Who Owns AI-Generated Art? Exploring Copyright Complexities with Jon Ellrose

Legal questions surrounding copyright and intellectual property in general have become more critical than ever. To shed light on these complexities, we sat down with Jon Ellrose, the Owner and CEO of Just A.I. and a leading consultant in AI law. 

Jon specializes in intellectual property law on a global scale, with a deep focus on the emerging challenges in both common and civil law systems as they grapple with AI's disruptive potential. His expertise spans jurisdictional analysis, legal methodology and case law related to AI. He also guides companies in implementing management systems and compliance frameworks like ISO 42001, NIST and the EU AI Act. Jon has worked in the educational field in Sweden most of his life as a teacher and as a principal before starting his legal consultancy firm Just A.I.

In this interview, Jon provides in-depth insights into the challenges and opportunities presented by AI in the context of copyright law.

Interview Q&A with Jon Ellrose

Q1: In your opinion, who should hold the copyright when AI creates art or music? Is it the developer who creates the AI models, the user who gives the prompts, or the AI itself?

Jon Ellrose: "I'm sorry to disappoint, but there is no simple answer. When it comes to copyright, it's all about examining originality and creative control when ideas get fixed in a tangible medium of expression. The view that we have on AI, compared to a camera for instance, is that you don't have control of the first picture that you receive if you talk about a diffusion model, and you don't control either a tune that you create or even a text that you create through prompts. You can have copyright on the actual prompt if that meets the standards of originality; that the prompt is not a fact and it is not a commonplace sentence used in everyday life or to short to qualify for an original expression. You can however get Copyright protection of a compilation of AI images if you view the work as a whole, but not the actual images generated by AI. That was the case for instance with the comic Zarya of the Dawn. If the AI part in a picture is de minimis as in the original picture is just improved by AI then you can possibly get copyright if the AI elements is not the actual starting creation. However, this is a complex issue with different interpretations across various jurisdictions grounded still in international harmonization of Copyright law, since most countries in the world adhere to the Berne Convention. What is interesting that contradicts a lot of Copyright law with international conventions (like Berne Convention and TRIPS to name a few) is a case in Beijing where a image created through AI actually got Copyright protection through the Beijing Internet Court."

Q2: How do copyright laws balance the rights of creators with the public good, and what are the key differences between the US fair use doctrine and international copyright frameworks like the TRIPS agreement? How do these frameworks apply to AI-generated content and practices such as text and data mining?

Jon Ellrose: "Copyright can’t exist without the purpose of public good. All jurisdictions about Copyright has this legal logic because there can be no artistery without a consuming public and no artistery can be created if there are no sources of funds - regardless of public or private sector funding since the Renaissance. So fair use - exceptions and limitations - has existed one way or the other since the inception of author´s rights when the publishers didn´t have monopoly rights, but the author started having more and more rights. With technological change the authorship concept really became the originator concept and both of these words are in the legislational frameworks of rightsholders. Copyright however is distinctly different in what it protects compared to author´s rights as it only protect the right for a limited time for mainly economic purposes to control copies, reproduction and derivative works and the likes. Author´s rights goes deeper and protect moral rights as well. This is very important to understand regarding natural and legal persons and their respective rights granted - and guaranteed - by states and the exceptions and limitations of those rights. The US Copyright system uses fair use doctrine (where the concept of transformative use could be especially important for certain AI models and systems) and the rest of the world mainly uses the three step test formulated in the Agreement on Trade-Related Aspects of Intellectual Property Rights (TRIPS) through the World Trade Organization (WTO). Both these perspective on public usage and societal development are also associated by law bodies of text and data mining (TDM), especially in the European Union through InfoSoc- and Copyright Digital Single Market directives in regards to what is premissable data analytics and temporary copying and other rules that guides the tangible medium of electronics, where creations are stored as copies during various time frames and continuity. The fair use doctrine does not rely on limitations on text and data mining, but if the use of data is deemed to be transformative in nature with a deeper analysis, then it is permissible. Harmonization about these two perspectives could be helped by both GDPR and the EU AI ACT, but that depends on jurisdiction and venues for lawsuit and the respective burden of proofs between the parties during litigation; between the plaintiffs and defendants.”

Q3: How do copyright laws balance the rights of authors and the public, particularly in the context of fair use and the three-step test? What is required for AI models to be considered legal and ethical regarding intellectual property?

Jon Ellrose: "The legitimate interest of any author as in the originator compared to the public is the spirit of the law with both fair use and the three step test, but the two methods approach these aspects with somewhat different analytics, but could also render the same conclusion of what is fair or permissable in ”special cases” to use copyrighted work without licenses. The legitimate interest of the public and consumer of originator creation is what these law bodies are all about; promote ”useful arts”. The interest of time as in generations to come on both sides of this dialectic relationship is something that is very important in the ad hoc fashion historically of Copyright law in international intellectual property law that has come about through innovation, piracy and technological development since the dawn of time (The USA pirated english books tremendously in the 19th century and Scotland in the 17th century pirated english books; the reason for enlightment coming about from Scottish thinkers and writers). This introductory background is of outmost importance to understand where we are at now trying to fit AI within the broader Copyright framework with respect for temporary and limited property rights within Copyright and various rightsholder perspectives and other author’s rights and at the same time respecting the consuming public and their interests. Can you have Copyright on your AI creations and what is required for an AI model to be legal in terms of intellectual property that natural and legal persons hold? We need to start of with an analysis of the concepts that determines the AI tools as legitimate tools for human authorship; legally sound and preferably ethical in nature while adhering to the foundational purpose of Copyright.

Q4: You mentioned the potential for data taxes and other economic measures to harmonize global markets. How do you envision these developments, particularly to AI-generated content?

Jon Ellrose: "Music was easier to manage because when recorded sound became available, collective licenses were forced upon us due to the inability to control every instance of use. That was the case in the US and for instance the transition from sheet music to self playing pianos. Technological disruptions created the need for just solutions. With AI, the initial training involves a bulk of data from the entire Internet, with varying degree of originality in data ingested by AI model. My thoughts are that we could consider a data tax that could support public arts programs or other contributions to the development of culture. However, the problem with copyright today is the increased anti-trust perspectives where certain entities control a lot of the distribution with rent seeking to various degrees, making it difficult for creators to get paid. AI could either erode culture if creators cannot sustain themselves, or it could enhance it if managed properly with the respect of the rule of law."

Q5: In your article, you referenced the ideas of Bentham, Mill and Locke with a discussion of their relevance to modern AI. How do these philosophical perspectives guide our understanding of AI today?

Jon Ellrose: "A utilitarian approach doesn't judge the quality of art or culture; rather, it focuses on what the consuming public wants. The idea of useful arts is what the consuming public finds attractive. AI should be seen as a tool, not a threat. It's a very advanced tool, but it's just that—a tool. Art is about the soul and society, something AI can not replicate easily. So AI touches on existential and religious concepts of consciousness and where art derives from. The real threat to artists isn't AI, but the market and royalty situation globally with decent incentives and compensation in the creator economy and how to raise capital."

Q6: In a recent case, the New York Daily News and Chicago Tribune sued Microsoft and OpenAI for misusing their reporters' work to train AI systems. Can you share any insights on similar cases?

Jon Ellrose: "An interesting case to note is the Viacom settlement with YouTube, which never actually tested whether YouTube is legal. This is relevant today as OpenAI and other models train on vast amounts of data, including YouTube videos. You have a notice and takedown system in YouTube as required by the Digital Millennium Copyright Act (DMCA) where Copyright holder either can request takedown or monetize content. The legal aspects of whether the data ingestion from YouTube in Open AI constitutes copyright infringement or unjust enrichment (or other FTC regulated aspects of unfair competiton and the likes) are still being debated. It's a complex issue because copyright is not an absolute property right; it is always balanced with public interest as in incentives to create useful arts with fair compensation, control, credit and consent regarding the rightsholder."

Q7: How should companies navigate the complexities of using AI models, especially in terms of ensuring they are not infringing on copyrights?

Jon Ellrose: "Companies should carefully review end-user license agreements and ask critical questions about how AI models were trained, especially concerning data origin. It's crucial to check whether the AI provider has proper licenses for the data used in training. Insurance companies will often not issue policies if these licenses are not in place, which can be a red flag. Additionally, companies should be wary of relying on transformative use arguments, as these can be legally precarious."

Q8: Looking to the future, what do you think is the most critical action developers, policymakers, and creators need to take to support innovation while protecting intellectual property rights?

Jon Ellrose: "It's essential to revisit the original purpose of intellectual property laws, which is to enable creators to monetize their work and contribute to society. The focus should be on ensuring a fair compensation system for creators rather than getting caught up in the fear of AI. We need to ask whether we have a fair market that encourages creativity or if monopolies are stifling innovation. AI is just a tool, and the real issue lies in the economics of how we reward creators."

Conclusion

As AI continues to evolve, so must our approach to copyright and other intellectual property evolve. Jon Ellrose's insights underscore the importance of balancing innovation with fairness, ensuring that creators are protected while embracing AI's possibilities. This interview is a valuable resource for anyone navigating the complex intersection of AI and copyright law.