In the era of language models, when ChatGPT and Gemini generate text at an industrial scale, or Dall-E synthesizes complex images in an instant, copyright is at the center of the controversy. As usual, technology arrives before the mechanisms to regulate it. OpenAI launched its AI model in open beta at the end of 2022. With it, versions of GPT proliferated, while other companies felt pressured to launch their own model, such as Google with Gemini or Anthropic with Claude. And it wasn't until the end of 2023 that the European Parliament and the European Council reached an agreement on the AI regulation.
The text is still pending its final ratification and gradual implementation over the next few years. It's going to take a while. In terms of copyright, it refers to the existing intellectual property law in each country. This contains the basis for judging all cases involving AI and copyright. But the European regulation establishes a clarity of concepts and obligations necessary. With these elements, it becomes easier to apply copyright law to scenarios involving AI.
Estefanía Asensio, a lawyer specializing in intellectual property and industrial property at Metricson, answers questions from Neosmart on this topic. The lawyer has been advising on intellectual property issues in technology projects for over eight years.
What are the limits on the use of AI that mark the intellectual property law (in text, images, and video)?
Before we talk about what a limit or exception is in intellectual property law, let's define what a limit is. In the Spanish Intellectual Property Law, limits are a series of exceptions that establish the text of the Law of Intellectual Property to the exclusive right that an author has to exploit a work. These are scenarios where it is not necessary to obtain authorization or consent from the creator to use a work. However, the Spanish Intellectual Property Law, like others, does not specifically address the limits on the use of an AI system.
Of course, this technology did not exist when the legislation was drafted in 1996 or when subsequent modifications were made...
But yes, the general principles of the Intellectual Property Law will apply. And they affect how an AI system or model can be used in relation to protected works. In the context of AI use, there are several limits that intellectual property law establishes, especially when we are talking about training AI systems or generating works. For example, AI systems developed for educational or scientific research purposes can make certain uses of these protected works, as long as certain conditions established by the law are met, specifically in Article 32.
To clarify, can a model not be trained with protected works by copyright?
The use of protected works by copyright to train any AI system must always respect copyright. This means that we must have obtained authorizations or licenses from the rights holders or that the use fits into one of the exceptions.
What difference does the purpose for which the AI model is intended make?
If I am developing an AI system for an educational purpose, I must see if it fits into the limit and each of the conditions established by the law for that exclusive purpose. In other words, the work must already be published and there are a series of questions that are established for use. Ultimately, the systems are designed to perform a specific task. When you obtain authorization from the author of the work, you must specify the uses that you will give, which will be permitted to use their work.
How does the EU AI Regulation complement the Spanish Intellectual Property Law?
The EU AI Regulation and the Spanish Intellectual Property Law complement each other. The goal is to harmonize the development and use of AI systems or models within the framework of respecting existing intellectual property rights. The Regulation will focus on establishing the legal requirements for the development, marketing, and use of AI systems, prioritizing objectives such as security, transparency, and respect for fundamental rights, including intellectual property rights.
The revised Spanish Intellectual Property Law is intended to protect creations, works. It does not specifically address AI-related issues. However, with regard to authorship, it does establish that authorship will always correspond to natural persons who create them.
Does this mean that an AI system cannot be the owner of copyright?
We already have the DABUS case, which determines that authorship cannot belong to a machine. You cannot consider a machine an inventor. Ultimately, works generated by AI will need significant human intervention to be protected. Only physical persons can be considered authors.
What other requirements must a work generated with AI meet to be protected by copyright?
The requirements are determined by the law. It must be original. This is an indeterminate concept, but the judiciary has been refining it. And it must be possible to fix it in some medium that allows reproduction and communication.
Regarding the training of AI models, how should a company that develops AI act in terms of intellectual property?
The company must obtain those licenses or authorizations from the rights holders, or apply one of the exceptions we've discussed, or limits, as established by the law. If not, it would be an infringing use of intellectual property. You could be required to pay third parties for making the wrong use.
To conclude, what general recommendations would you give for complying with intellectual property law in this era of generative AI?
My recommendation would be to always ensure that we have the necessary licenses from the rights holders or study whether we can apply one of the limits when using works that we think may be subject to some right, because they are not in the public domain.
Secondly, it would also be recommended that users and developers categorize the system very well to determine the specific obligations that, according to the regulation, we must adhere to. And another case, when we are users of a system, as responsible for deployment, we must study very well who our provider is and what information and documentation they have.
