Investments in the field of artificial intelligence (AI) are actively growing. According to the survey "Goldenman Sachs", by 2025, the volume of investments may reach 158 billion dollars. Big companies like Microsoft, Adobe, and GitHub are integrating AI into their products, and startups are raising vast amounts of money to compete. Most AI systems learn on large amounts of data from the Internet, be it text, code, or images. This data belongs to other persons. Copyright holders must obtain permission to copy and use the content.
In September 2022, the US Copyright Office approved the registration of a unique comic book created with the help of AI Midjourney.
After that, the trio of artists filed a lawsuit against Midjourney and the DeviantArt platform. They claim that these organizations violated the rights of millions of artists by training their artificial intelligence tools on five billion images collected from the Internet without the authors' consent.
In a blog post, the plaintiffs' attorney describes the case as another step toward making AI fair and ethical for all. He believes that the ability of AI art tools to flood the market with an unlimited amount of infringing images will harm the market for art and artists.
Today, it is known that the US Copyright Office is reviewing its decision. The artist who created the comic said the US Copyright Office asked her to provide more details about the process. It will help prove that there was a significant human contribution to creating a graphic novel.
Copyright in the results of an AI model will likely depend on the degree of human involvement. If the author can prove that he spent several weeks refining the keywords and manually editing the finished work, this will indicate a relatively high level of intellectual involvement.
In our opinion, training AI on copyrighted works falls under the fair use doctrine. It allows limited use of such materials without the permission of the rights holders.
Factors considered when determining fair use:
1. Transformative nature of use. New content created by AI must be different from the original work. In addition, such content must be created to comment on, criticize, or parody the original work.
2. The nature of the author's work. It is essential to determine whether the original work (based on which the AI output was generated) falls under the strongly protected author material category. The degree of protection is an evaluative concept. It is determined by the law enforcement body, considering the text's nature. If it is, for example, data from a documentary work, then this can be an argument in favor of fair use. On the other hand, if information is borrowed from a literary text, the court will probably not recognize this as fair use.
3. Scope and materiality of use. The magnitude and significance of the used part of the original work in the context of the AI output also matter. If the AI uses only a limited part or an insignificant amount of the original work, this may affect the fair use decision. However, using even small pieces may be considered bad faith if it uses the "heart" of the copyrighted material.
4. Impact on the potential market or value of the work. Consider whether the result of the AI generation may affect the market or the value of the original work should be considered. If the AI result does not compete directly with the original and does not cause damage to its market, this can contribute to the court's recognition of such actions as fair use
When applying AI to generate your content, it is worth trying to edit the finished work using a small part of the author's works and the least protected categories of information. Under such conditions, the chances that the copyright for such a work will be registered increase significantly.