Beyond Human Creativity Rethinking Intellectual Property in The Age Of Artificial Intelligence

Introduction : The essay explores the myriad legal conundrums arising from the gradual adoption of Artificial Intelligence in public, professional, and creative spheres to conceive of various categories of Intellectual property. The essay re-iterates quintessential intellectual quandaries regarding ownership rights in Generative AI’s training models and algorithms, and whether works created using AI would be eligible under cyber law and international law to gain copyright protection. The essay highlights the various concerns regarding the law where AI and Intellectual property interact, including jurisdictional concerns since this technical confluence falls in multiple legal geographies, whether there must be a uniform set of copyright laws for cyber-assisted intellectual property, and what that must be. Furthermore, the author conceptualizes certain legal paradigms the legislature and judicial benches can pragmatically approach so that the law can fairly catch up to the exponential advancement of the Metaverse and AI and provide for some good safeguards against unfair use, plagiarism and stealing of intellectual property in a post-Meta legal ecosystem.

The Emergence of Artificial Intelligence as a Challenge to Traditional Copyright Frameworks

Intellectual Property refers to any intangible, physically immaterial, indivisible, and innumerably consumable property that is conceived using mental, creative, and intellectual faculties. They are legislated differently from traditional property like land, money, and material possessions because of these marked distinctions and characteristics. Internationally recognized codified law governing intellectual property has also been fairly recent, with the Paris Convention for the Protection of Industrial Property (PCPIP) being one of the very first attempts at defining and ratifying cross-jurisdictional legal standards for intellectual property rights in 1883. It was succeeded by the Berne Convention for the Protection of Literary and Artistic Works in 1886, then only adopted at ratified by 10 signatories because of the lack of legal awareness and discourse on intellectual property rights.

With over 110 members a part of the Berne Convention as of 2024, it is notable that it serves as one of the most important pieces of international legislation governing intellectual property and copyright laws. Membership into the Convention and non-binding adherence to the statutory mandates listed in the treaty are to be considered for membership in the World Trade Organization. This way the Berne Convention has been fairly successful in formulating and devising a standardized legislation of copyright law that member nations must incorporate into their legal systems to ensure uniformity in the law governing intellectual property disputes regardless of what the jurisdiction of the parties in question might be. Not only does standardized legislation ensure swift dispute resolution of intellectual property-related cases despite the specific geographies of the litigants, but it also sets certain expectations of the myriad aspects of the law that must be covered in their entirety when a member nation is devising their acts, statutes, and bills governing copyright law. All member nations of the Convention loosely adhere to the statutes laid down in the treaty within their legal systems governing copyright law.

Copyright law has an extensive colonial legacy in India. Pre-1947, India had three distinct sets of copyright legislation to keep up with consequential technological advancements that were plaguing the rapidly industrializing world of the 19^th century. The very first piece of legislation governing copyright law in India was the India Copyright Act of 1847, the next two almost a century later, the Copyright Act of 1911 and the Copyright Act of 1914. Post-colonial India passed another legislation on Copyright Law, namely the Copyright Act of 1957, which alongside several amendments to the original act continues to be in force today. 

Legal academicians historically recognized the gaping vacuum in the law for the lack of recourses an owner could take in cases where their intellectually devised property has been infringed upon, appropriated without their consent for economic incentives or just stolen. The extensive legislation separate from that of traditional property was emblematic of the ways academics perceived intellectual property as a distinct but expansive field in and of itself. Because of the immateriality and intangibility of intellectual property, any technological advancements threatening the scope and applicability of copyright protection from unauthorized and unfair use were inspected and the legislation was re-evaluated in a new technical context and how that would affect the production and dissemination of protected creative and scientific works under fair use.

The law aimed at fostering innovation and widespread dissemination of IP but wanted to discourage plagiarism and appropriation of property. Emerging tech constantly forced the law to reevaluate the boundaries of what would differentiate ‘inspiration’ from ‘stealing’, of what would constitute ‘original’, and with the advent of AI, what would constitute ‘art’ in the context of AI art and Generative AI being employed extensively to create and consume art. The exact answer is debatable among legal academicians but the various aspects of AI on Intellectual Property and Copyright are fascinating to explore through.

Copyright Implications of AI Training Models and Datasets

With the rapid advancement of Artificial Intelligence comes certain intellectual and philosophical quandaries, some of which have to be dissected to understand the essence and crux of the problems AI poses to the law. In 2023, Open AI, a US-based artificial intelligence tech company faced several class action lawsuits alleging misappropriation of the ‘fair use’ clause of the US’s Copyright Act.  The petitioners alleged that Open AI had used both licensed and unlicensed copyright-protected works to train their AI models, something that would squarely fall under ‘copyright infringement’ as per US law. Subsequently, the petitioners demand newer interpretations of Section 107 of the Copyright Act, which was identified as the vacuum and lapse in the law that would lead these AI companies to get away with theft.

There were discourses in both academia and the media surrounding AI art, or art created using generative AI models, and the various ways this could infringe upon an artist’s right to own their work. AI models like Open AI, Midjourney, and Chat GPT use text-based prompts to conceive visual art. The AI doesn’t create this art originally but uses thousands of images and copyrighted artwork available on the internet to train its model and algorithm for the AI art’s very creation. AI companies were hit with lawsuits when artists started seeing similarities between works conceived by AI models and original human art created by some of these artists. Legal academics grapple with the question of how the law must distinguish ‘human art’ from ‘AI art’. What is the amount of AI involvement needed in the creation of an artwork for human art ‘to become’ AI art? What is human art? Is it art made without AI assistance or will some AI assistance be deemed okay to create an artwork to constitute human art? Should only human art be protected by the law? Or should AI art be conceptualized as a new kind of art that is part of an ever-expansive, interpretative, and expansive philosophical definition of art?

The debate was highlighted in the media in August of 2022 when Jason Allen, an AI artist who created his artwork using Midjourney, won the First Prize for his AI artwork at the Colorado State Fair after defeating human artists in the competition. One year later, Midjourney would be sued by artists for copyright infringement and appropriation. 

Intellectual property that is created by AI now has steadily encompassed all types of art, from textual to audial. Generative AI models now produce works of literature, music, videos, and even movies in a single prompt. As authors, musicians, video and movie makers feel they would be put out of work as AI will replace them, domestic and municipal laws find it increasingly hard to catch up to the rapid and exponential development in the quality of art developed by AI, as even in its nascent phases AI art is virtually indistinguishable from human art, even to AI experts. Artists contend that the AI models have illegally trained themselves from their works, only to go on to fundamentally replace them. 

Determining Liability for AI-Related Intellectual Property Infringement

AI infringement legislation is fairly new to the law, and most landmark suits concerning the intersections between intellectual property, copyright law, and AI have remained pending. However, newer interpretations of existing copyright laws in the US and India have been made to incorporate AI into the copyright provisions. In Anderson v Stability AI, the petitioners who were visual artists brought a class action lawsuit against Midjourney, Deviant Art, and Stability AI for allegedly infringing upon their copyrighted artworks to train their AI models without any prior consent or knowledge of the artists themselves. The petitioners argued that just because their visual art existed as algorithmic representations in the input stage before the AI ingested and processed their artwork to create new artworks, that cannot be employed as a basis to dismiss claims of direct infringement. Past case pronouncements dealing with Section 107 of the US Copyright Act cannot be used to give injunctions on cases involving AI since the technology is something the law has never dealt with before.

Unlike past technological advancements that were hit with similar lawsuits, like VCR and xerography, AI cannot be let off because it involves the processing of actual data before producing new work whereas with VCR technologies the work was made by the consumers of the work themselves, not by VCR companies before reaching the consumers. It was a notable contention in the case that the reason why these cases don’t succeed is because of the myriad ways different AI models ingest different types of information, all having different algorithmic and mathematical structures, and all ingesting different kinds of copyright material at the same time. In the case of Getty Images v Stability AI, the issue of altering, removing, and falsifying copyright management information was evaluated. As per the Digital Millenium Copyright Act, it is illegal to falsify information that could potentially mislead consumers about copyright and fair use authorizations of a piece of work, for example using watermarks on someone else’s work, or in an antithetical scenario, using someone else’s watermarks on one’s original work.

Getty Images sued the AI company in 2022 for not only violating copyright infringement rules by using Getty Images without the company’s consent to train Stability Diffusion’s AI model but also for misappropriating copyright information by using Getty’s watermarks on unflattering AI artworks that could potentially lead to misrepresentation regarding Getty’s commitment to quality in the images they produce.

So much of the discourse surrounding AI has to do with the apparent similarities between the output produced and the input that was ingested and subsequently processed by that same system to produce that very output. There are a myriad of challenges when dealing with this, first off the law is unsure about the liability of AI infringement cases. Is the company to be held squarely responsible for the copyright infringement, or does the prompter or the AI developer to some extent share the load of liability? Where does infringement of work start and where does it end? Is deriving someone’s work and interpreting it to create new works that hold the essence and crux of the original work to be taken as an act of plagiarism? Under US law, similar reproductions of intellectual property violate the rightful owner’s right to publicity, an entitlement vested to a person to be able to own the essence and crux of his work and image. Interpreting this legal provision, AI companies could find themselves in more trouble given the reproductions of outputs, be they audiovisual or textual, often hold in the essence and crux of the works of the original authors in question, for example by reproducing art in the art style reminiscent of a creator (their most notable representation to the world), therefore violating their right to take control of their public image. There is a need for greater transparency in the tech industry regarding the source materials that AI models use to train themselves. If there is greater public awareness about whose work goes in to be ingested into these models, it will adhere to the open-source licensing terms in relevant statutes. 

In Tremblay v Open AI, the petitioners who were world-renowned authors, sued Open AI for allegedly using 300,000 books to train its AI model without the authors' consent. The contention was around reproductions of copyrighted material and whether that would amount to fair use. Similarly, in New York Times v Open AI, the newspaper contended that the AI company had used articles and writeups from the paper as inputs to the AI’s algorithmic input to train the AI’s response to prompts asking for text-based outputs. More suits concerning the usage of copyrighted and protected textual material from literature to newspapers are coming to federal US courts. A landmark injunction deciding the interpretative clauses of the Digital Millenium Copyright Act is yet to remain, but questions regarding whether AI companies will have to unexpectedly be licensed to be able to use copyrighted material is an ongoing legal debate. The petitioners demand that their consent must first be used before their work is used to train these models for AI algorithms, and that appropriate recognition and compensation be given to the original works that have been used to extensively train these models in the first place. The petitioners also demand more transparency and lucidity in AI companies' methodology of extracting the source material required in training these models, with the explicit consent of the authors and a licensed authorization recognized under the law to be able to use protected copyrighted material under fair use.  

There have been certain statutory roadblocks that plaintiffs have found difficult to prove to be able to bring a legitimate claim against AI developers, their co-partners, their parent companies, and the corporate entities who invest in them. Under US copyright law, a plaintiff has been often forced to limit their claims to unfair competition and false advertising because of statutory limitations present in the Digital Millenium Copyright Act. If a claimant’s intellectual property is not already registered by the law, he cannot sue the defendant based on copyright infringement of their work. Similarly, Courts have interpreted and rejected the claims of various plaintiffs since conventional understandings of copyright law depend on exact reproductions of the original works. Many cases in US courts have been dismissed on the pretext and rationale that AI outputs are rarely regurgitations of the original work, so that cannot possibly constitute copyright infringement.

With AI developers catching onto this legal loophole, there have been incentives in the tech industry to be able to train the models to such an extent that no exact reproduction of an original piece of artwork can be produced by the AI. In cases of addition and subtraction of watermarks, as listed under copyright management provisions of the DMCA, it has been increasingly difficult for the plaintiffs to prove that their work has indeed been infringed upon, violated of fair use, or misappropriated since it doesn’t matter whether the watermark is present or absent, the output of an AI prompt is rarely, if ever, exactly similar to the original copies of inputs fed to the model to ingest, and therefore getting a claim against AI companies remains to be difficult.

Landmark judgments on AI models employing copyrighted material to train themselves and their datasets are yet to be seen. However, in cases like Thomas Reuters v Ross Intelligence, the doctrine of fair use in copyright provisions and statutes will be relevant in how courts interpret them. Under US law, copyrighted material can be used without the consent of the original authors in question if the use is regarding an ‘educational and transformative’ purpose rather than a strictly commercial one. The plaintiffs have argued that using West Law’s written material to train Ross Intelligence’s AI model has been purely for commercial reasons, fostering unfair competition antithetical to provisions of antitrust legislation.

The defendants have claimed that the usage of copyrighted material to train these models has strictly amounted to fair use since these source materials have been used to ‘teach’ the AI models to produce better and more transformative outputs to text prompts. To reduce the litigation costs of bearing class action lawsuits, many AI companies might start considering expansive licensing schemes to be able to comprehensively pay and compensate authors and owners of copyrighted material after using their work in their models. The law has to be particularly circumspect while devising new laws, revising and re-interpreting old ones to reduce the hostilities between creators and Generative AI developers, and to allow wide dissemination of intellectual property that can be consumed by many people at the same time through the internet. The law has to also carefully tread on the new technical landscape and how it changes the law’s interpretation of existing copyright provisions to not stall AI innovation and growth of AI in the market economy. How law aims to achieve these balancing acts is yet to be seen, and the contention remains.

In a landmark case of its very first kind in India that is yet to be decided, ANI v Open AI, several questions have been raised as they fall into the aforementioned Copyright Act 1957. It to be seen how the Court interprets Section 57 of CA 1957, which talks about the criteria amounting to infringement of copyrighted material and what liabilities might be posed if the defendants are found to be guilty as per Section 63 of CA 1957 which imposes liabilities for the act of infringement. Not only these various interpretations are to be adjudged, but the fair use as instated in the Copyright Act would be applied to the scenario. Open AI argues that the usage of ANI’s source materials and writeups to train their model has been for ‘transformative’ purposes as the output prompt is fundamentally dissimilar to what was fed into the system before it was ingested and processed. The Court might have to contend with certain jurisdictional and trans-legal problems as the parties to the case are situated in two wholly different legal geographies. 

Final Conclusion

The essay decisively demonstrated the myriad intersections and web of issues that AI and IP law are intrinsically enmeshed. There are a lot of contentions to be addressed, from inter-jurisdictional and transnational IP disputes and infringement lawsuits to interpretations of what constitutes fair use under US and Indian law governing copyright. With more cases lining up in federal courts across municipal legal entities, there will be more statutory interpretations to conventional understandings of what constitutes copyright infringement. Furthermore, a standardized mechanism can be instituted to squarely position legal liability in the hands of certain parties in specific cases. Such judicial pronouncements and precedents will serve as comprehensive sources of law determining resources to intellectual property disputes across legal geographies.

Author: Harshita Shukla in case of any queries please contact/write back to us via email to content@khuranaandkhurana.com or at  Khurana & Khurana, Advocates and IP Attorney

References

1. The Copyright Act, 1957

2. Copyright Rules, 2013

3. Copyright Act of 1976

4. Digital Millennium Copyright Act

5. Berne Convention for the Protection of Literary and Artistic Works, 1886.

6. Paris Convention for the Protection of Industrial Property, 1883.

7. Agreement on Trade-Related Aspects of Intellectual Property Rights (TRIPS), 1994.

8. WIPO Copyright Treaty (WCT), 1996.WIPO Performances and Phonograms Treaty (WPPT), 1996.