Artificial intelligence (AI) is becoming an increasingly important part of our daily lives, from search engines and automatic driving systems to generative AI (GenAI) such as ChatGPT, MidJourney, and DALL-E. The growing use of these AI systems raises major questions in the legal field. This is particularly true when it comes to intellectual property (IP) and, more specifically, copyright. While artistic creation has traditionally been associated with human beings, today we are faced with machines capable of producing works comparable to those created by humans. The constant improvement of AI makes it harder to distinguish between human creations and those produced exclusively by AI systems. It is therefore crucial to consider how these new technologies fit into our legal system. The relationship between AI and IP raises numerous questions relating to trademark law, patent law, and copyright law. This post focuses on the relationship between GenAI and copyright by giving some technical background and then exploring how different jurisdictions around the world approach the tension between GenAI and IP. Some of the most apparent IP issues lie in the use of GenAI systems, which according to the European Parliament version of the AI Act are ‘AI systems specifically intended to generate, with varying levels of autonomy, content such as complex text, images, audio, or video’. There are two major issues between GenAI and the copyright legal system. The first problem arises before any content is even generated; it takes place during the training phase of GenAI. In training these systems, a huge amount of data is needed to ensure the best quality performance. In this process, copyrighted content will likely be used, which raises the question of fair compensation for creators whose works contribute to the training of GenAI. The second major issue is raised once AI generates the content. Whether these works meet the conditions necessary to trigger copyright protections is unclear. The questions of originality or whether the creation can truly be attributed to the user who generated the content using AI remain debatable as well as if it can even be subject to copyright protections due to its origin. There are currently two positions held by those in the field of technology law: those who consider the existing legal framework sufficient to handle AI-related IP cases and those who believe that new legislation is needed to adapt to the new challenges. As such, it is interesting to see how different countries react to these different challenges and what their positions are regarding existing legal frameworks. The AI Act describes an AI system as ‘a system that is designed to operate with elements of autonomy and that, based on machine and/or human-provided data and inputs, infers how to achieve a given set of objectives using machine learning and/or logic- and knowledge-based approaches, and produces system-generated outputs such as content (generative AI systems), predictions, recommendations or decisions, influencing the environments with which the AI system interacts.’ GenAI is a branch of AI that generates new content based on the input data it has been trained on. This is achieved by using a training database which constitutes the input, i.e. all the data provided so that characteristics can be extracted and patterns drawn. Once the system has learned from the training database it can produce new content that shares the characteristics of the database. The content resulting from the use of this model is the output. Before a GenAI system can produce any kind of content, it must first ‘learn’, a process called unsupervised learning. Unsupervised learning is a type of machine learning that learns from data with no pre-existing labels or categories. The goal of unsupervised learning is to discover patterns and relationships in the data without any explicit guidance. This takes place through the input of huge databases, within which the AI determines patterns and identifies characteristics. The AI system is thus trained using existing content to generate new results. It is during this unsupervised learning process that the first copyright issue arises, which relates to the presence of copyrighted works in training databases. This raises the issue of consent from the artists and fair compensation for those whose work is used in the process. Acknowledging the importance of consent is essential, particularly concerning sensitive information, as with any type of data which would allow a person to be identified. As such, pictures, videos, or any other content containing elements such as a voice, a face, or any individual information making a person identifiable should be treated with particular care. It is even more important in the case where the content owner is not the person visible in the content, as this relates to personality rights, or your right to your image. Since in most cases, a portion of the training data consists of copyrighted works, it is essential to ensure fair use of those works while respecting the content owner as well as the rights of the person on display. Countries have varying elements in their legislation allowing for different levels of protection for copyrighted content. While this topic has not led to any legal battles in Europe as yet, it is a topical issue in the United States where in recent years, multiple companies have been sued for the unconsented use of copyrighted content. In December 2023, for example, The New York Times sued OpenAI and Microsoft for copyright infringements. The New York Times alleges OpenAI and Microsoft used a large number of its articles to train their generative models. Not only is this a copyright violation regarding the unauthorised use of their articles, but The New York Times also alleges that the chatbots are now competing with the media company’s content as a source of information. This is not the first lawsuit against the company. In September 2023, an author’s guild composed of famous American writers such as John Grisham and Jonathan Franzen sued the company for very similar motives – the unauthorised use of their novels. So far, American jurisprudence has favoured the tech companies as the use of content has often been justified under fair use; these companies have yet to be punished. Whether these new cases will bring forward an updated jurisprudence or whether the American courts will stick to their current position remains to be seen. Each legal system has a list of criteria regarding what kind of works trigger the protection granted by copyright, allowing the creator of the work to benefit from the advantages and the protection tied to it. Most legislations include three key criteria. First, the creation must be tied to some form of human intellect or action. Some legislations are more rigorous than others on the importance of human involvement. The second condition is that the creation should be recognisable as a ‘work of art’. Examples of what are considered works of art are usually included in the legislation, but the main idea is that it should be a creation related to a creative field, such as literary, musical, or visual creations. Finally, the work of art must be original; copies and non-original works are subject to a different kind of protection. If we take the Swiss legislation as an example, article 2 of the Federal law on Copyrights and Neighboring Rights defines the notion of a work of art as ‘[w]ork, regardless of its value or purpose, means any creation of the mind, whether literary or artistic, which has an individual character’. A ‘creation of the mind’ means that the work must originate from human creativity. If we take a look at the US legislation, Article 102A from Chapter 1 under Title 17 of the Copyright Act of 1967 states that ‘Copyright protection subsists, in accordance with this title, in original works of authorship fixed in any tangible medium of expression, now known or later developed, from which they can be perceived, reproduced, or otherwise communicated, either directly or with the aid of a machine or device.’ Similarly to what the Swiss legislation meant by the ‘creation of the mind’, the term ‘original work’ in the US legislation means a creation which is ‘independently created by a human author and [has] a minimal degree of creativity’. The element of human creativity and cognitive labour can also be found in China’slegislation on the matter as well as those of Brazil and France. While expressed differently in each text, the core idea remains that any work protected by copyright law has to stem, to some degree, from human creation and creativity. This is where the IP dilemma comes into play. The question is to determine whether individuals using GenAI are making use of sufficient creativity for the work to be attributed to them and protected by copyrights. Different jurisdictions have different ways of assessing human involvement in the use of GenAI. While some jurisdictions seem to consider there to be sufficient human intervention, some seem to think the level of human-based creativity is insufficient to qualify a work for copyright protection. As such it is interesting to look at the most recent cases to see the direction in which different jurisdictions are taking in the matter of GenAI use. In August 2022, Brazil faced a particular case, also raised in several other jurisdictions such as the UK, South Korea, and Australia. This is the case introduced by Dr Stephen Thaler, the president and CEO of Imagination Engines Incorporated. He is the inventor of an AI model called Device for the Autonomous Bootstrapping of Unified Sentience or DABUS for short. In 2019, Dr Thaler filed applications in various patent offices around the world to list his AI model, DABUS, as the inventor of two patentable inventions. This was unprecedented as no one had previously attempted to have an AI model recognised as an inventor. Except for South Africa, Brazil and all other countries to which Dr Thaler applied rejected his request. The Brazilian court rejected Dr Thaler’s application because the title of the inventor must be tied to a holder of personality rights. While Dr Thaler claimed the Brazilian law didn’t explicitly mention the necessity of a human inventor in the case of a patent application, as such opening the possibility for AI to be recognised as the inventor, the court claims this can be deducted by reading the law. Implicitly, it mentions the existence of heirs, for example, or the fact that the title of the inventor is tied to some extent to personality rights and, as such, not applicable to an AI model. In this case, it is important to note that it is currently impractical to attribute copyright-related rights and protections to AI as there is no justification, economic or social, for attributing rights to a system capable of automatically generating works. It also makes no sense from the point of view of liability or indemnification, which are important features of the rights of others, and which would not, in practice, apply to AI models. That being said, in February 2024, a bill was introduced by a Brazilian Congressman to amend the national IP framework. If so far the patent office has denied the title of the inventor to AI models, the Congressman hopes that ‘by allowing that these systems be recognized as patent inventors, we are incentivizing innovation and research on this field, at the same time as we guarantee a legal system that is effective in the protection of intellectual property rights.’ On 23 November 2023, Beijing granted copyright protection to various works created by GenAI. In this case, the plaintiff generated several images using the generative AI model Stable Diffusion; the images were then used by the defendant in an article without crediting the plaintiff. The plaintiff, Mr Li Yunkai, claimed infringement of his rights as the author of these images. The Beijing Internet Court concluded that the efforts made by the plaintiff were a sufficient basis for fulfilling the criterion of human creation as Mr Li Yunkai had to formulate the prompts several times in different ways and in a very precise order to obtain the desired results. The court held that the work reflected the intellectual contribution of the creator and considered this criterion met as Mr Li Yunkai chose the AI service provider among others, selected the prompt after multiple attempts, and modified the parameters to produce the desired images; therefore, his complaint of copyright infringement was justified. Switzerland, contrary to the countries mentioned previously, does not yet have any significant jurisprudence on the matter of IP rights and AI use. But the confederation has made its stance clear on the matter as it considers the current legislation to be sufficient and as such needs no change in the foreseeable future. While this is a position held by many, the frustrations tied to the unclear status of input and output data will most likely lead to some change in the future as people become increasingly aware of these issues. The United States holds a clear stance on this topic as it has repeatedly refused to grant copyright to artistic creations generated by AI. This was the case, for example, of Jason Allon, whose work Théâtre d’Opéra Spatial won first prize at the Colorado State Digital Art Fair in the United States in 2022. It took a considerable amount of time to find the prompt that enabled the GenAI MidJourney to generate this work. Despite this effort, the United States Copyright Office ruled that the work lacked any connection with its human author and, therefore, fell outside the scope of US copyright law, which excludes works produced by non-human entities. This is not the first time the Copyright Office has ruled in this way, its jurisprudence showing that it considers the use of AI models to generate works of art to fall outside the scope of the law which seeks to protect individual intellectual creations. Another example would be the case of A Recent Entrance to Paradise, an image generated by DABUS, the AI model designed by Dr Thaler. Similarly to the previous case, the US Copyright Office refused to grant Dr Thaler authorship of the image due to a lack of human involvement. The judge explained that human involvement is one of the core elements of US copyright law and it is as such impossible to attribute copyright protection to this work. These different cases give us an overview of the current challenges faced by jurisdictions in the matter of AI and IP. While there is no clear international consensus on the matter, it is obvious governments are becoming aware of the growing impact of AI in the IP space and are taking a stance on the matter. It will be important to come up with solutions that can satisfy all parties involved, as much the copyright owners as the AI companies who would like to make use of the content. The question of the ownership of AI-generated content should also be answered to give users the necessary clarity when using such models.Introduction
Technical background
Input
Output
Legal foundation
Case law regarding copyright and the use of AI
Brazil
China
Switzerland
United States
Conclusion