI-generated content.

AI service providers that fail to do these three things may trigger a tort liability.

Liability for damages

In addition to the standard orders to cease infringement and mitigate impact, the awarded damages in copyright infringement cases can offer insights into a court’s special consideration.

In addition to conventional factors such as the popularity of the infringed work, the seriousness and impact of infringement and expenses incurred for safeguarding rights, the court in case 2 held that AI “is considered as a major field for developing new quality productive forces … both copyright protection and industrial development should be taken into account, and it is inadvisable to unduly aggravate the obligations of service providers”.

In both cases, the awarded damages reflected a principle of equitable compensation, striking a balance to deter infringements without impeding the growth of the AI industry.

Frank Liu is a partner at Shanghai Pacific Legal

Room 2709, 27/F, Plaza 66 II
1266 Nanjing Road West
Shanghai 200040, China
Tel: +86 21 6086 0199
Fax: +86 21 6086 0111

Copy link