As we have indicated, in the current state of law, as a rule, works created with the use of AI will not constitute works within the meaning of copyright law. Meanwhile, their creation may often lead to infringement of third-party copyrights.
1. Copyright infringement by AI
Very often, AI-based tools that are responsible for creating images or photographs use millions of images on the open web and generate new images in the same style using algorithms and pattern recognition. Midjourney and DALL-E, inter alia, work in this way.
AI systems that are capable of creating new music or films also work in a similar way. AI generates music or short films on its own, but does this thanks to the huge amount of data that has been dedicated to training the AI system in question. This is the principle behind the Jukebox and MusicLM applications.
With this in mind, it is problematic to use a huge database of copyrighted songs to train AI. In addition, there is a risk that during the process of generating a new work, the AI system will make too many references from the existing works, thus creating so-called plagiarism.
The possibility of using other people’s works is only created by the regulations of fair use or a licence agreement. Meanwhile, in the case of using works to train AI, creators point out that no one has asked their permission to do so. A growing number of legal disputes appears to be the obvious consequence in this situation.
Lawsuit against programming app – GitHub Copilot
GitHub Copilot is an AI-based coding tool created and developed by Microsoft, GitHub and OpenAI. The purpose of the tool is to suggest to programmers what code they should write. However, many people noticed that GitHub Copilot’s suggestions were copies of works already created, which the application could access thanks to the Microsoft Azure cloud.
With this in mind, in November 2022, a group of open-source developers filed a class action against the companies accusing them of violating intellectual property rights and open-source licenses. Currently, the case is still pending in a US court.
Lawsuit against image generating app – Stable Diffusion
In January 2023, a class action was also filed by artists Sarah Andersen, Kelly McKernan and Karla Ortiz, accusing London-based Stability AI Ltd and its US subsidiary Stability AI Inc of copyright infringement. The claimants point out that the AI software created, called Stable Difusion, which is dedicated to generating images, has downloaded millions of copyrighted images from various sources on the Internet.
A copyright infringement lawsuit against Stability AI has also been filed in London by Getty Images.
The claimants insist that the images generated by AI may be copies of their works and that they are competing in the market. As a counterpoint, the defendants’ attorneys pointed to the applicability of fair use provisions.
Final rulings in the above-mentioned cases have not been handed down yet.
Lawsuit against Chatbot apps – Meta and Open AI
On 7 July 2023, a lawsuit was filed in San Francisco Federal Court by Sarah Silverman and writers Richard Kadrey and Christopher Golden. In the lawsuit, the claimants allege that their copyrights have been infringed. As they pointed out, the two companies used their books without permission to create language models for AI-based applications. As evidence, the lawsuit includes Chat GPT’s generated responses on the summary of the claimants’ books.
This is another lawsuit against OpenAI. Earlier, on 28 June 2023, a lawsuit was filed by writer Mona Awad and writer Paul Tremblay. In that case, the authors also indicated that their works may have been used to train GPT Chat, which they did not agree to. As one of the arguments that their works were used to train GPT Chat, they pointed out that an app from OpenAI was able to accurately summarize their works.
2. Legal basis for using works to train AI
Some commentators point out that the legal basis allowing the use of copyrighted works for machine analysis of text and data may be the EU Digital Copyright Directive of 17 April 2019 (790/2019), which has not been implemented in Poland. The directive creates the possibility of reproducing databases or works under copyright law for the purpose of machine learning.
However, the conditions must first be met: one must have legal access to the work, while the rightsholder himself does not object to such action. This means that the rightsholder will be able to prohibit companies from training AI based on their works.
The above court cases are only a few selected examples. Currently, the vast majority of disputes are pending in the United States. Since most of the cases date from late 2022 or 2023, final rulings have not yet been made in these cases. However, there is a possibility that the first rulings will be issued later this year.
At present, no similar case involving copyright infringement by AI has been recorded in Poland.
Undoubtedly, the above court cases may currently be problematic for the courts due to the lack of adequate regulations on copyrights and AI. Nevertheless, it seems that the use of a work to provide data for AI to perform a specific task is one way to exploit the work.
Currently, in the absence of specific legal regulations, it will be up to the courts to delineate the boundaries of fair use of works when creating a training database for AI. Accordingly, we will closely follow the case law to be issued in that respect in various jurisdictions around the world. Certainly, the first positive rulings for creators – being the claimants – will encourage other creators to take legal action. Hence, a rapid rise in the number of cases of this type can be expected.