ChatGPT and similar AI applications are now commonly used both at work and for leisure. Instead of using web-based search engines, we seek answers to our questions using these AI applications. We even go beyond that, obtaining songs, poems, and pictures from them. Although controversial, we have AI write scientific articles and other literary works. In return, neither we nor the AI face any copyright restrictions. However, it’s a fact that this situation often leads to copyright issues.

First and foremost, there is no regulation regarding who should own the intellectual property rights of a creative or artistic work shaped by AI according to our requests. Therefore, whether there is a copyright on reusing a work produced by someone else’s AI, and if so, who owns it, is still unclear. But this issue seems less significant compared to another crucial problem: the sources from which AI draws inspiration, references, or even directly copies.

The operating principle of AI involves collecting vast datasets, analyzing these data, and processing them through machine learning with large language models (LLM) to provide outputs appropriate to commands. These datasets can include texts, images, sounds, or other digital content. In essence, AI platforms extract patterns and relationships from this data, form rules, and make judgments and predictions when responding to commands. AI models are typically trained on copyrighted content, which can lead to copyright infringements when AI uses this content to generate new works. For instance, a language model might create new sentences or stories using copyrighted texts. Naturally, using these data without the copyright holders’ consent can become a legal issue. For example, Getty Images, a globally recognized visual media company, has sued the developers of Stable Diffusion, an AI company, over the unauthorized use of its licensed photos. Similarly, in the Andersen v. Stability case, artists Sarah Andersen, Kelly McKernan, and Karla Ortiz initiated a class-action lawsuit at the end of 2022 against AI platforms like Stability AI, Midjourney, and DeviantArt, alleging these platforms used their copyrighted works without permission to train AI models. The core issue here is determining to what extent the writings and works produced by AI are derivative. The plaintiffs argue that the works produced by AI are derived from their copyrighted works and thus constitute copyright infringement because these works are not sufficiently original or transformative. The primary problem is to determine how much AI-generated works are derivative or fair use.

Although these cases are still in the judicial process, legal uncertainties complicate matters further, leading to highly subjective evaluations. Moreover, the ownership of works created by AI remains undefined. While copyright laws provide a framework for humans, they do not offer clear legal guidelines for content generated by AI. Courts have yet to establish precedents on how to apply existing laws in these scenarios.

In addition to the ambiguities regarding the ownership of AI-generated works and the use of unlicensed content in training data, another discussed issue is to what extent we, as users, infringe on copyrights when we use works created by AI. While these topics are still under debate, adhering to ethical standards in AI use can always help minimize risks. AI developers also need to be transparent about the source of their training data and clearly state the origin and licensing status of the data used. In this era of rapid technological advancement, the fact that legal rules often fail to provide timely solutions and are lagging behind is another issue that requires consideration.