-Harshal Chhabra* and Kanishk Gaurav Pandey**
Abstract
This post explores the complexities of copyright law in India in relation to AI-generated content, focusing on literary works. We address the challenge of determining copyright ownership when AI plays a significant role in content creation. We propose the concept of a "Significant Human Input" test. It seeks to strike a balance between fostering innovation and honoring human creativity. Through legal analysis and practical examples, we propose an approach that acknowledges the evolving landscape of creative processes while upholding the principles of originality and human authorship. By advocating for a framework that integrates AI-generated content within existing copyright standards, we offer insights into how India can adapt its copyright regime to accommodate the rise of AI-driven innovation.
Introduction
“In depths of soul, creativity blooms,
A tapestry woven with passion looms.
With words and brush, visions take flight,
Imagination's spark, a beacon of light.
In each creation, a glimpse of our worth,
Humanity's art, an endless rebirth.”
- Robert Frost
In the above excerpt from one of his classic poems, Robert Frost highlighted the importance of human creativity in art (or did he?). While the above lines may seem similar to Frost’s other popular works to a casual reader, his fans will tell you that no such poem exists – and the above is merely a ChatGPT output based on a one-line prompt.
The recent emergence of highly advanced chatbots has given rise to significant questions about the boundaries of copyright law. A question as simple as – who owns the copyright to the above poem? Is it ChatGPT because the output was written by it? Or is it OpenAI because their software generated this output? Is it us because we wrote the prompt that generated the output?
Indian copyright law is still in its nascent stages of familiarizing itself with the rise of generative AI such as Bard and ChatGPT. This blog post will clarify the answers to the above-posed questions under existing Indian laws. However, those answers will make it clear that there is no place for AI-generated literary works under the current copyright regime. While, the ideas we propose can be extrapolated to other non-literary forms of copyright, such as music, that inquiry is beyond the scope of this post.
Therefore, we will attempt to fit in copyright for generative AI for literary works without suggesting any radical changes and staying within the confines of current standards with slight alterations. We argue for a “Significant Human Input” test, which allows users of generative AI to get a copyright for AI-generated work in certain cases. It evaluates the extent of human input in the creation process against a “Significant Input” standard. The test has been inspired by the recent standard for granting AI-generated works developed in the USA, albeit with modifications to better suit the Indian Copyright Law. We are choosing to limit the scope of this test to literary works for the purposes of this paper.
In the upcoming section, we will summarize the current position of law in India by answering the questions asked earlier. In the third section, we will lay out the test and put forth an argument for allowing authors of AI-generated works to be granted copyright if their work meets the standard prescribed in our suggested test. Additionally, to justify the test, we also address two major counterarguments to grant copyright to AI-generated works, namely – Lack of Originality and absence of human intervention.
The Current position of law
In the introduction, we pointed out the potential candidates who could be granted the copyright for a piece generated by AI. In this section, we will try to answer that by analyzing the current position of law regarding AI-generated works in India.
ChatGPT as author?
ChatGPT cannot be considered the author as Indian copyright law requires the person claiming the copyright to be a natural person. Section 17 of the Indian Copyright Act states that only persons can be authors. Usually, the definition of ‘persons’ is restricted to individuals, but entities such as companies can be assigned copyright (Section 18) by the individual, based on an agreement, for a limited period. Section 17 also states that unless a contract to the contrary exists, the human individual will always retain the first copyright to the product. Additionally, the scheme of the act is clearly human-centric. For example, the application for copyright registration (Form-XIV) requires disclosure of the claimant's name, nationality and address. (This issue has been discussed in-depth here, here[Article number 4 of this issue]). Regardless, the consideration of whether authors can be natural persons (humans) only or entities is outside the scope of this paper. Our query here is to determine the copyrightability of products that have been produced by humans using generative AI.
Or, Is it the AI’s Developer?
The developers’ claim would largely depend on their ‘Terms-of-Use’ policy. If a developer mentions that they will retain the rights to work generated by their AI’s, they will have the copyright for it, unless a contract to the contrary already exists. However, the ‘Terms-of-use’ for most commonly used AIs such as ChatGPT/BingChat does not claim copyright over the outcomes generated. Even logically, it does not seem fair to grant developers the ability to assert copyright ownership over AI-generated products. An analogous situation can be seen in the creation of a painting using Microsoft Paint – it would not be reasonable for Microsoft to claim copyright over the content produced. Even though, similar to a chatbot, the services Paint provides is very substantial for the creation of the painting (filling of colours, shapes, etc). The resultant creative product is a fruit of the human’s individual application of mind. Similarly, OpenAI is merely providing a service, the resultant product is a result of how the human chose to use that service.
But, can the prompt giver be granted the copyright?
Here is where the law, and its interplay with generative-AI, gets more interesting. The Indian Copyright Act was amended in 1994 to account for situations where artistic works may be generated by a computer. The amendment introduced an additional provision, Section 2(d)(vi), to the act, explicitly stating that the authorship of computer-generated works belongs to the person who caused the creation of such works.
One may be tempted to think that this clarifies the position of law in India as the ‘prompt-giver’ is causing the work to be generated. However, entering a single-line prompt is not enough to grant copyright protection to a work. Protection under copyright law is granted to an author if their artistic work/product meets the requisite standards to constitute an ‘original’ product. The lowest standard for granting copyright protection to the Author (this standard is not applicable in India) is the ‘sweat of the brow’ doctrine which states that an author gains copyright rights based on simple diligence and effort put in by them. An important case for understanding it is University of London Press Ltd v. Tutorial Press Ltd, where a publisher made a compilation of question papers released by the University of London. The university challenged it on the grounds that it was infringing the copyright of professors who put in their skill and time to create those papers. In response, the publisher argued that the papers were created from an existing body of knowledge and thus were not “original” and could not be copyrighted. However, the court held that even if the paper was created from an existing body of knowledge, a certain amount of effort was exerted, and this is enough to qualify the papers for copyright protection. This standard of ‘originality’ grants copyright on the basis of diligence and effort and does not include a requirement for subsequent creativity. However, an output based on a simple prompt – no matter its distinctiveness from other work – is not copyrightable due to the lack of effort put in by human authors in writing a single-line prompt. Originality will be further explored in part A of the next section.
“Significant Input” Test For Determining Copyrightability
AI has swiftly become an integral tool across industries such as software, art, literature, and various other domains. Given its ever-increasing presence in shaping our work processes and determining end products, revisiting existing laws has become a pressing necessity due to the importance of copyright in incentivizing creativity and innovation. AI is bound to be everywhere soon, in the software for the technology we use, in the books we read, and even in the art we see and hear.
Recognizing the permanence of AI, it is imperative to rethink the current position to encompass AI-generated works. AI is poised to catalyze significant developments and disregarding them could be disastrous for development. It would be rather ignorant for a country like India, which is aiming to present itself as a commerce and innovation hub to the world. However, while reconsidering the current position, it is equally important to ensure that human effort and involvement continue to hold value in creative processes.
To strike a balance between the competing principles of safeguarding human involvement in creative works and recognizing the copyrightability of AI-generated content, we propose the “Significant Human Input” test. It is a simple test that seeks to check the extent of human involvement in the creation of an “original” product. To clarify, we are not suggesting a completely new framework to assess copyrightability, rather this test operates within the existing framework of the “Skill and Judgement” but with an additional query attached which evaluates whether the “original” product in question would exist in the same tangible form without human intervention (these standards will be explained in the upcoming section). If the product in question meets the standard of “Significant Input” set in the test, it can be granted copyright protection. The test has two components – a) Determining whether the AI-generated product is “original” and b) whether the extent of human involvement in the process is significant enough.
The rest of the section will lay out the test in further depth. We will, first, analyse the standards for originality under copyright law and; second, explain in further detail the understanding of the ‘significant’ part of the ‘significant input test’. We will also justify its suitability by countering two major arguments- lack of originality and human input, made against granting copyright protection to AI-generated products.
A. Determining “Originality”
One major argument against protecting anything created by generative AI is the lack of originality. These chatbots are not capable of independent thought and thus, there is no novelty in any output. The output we receive is a combination of pre-existing material found over the internet, thus, it is contended that AI-generated works cannot be copyrighted. However, that contention does not hold true if we analyse the standard of “originality” required in India for claiming copyright protection.
The current standard for originality in India was established in Eastern Book Company v. D.B. Modak. In this case, SCC, a journal run by the EBC since 1969, published all Supreme Court cases with a few additions such as paragraph numbers, headnotes, formatting and cross-references. In 2004, the respondents launched software which contained copies of judgements lifted directly from SCC. Eastern book company filed a copyright violation suit for copying “original” elements. In response, Respondents argued that since government documents are not copyrightable, that will apply to judicial decisions as well and thus, SCC held no copyright over them.
The Supreme Court considered different standards of “originality” to determine the standard for India. It felt the ‘Sweat of the Brow’ was too low a threshold for giving copyright protection and shifted the balance of copyright protection too far in the favour of the owner’s right. It also looked at the American standard of ‘modicum of creativity’ which posits a requirement for a certain level of creativity to make the product copyrightable. The court decided that it was too high a standard. It finally settled on the Canadian test which says that the work in question must be a product of the author’s skill and judgement. The exercise of skill and judgement should also not be so trivial that it can be purely exercised as a mechanical exercise. This also re-asserts the argument for why an output generated by a single line prompt cannot be granted protection. Ultimately, it was held that the human skill and judgement involved in creating the additional elements involved legal knowledge, skill and the author’s judgement and thus the SCC version of judgements is copyrightable.
Novelty is considered too high a standard for copyright, and there is no reason why a different standard should be set for AI-generated works. Therefore, AI-generated works can meet the requisite threshold of “originality” in India irrespective of whether they were derived from pre-existing bodies of knowledge.
B. Significant Input and Human authorship
It is also important to remember that the copyright act places a requirement for a Human author. Usually, AI-generated work involves very little human involvement as much of the labour is done by the AI itself. Most of the requirements posited in the standard created in D.B. Modak are met by the AI, and not the human. In Navigators Logistics Ltd. v. Kashif Qureshi, a copyright claim made on a list compiled by a computer was rejected due to a lack of human intervention. It was held that human involvement in the creation process is essential for the grant of copyright protection in India.
The “Significant Input” test establishes two essential requirements for determining whether an author, who utilized AI assistance in creating a product, can claim copyright or not. The first is an objective criterion where it should be checked whether there was human involvement at all in the creation process. Second, the extent of that human involvement should be considered. The ‘extent’ of human skill, judgment, and labour invested in its creation must be significant enough that the product would be fundamentally different or non-existent without it.
An example will make the above position clearer. Let’s consider an example in the field of legal academia. Legal commentaries provide a detailed, comprehensive and contemporary overview of the law. Writing a legal commentary is a tedious process. It involves reading almost every important case, analysing the statute, as well as the academic debates related to the topic. By using AI tools to summarize judgments and extract vital points from scholarly works, the author can streamline their research process. This allows authors to focus more on critical analysis and interpretation of the law, improving the overall efficiency and quality of their work. Now, let’s introduce an additional layer of complexity to this scenario. The author in question is a non-native English speaker who has completed the initial draft and wishes to improve the language and grammar of certain sections of the text using ChatGPT. It is important to note that ChatGPT can only assist with form, structure, and grammar but not make any substantive improvements. The author heavily relies on ChatGPT to enhance the work, and without this generative AI assistance, the commentary would not have taken its current form.
The question which arises now is whether the author can claim copyright over the commentary. According to the test, the answer is yes. Although AI assistance was utilized in the process, the author fulfils the twin requirements of the test. Firstly, there was human input in the creation process as the author conducted research, found relevant judgments, and composed a draft. Secondly, the extent of human input, in terms of labour, skill, and cognitive ability exercised was significant enough that the commentary would have either been fundamentally different or non-existent without it.
C. The Zarya Standard: Similar but not the same
The USA recently recognised a similar right following the Zarya of the dawn (‘Zarya’) case. In this case, the US copyright office recognized protection for a graphic novel, even though all the images were generated by AI. The Copyright Office’s decision was based on the assessment of the skill, judgment, and creativity exercised by the author in arranging, selecting, and editing of the AI-generated images, which ultimately transformed them into a tangible form—a graphic novel. Following this, the USA copyright office issued a formal policy statement which said that a work containing AI-generated material may be granted copyright if it contains sufficient human authorship to meet the standard required for copyright protection.
Despite similarities with the ‘Significant-Input’ test, there are a few differences between our proposed standard and that devised by the US copyright office.
First, the ‘Significant Input’ test is a general standard as it checks the extent of human involvement in the creation process and such a standard would be applicable in many copyrightable products such as music, books, software etc. (even though this paper specifically analysed it in context of literary works). Such a broad application of the Zarya standard might not be possible because it attributes originality on the basis of specific acts undertaken by the author (usage of specific words like ‘arrangement’ and ‘selecting’) in the creation process of a graphic novel specifically. The significant input test applies more broadly and as a general standard. It does not prescribe specific overt acts that the human author must undertake to claim copyright. Instead, it focuses on evaluating the overall level of human creativity, judgment, and effort exerted in the creative process. It acknowledges that originality can arise from various forms of human input, not limited to specific actions like arrangement or selection.
While the acts of selection or arrangement may be crucial for establishing originality in the context of a graphic novel, they may not be directly applicable or relevant to other types of creative works where human involvement operates differently. An author might not have to do specific overt acts such as arranging or selecting specific AI-generated parts to claim copyright. For instance, as highlighted in the legal commentary example in the preceding section, let’s consider how the author’s process diverges from the specific acts emphasized by the Zarya standard. The author’s initial involvement in the research process, which includes reading cases, analyzing statutes, and engaging with academic debates, is where the substantial human input lies. This phase of the process involves the exercise of skill, judgment, and creativity, essential elements for copyright protection.
However, when the author utilizes AI tools to streamline the research process, the subsequent actions may not align with the specific acts highlighted by the Zarya standard, such as arrangement and selection of AI-generated content. Instead, the author might simply input the gathered information into a Generative AI tool, requesting corrections for language, grammar, and style without explicitly engaging in the arrangement or selection of specific content.
This distinction is crucial because, under the Zarya standard, the emphasis is placed on the author’s specific actions in the creation process. On the other hand, in the legal commentary, the author’s primary contribution lies in the initial research and engagement with the subject matter rather than specific acts of arrangement or selection.
Therefore, while the author may not have undertaken the overt acts prescribed by the Zarya standard, their substantial involvement in the research process prior to utilizing AI tools is sufficient to meet the originality test.
Second, the proposed test does not intend to import the USA’s ‘modicum of creativity’ standard for determining the originality of AI-generated works, as Indian courts consider it too high a standard for copyright law. In the D.B. Modak case, the court also clarified its reason for adopting the Canadian standard by claiming that the USA’s ‘modicum of creativity’ standard implies that the product in question must be ‘Novel’ or ‘Non-Obvious’ – both concepts which are associated with patent law rather than copyright law (See generally for understanding the difference between patent and copyright). It also claimed that the current standard borrowed is more closely associated with the objectives of the Indian Copyright Act. Thus, the test retains the ‘Skill and Judgement’ standard while determining the copyrightability of AI-generated products by asking a further question adequate for the new era of AI-generated works. Therefore, due to its wider applicability and proximity to the objectives of the copyright act, our proposed test is better suited to help Indian copyright law transition into an era where it might have to grant copyright protection to AI-generated products.
Conclusion
Generative AI, in a matter of months, has entrenched itself into our lives to an unimaginable level. With new developments and big-tech players such as Google and Amazon entering the race, it is bound to grow more significantly. India’s current position of not granting copyright protection to any AI-generated material will not be feasible anymore as much of the innovation and creation from now on will take AI assistance to make tasks easier.
In this blog, we aimed to further discussions on this issue by presenting a solution that maintains the current requirements for human authorship and originality while also considering that AI is now an intrinsic part of most innovation. The blog reasoned that AI-generated material may be given copyright protection if the human involvement in the creation process is significant enough. It also explained how AI-generated works may meet the originality requirement under copyright law despite being derived from pre-existing work. The position of the decision-makers on this matter remains unknown, but there is a hopeful expectation that India will embrace a position that allows users to claim copyright protection for AI-generated works. This pivotal decision could have far-reaching consequences in defining India's international reputation as a leading innovation hub in the decade to come.
*Harshal Chhabra is a second-year student pursuing his B.A., LL.B. (Hons.) from Gujarat National Law University. He is an Intellectual Property Rights enthusiast and is deeply immersed in policy research, particularly exploring the dynamic intersection between law and technology.
**Kanishk Gaurav Pandey is a second-year student pursuing his B.A., LL.B. (Hons.) from National Law School of India University, Bangalore.