Farangiz Bakhtiyorova

Image source: www.executiveeducation.blog

Introduction

These days generative AI systems find widespread application not only in Uzbekistan but also globally across diverse professional domains. Generative AI is a set of “foundation models used in AI systems specifically intended to generate, with varying levels of autonomy, content such as complex text, images, audio, or video.” (Art. 28b (4) EU AI Act).  Research indicates that an estimated one-sixth of world’s populace, amounting to approximately 1.33 billion individuals, use generative AI technologies on a daily basis, prompting significant reconsideration of how copyright law principles—such as authorship, infringement, and fair use—apply to content created or utilized by AI.

This blog post will explore potential legal issues, regulatory gaps, and copyright protection risks associated with generative AI, alongside potential solutions to these challenges.

Challenges in the Context of Copyright: Legal Dilemmas and Solutions

Generative AI programs, including OpenAI’s DALL-E and ChatGPT, Stability AI’s Stable Diffusion, and Midjourney’s eponymous program, possess the capability to produce new images, texts, and other forms of content in response to user-provided textual prompts. The prevalent utilization of generative AI programs prompts inquiry into copyright infringement and rights of use issues, uncertainty about ownership of AI-generated works.

Firstly, generative AI technologies have advanced to the point where they can produce texts, images, and videos of remarkable quality, often indistinguishable from human-created content and possessing unique aesthetic attributes. For instance, platforms like Stable Diffusion and Midjourney excel in harmonizing colors to generate visually striking images on demand. This technological feat has led to remarkable instances in the art world such as the Museum of Modern Art in New York has already showcased an AI-generated installation derived from the museum’s own collection.

Such developments, however, bring forth complex legal questions regarding authorship and copyright in the context of generative AI. Specifically, the determination of authorship for AI-generated works, the applicability of existing copyright laws to these creations, and the potential recognition of both AI and human contributors as co-authors are critical issues for consideration. This evolving landscape necessitates a thorough examination of copyright legislation to address the unique challenges posed by generative AI in the creative domain. Whether generative AI’s content is copyrightable or not is a subject to different laws and regulations from different countries. For example, according to Law No. 9.610 from 19 February 1998 (Brazil), Chapter II, Article 11, the author of a literary, artistic, or scientific work is defined as the natural person who created it. According to the Law on Copyright and Related Rights (Serbia) the holder of the copyright must be a natural person. In contrast, China, the United Kingdom, and several other jurisdictions have established mechanisms pertaining to the acknowledgment of authorship in the realm of generative artificial intelligence (AI), as well as addressing the copyright implications arising from collaborative endeavors involving human-AI interaction.

According to Law on Copyright and related rights in Uzbekistan, a holder of the copyright must be a natural person, with an exception of films, where the copyright holder may also be a legal entity, such as the producer whose name is indicated in the film itself.  An object of copyright must be a spiritual creation of the author, which is incompatible with the nature of AI. But the law does not specify any certain rules regarding the copyright of outputs created by generative AI or generated as a result of human requests and instructions. It is crucial to clarify the issue of generative AI and authorship, as the significance of authorship is paramount in several contexts. These include the commercial use of AI-generated copyright objects, the formation of contractual relations, the exercise of any authorial rights, and matters of obligation and liability. For instance, in 2018, a French art collective called Obvious used an AI system to create a portrait titled “Edmond de Belamy.” The artwork sold at auction for over $400,000, raising questions about ownership and attribution of AI-generated art. While the collective was credited as the creator, the AI system’s role in the artwork’s creation remains unclear.

Problems related to copyright infringement

Secondly, unlike the previous problem, which highlighted an authorship loophole for human-AI created works, this concerns potential infringements of copyright by generative AI outputs. ChatGPT was trained on approximately 45 terabytes of text data, equivalent to about one million feet of bookshelf space. If generative AI utilizes copyrighted material without content, it engages in copyright infringement. However, generative AI companies argue that training AI on copyrighted works does not constitute copyright infringement. These models do not copy the training data but rather learn associations between elements such as words and pixels. Companies like Stability AI, the creator of the image generator Stable Diffusion, contend that the output images generated in response to specific text prompts are unlikely to closely match any particular image in the training data. Despite these arguments, several cases of generative AI copyright infringement have been litigated globally.

Case Studies: Insights into Copyright Infringement

Case 1

In a recent lawsuit in US against OpenAI and Microsoft, former journalists Nicholas Basbanes and Nicholas Gage allege infringement on writers’ exclusive rights. Basbanes, in an AI Journal interview, emphasized his motive to defend writers’ rights under Clause 8 of the First Amendment. This case highlights the writing community’s concerns over AI’s use of copyrighted content and underscores the importance of US federal law in protecting these rights to maintain writers’ trust in constitutional safeguards.

Case 2

In China, the Guangzhou Internet Court ruled that an AI company infringed on the plaintiff’s “copyright and adaptation rights” to the Ultraman works while providing generative AI services. The court ordered the defendant to pay 10,000 yuan (approximately $1,389) in compensation for the plaintiff’s economic loss.

Liability issues

Thirdly, critical question remains regarding who should be held liable if generative AI outputs infringe on existing copyrights. In Uzbekistan’s legislation, there is no explicit obligation for the creators of chatbots or similar algorithms to avoid violating copyright while using data. As a result, it is challenging to apply legal grounds if such a claim were to be brought before a court. This gap in the legal framework highlights the need for specific regulations addressing responsibilities of AI developers and users in relation to copyright infringement to ensure comprehensive protection of intellectual property rights. As an example, the lawsuit against Stable Diffusion implicates AI companies for copyright infringement in the USA. According to legal doctrines, both AI users and companies could be held responsible. For instance, even if a user is directly liable, the AI company could face liability under the doctrine of “vicarious infringement.” if it has the right to supervise the infringing activity and a direct financial interest in it. In light of AI-generated content, it is crucial for legislators in Uzbekistan to clarify the scope of copyright protection in the context of AI systems.

Feasible solutions

The legal complexities of AI authorship, attribution, and copyright infringement highlight the urgent need for regulatory development. Despite initiatives  “Digital Uzbekistan – 2030” Strategy and AI implementation decrees to develop AI, a comprehensive legal framework for AI remains lacking. This gap poses significant legal challenges as AI adoption surges. To mitigate these challenges, following decisive measures are essential:

1. It is crucial to delineate AI’s status within copyright laws and offer guidance on the ownership of AI-generated or human-AI collaborative copyright objects. As an international experiences, the UK’s Copyright, Designs and Patents Act (CDPA), Section 9(3) clearly provides that the person who made the “arrangements necessary for the creation of the work” is the author of the resulting copyrightable work. Consequently, if a person has trained or directed the AI to perform a particular task, that person must be recognized as the author of the resulting work. This particular recognition is also allowed in China, as a recent case shows, the Beijing Internet Court ruled that an image produced by Stable Diffusion satisfied the criteria of “intellectual achievement” and “originality.” This determination stemmed from the user’s intellectual contribution through inputting prompt texts and setting parameters, resulting in the AI-generated image reflecting a personalized expression of the user.

2. AI developers, both in Uzbekistan and worldwide, must proactively address different forms of data sourcing practices. Notably, models like Stable Diffusion and Midjourney have leveraged datasets like LAION-5B, comprising vast amounts of web-scraped images, including copyrighted content. Therefore, conscientious data acquisition strategies are indispensable for upholding ethical and legal standards.

3. Establishing liability frameworks for copyright infringement by generative AI systems is pivotal. Moreover, establishing rules for transparency and academic integrity for generative AI users. Jurisdictions like the US, South Korea, and Japan have enacted copyright laws to safeguard publishers from AI-related copyright infringements. The European Union’s EU AI Act categorizes generative AI models, such as ChatGPT, as non-high-risk entities but subjects them to the transparency requirements and EU copyright law.  According to these requirements, the AI’s role in content generation must be disclosed, and the summaries of copyrighted datasets used during the training of generative AI must be provided. Besides, the AI-generated or modified content, like deepfakes, necessitates a clear labeling to ensure awareness among its users.

Conclusion

To conclude, the legal problems mentioned above are of concern not only in Uzbekistan, but around the entire international community. Given Uzbekistan’s AI development policy and its country’s current extensive usage of generative AI, the enactment with a separate section on AI regulations or the amendments to existing copyright legislation are imperative in order to further successfully navigate inevitable copyright-related complexities.

Cite as:  Farangiz Bakhtiyorova, “AI vs. Copyright: Navigating the Legal Maze of Generative AI Content Creation”, Uzbekistan Law Blog, 30.06.2024.