Author: Yingying Zhu, Partner at BEIJING MINGDUN LAW FIRM

Email: zhu.yingying@mdlaw.cn

Date: July 24, 2024

With the flourishing of AI technology, works created by AI could outpace and exceed those done by humans in many aspects, as AI is being trained using deep learning algorithms to analyze vast amounts of data and to learn patterns, styles, and structures, while human brains normally cannot compete in that depth of training. While people around the world are enthusiastic about the continued pushing-back of boundaries for literary and artistic creations by AI, in the copyright realm, some fundamental questions remain to be answered-Who is the author of the works generated by AI? Who owns the copyright to the AI generated content? Currently, those questions could have very different answers in China and in the US.

Beijing Internet Court’s Judgment vs. U.S. Copyright Office’s Decision

1. The AI-generated Image “Spring Breeze brings Tenderness" Case in China

In a copyright infringement case before the Beijing Internet Court (“BIC”), the plaintiff Li used artificial intelligence Stable Diffusion to generate the disputed image by inputting prompt words and published the image under the name of "Spring Breeze brings Tenderness". Later, the plaintiff Li found that the defendant Liu published an article in a content sharing public platform and used the disputed image in the first page of the said article. The plaintiff claimed that the defendant used and published the image the plaintiff created without permission, which violated the plaintiff's right of information network communication and also the plaintiff's right to authorship.

After trial, the BIC held that the disputed image in this case showed differences from the previous works. From the perspective of the generation process of the image, the plaintiff Li obtained the first image by entering prompt words and setting relevant parameters, and then the plaintiff continued to add prompt words, modified parameters, and constantly adjusted and amended, and finally obtained the image. The BIC considered that “this adjustment and amendment process also reflects the plaintiff's aesthetic choice and personal judgment”. In the absence of contrary evidence, the BIC concluded that “the disputed image was independently completed by the plaintiff Li, reflecting the plaintiff's personalized expression and it belongs to the works of art under the plaintiff’s authorship”. Accordingly, the BIC ruled that the defendant infringed upon the plaintiff’s copyright and the defendant is ordered to post a public statement of apology, eliminate the negative impact, and compensate the plaintiff for economic losses1.

2. The AI-generated Images “Zarya of the Dawn” Case in the US

In this US case, shortly after registering the Work titled Zarya of the Dawn, a “comic book” consisting of mixed texts and images, the U.S. Copyright Office (“USCO”) became aware of statements on social media attributed to the artist Kashtanova that she had created the images in the comic book using artificial intelligence Midjourney. The USCO, after reviewing the letter from Kashtanova’s lawyer, decided to cancel the copyright registration, arguing that “the process (of creating the images) is not controlled by the user because it is not possible to predict what Midjourney will create ahead of time.” The artist Kashtanova objected that she has played an active role in guiding the creation of the images through “hundreds of thousands of prompts” and through making corrections to some of the outputs using image-editing software. The USCO was unconvinced, asserting that “the process described … makes clear that it was Midjourney – and not Kashtanova – that originated the ‘traditional elements of authorship’ in the images”. Then the USCO approved the registration for the texts and arrangement of images2 of the comic book under the authorship of Kashtanova, but denied protection for the individual images within the comic book. 

Based on the cases discussed, it can be inferred that in China, the judge holds that the human-user’s act of entering and adding prompt words and setting and modifying parameters for the AI-generated work is considered a sufficient creative input to qualify for copyright protection while the US Copyright Office does not currently offer protection for literary and artistic works generated by AI because such works are not considered the product of human authorship. The cases above illustrate a lack of coherent global approach when it comes to the determination of authorship over the AI-generated works.

Discussion

1. The Idea-Expression Dichotomy 

In determining the appropriate scope of copyright protection in the AI era, it seems essential to come to the very basics of the copyright law, that is, the idea-expression dichotomy, which is a widely recognized principle in copyright law defining that ideas are not protectable under copyright laws, but expressions of those ideas can be legally protected.

In a typical AI generated content scenario, human-user provides prompts in the format of short phrases or sentences to AI, while AI, by processing those prompts, can automatically generate original image or literary contents that imitate human-created artworks or articles. The question of who the author is seems simpler if it could be determined who contributes to the ideas and who contributes to the expressions.

In the opinion of the renowned Chinese copyright law scholar Mr. Wang Qian, for freehand sketching and drawing software, when the human user creates, the user’s thoughts directly act on the expression. For example, when an artist wants to draw an "apple", the artist needs to choose the color of the "apple", draw the general outlines, and deploy his or her thoughts to dominate the forms of expression, and then the final creation as well.

However, in AI drawings, the thoughts generated in the user's mind cannot dictate the specific forms of expression generated by the AI. Although the two, the drawing software and AI, are both tools, in the traditional drawing method, people are still involved in the creation and the use of tools to express, but in AI drawings, people only provide ideas, and AI completes the expression part independently3. 

Deducing from the Idea-Expression Dichotomy, which is a globally recognized cornerstone of copyright law, the human-user contributes to the ideas while AI contributes to the expression of such ideas. If the authorship could ever be vested in a non-human, that is, a software application or a machine, AI should be the author of the works generated by AI, and AI owns the copyright to the AI generated content.

2. AI the Author or AI just pieces Everything together?

In the view of the author (who is not AI, by the way) of this Article, AI does not do the expression part either-the expression has been done by numerous human authors and artists who originated the various elements that confirm to the prompts entered into the AI by the user. AI is not the author. AI just pieces everything together.

As mentioned above, AI is being trained using deep learning algorithms to analyze vast amounts of data and to learn patterns, styles, and structures created by humans. AI is good at mimicking but not so good at creating. The creation process has been done by numerous human authors and artists over the centuries in human history. AI searches from its training materials these prior works and grabs the elements confirming to the prompts entered by the user and then pieces the relevant elements together to generate the final “work”, which, in the author’s view, is hardly a creation process.

That is why protecting the “real author” behind the content generated by AI is so important to accommodate the tremendous advancements in AI technology. In this direction, recently, a bipartisan group of senators in the US introduced a new bill The Content Origin Protection and Integrity from Edited and Deepfaked Media Act (“COPIED Act”) to make it easier to authenticate and detect artificial intelligence-generated content and protect journalists and artists from having their work gobbled up by AI models without their permission. The COPIED Act would direct the US National Institute of Standards and Technology (“NIST”) to set up standards and guidelines that help prove the origin of content and detect synthetic content, like through watermarking4. 

The danger of “AI plagiarism” also poses another problem - if the human user becomes the legitimate right holder of any AI generated work just by entering prompts, setting parameters and making adjustments, AI might be utilized as a tool of plagiarism at no cost to the human user, leaving the copyright and title of the “real author” behind the content generated by AI in great peril.

Conclusion

With the breathtaking development and advancing of AI technology, the lingering doubts and fierce debates over the authorship of AI generated works have been on the rise globally in recent years. The vastly different approaches of legal authorities in China and the US shows that there is a lack of coherent global recognition of the role played by AI in the creating or generating of literary and artistic works. Simply put, AI is developing too fast, while the law is left behind.

When one attempts to answer the question of authorship within the context of AI-generated works, it becomes apparent that there is no straightforward answer, and it would be difficult to find out or set a unified global approach at the current stage. Furthermore, the protection of human copyrighted content and other fundamental rights such as the personality rights (images, likeness, voice, etc.) of people from “AI plagiarism” during the AI training or data feeding processes also presents significant legal, ethical and practical challenges.

 1.Beijing Internet Court, (2023) Jing 0491Min Chu No. 11279.

 2.Robert J. Kasunic, U.S. Copyright Office, Zarya of the Dawn (Registration # VAu001480196) (2023).

 3.See https://baijiahao.baidu.com/s?id=1802456776111016522&wfr=spider&for=pc.

 4.See https://www.theverge.com/2024/7/11/24196769/copied-act-cantwell-blackburn-heinrich-ai-journalists-artists.