The legal implications of generative artificial intelligence—such as ChatGPT, Dall-E, and Midjourney—are countless: From privacy issues relating to the processing of personal data without the permission of users, to civil and criminal liability of the information generated by these tools. However, one of the most interesting legal issues is the extent to which these new technologies conflict with copyright law, especially as AI tools such as ChatGPT have made it extremely easy to create what appear to be new works based on information freely available on the internet. The question here is simple: Do the works produced by ChatGPT violate the copyright protection of the underlying works from which ChatGPT obtained its information? Spoiler alert: Under Omani Copyright Law, the answer is yes.
ChatGPT incorporates many technologies to be able to respond to a question written in natural language and hold a conversation about it, but the most important aspect of these technologies from a copyright perspective is the training component, which is a data-gathering process that involves feeding the model massive amounts of information so that it learns from it by identifying patterns and trends. ChatGPT was trained by scanning millions of websites, books, and articles, and it has learnt its information by interpreting the content of these sources and identifying the patterns within this content. So when you ask ChatGPT to tell you how the term “ship” is defined under public international law, ChatGPT uses the information it has obtained from previous training on the vast information available on the internet on this topic to generate an answer.
If ChatGPT simply copied the ideas and abstract facts from underlying works, that would not be in itself an infringement because copyright law does not protect abstract ideas, but the expression of these ideas. This is explicitly stipulated in article 4 of the Omani Copyright Law:
Protection does not extend to mere ideas, procedures, methods of operation, mathematical concepts, principles, discoveries, and data. […]
Copyright and Neighbouring Rights Law, Royal Decree 65/2008 (issued 4 May 2008, published 17 May 2008) OG 863, art 4.
However, what ChatGPT does in the training part of its process is copy—whether temporarily or permanently—all sorts of works found on the internet, and this very aspect of the process infringes on the exclusive economic right of reproduction granted by article 6(a) of the Omani Copyright Law.
Of course, the fact that a work is made available for free on the internet does not mean the work can be legally copied by others. Copyright protection is automatic and does not require the author of the work to undertake any formality for the protection to be effective. This practically means that any work posted on the internet, from articles published in the Wall Street Journal, to photos posted on Instagram, are protected by copyright.
A common defence that people have in regard to using the works of others is “fair use”, but fair use is an American legal doctrine that does not exist under the laws of Oman, the UK, France, Japan, and the majority of countries around the world. What we have in Oman is a specific list of exceptions to copyright protection under article 20 of the Omani Copyright Law that permit certain specific uses to be carried out without the need to obtain the permission of the author, such as the right to quote a small part of a work for the purpose of criticism or review, the right to perform a work for the purposes of face-to-face education, the right to make a single backup copy of a software for use in case the original copy of the software is damaged, and other very specific exceptions. Outside these exceptions, it is not permitted under Omani Copyright Law to copy a work without permission, even if the use is personal, non-commercial, and attributes the author.
The situation under the copyright laws of other countries may be different. For example, the United States has the fair use doctrine under its copyright law that makes it possible for any use to be permitted without the permission of the author as long as the use satisfies flexible criteria relating to the purpose of the use, the nature of the original work, the portion copied, and the effect of the use on the exploitation of the original work. Fair use provides flexibility, but also results in a degree of uncertainty, and at this point, it is still not clear if ChatGPT is permitted under the fair use doctrine in the US as the matter has not been tested by the courts.
The position under Omani Copyright Law as it stands is very clear. Training AI models involves creating copies of copyright-protected works, and since the permission of the authors of these works has not been obtained, this use violates the law. Whether or not ChatGPT should be permitted by law is not an easy question to answer. ChatGPT and other AI tools are, without a doubt, a groundbreaking development in the use of technology, and it is humanly impossible for a model such as ChatGPT to go and seek the permission of the owner of every single page on the internet before that page is used to train the model. However, the extent to which ChatGPT takes content from other sources, repackages it, and eliminates the need for users to go back to the source is not reasonable. If the source websites are unable to generate money from advertisements and other sources of revenue because ChatGPT absorbed all their content and gave it to the users in the form of short answers, these websites would cease to exist, and that would not only be bad to the economy and to society, but ChatGPT itself would not be able to find new sources of content to gain new knowledge from.
This is not the first time a groundbreaking technology tests the limits of the legal system. When Napster came out more than 20 years ago, it disrupted the music industry, but it was shut down because the courts decided that it was, in fact, illegal—and that only happened after it became a worldwide phenomenon.
The legal system is very slow, and it is impossible to predict how it will respond to ChatGPT. Countries would have to revise their laws to permit machine learning under copyright law, at least under certain conditions. The technology providers would also have to figure out ways to attribute the sources of the material and allow them the opportunity to obtain some revenue streams to make these sources sustainable.
