I. ABSTRACT
Copyright laws around the world are responding to the surge of generative artificial intelligence. While there has been significant discussion on whether an AI can be an “author”, there has been little focus on whether AI-assisted work can be copyrighted. In light of this evolving landscape, this paper specifically focuses on the copyrightability of works having both human and AI input. It comparatively analyses the standards for the copyrightability of these works in the US and India. This paper lays down the US standard for the copyrightability of Human works and argues that it is doctrinally inconsistent with creativity and expression. It builds on this analysis to argue that Indian copyright law must diverge from US copyright standards and instead adopt standards that are aligned with Indian copyright’s roots in creative and original expression.
II. INTRODUCTION
IP laws around the world have been shaken by the rise of Generative Artificial Intelligence (‘AI’).[ii] Particularly, they find it difficult to accommodate the idea that a machine can compile original work and show ‘creativity.’[iii] One of the contentious points in this issue is the copyrightability of works having both human and AI input (‘Human-AI work’). When a human uses AI to create a work, should this work be eligible for copyright? To what extent or in what manner should this AI be used? Are all AI-assisted works eligible for copyright, or only those that have significant human involvement? While India has not come up with answers, the U.S. Copyright Office (‘USCO’) seems to have come up with some of them.[iv] In this paper, I comparatively analyse the US and Indian IP regimes on the usage of AI in copyrightable works.
Firstly, I explore the question of whether an AI can be an ‘author’ of a copyrightable work. In outlining the differences between US and Indian law, I conclude that AI cannot be an ‘author’ in Indian law, but it can be used to provide inputs in human-involved copyrightable work. But in what circumstances can humans claim copyright over works with AI input? Secondly, to answer this question, I analyse the standards adopted by the US to allow humans to claim copyright over such human-AI work, as well as the policies and doctrines underlying its standards. Here, I show that the US takes the focus of copyright away from “creativity” and back to the “sweat of the brow” doctrine. Finally, I argue that Indian law must diverge from the US standards, given the underlying policy objectives of its copyright regime. In doing so, I conclude that Indian law must build upon its “originality” and “creativity” standards to allow AI to facilitate original expression instead of stifling it.[v]
III. AI AUTHORS? THE INDIAN POSITION
There is no doubt that AI can create all sorts of works; be it artistic, literary, or dramatic.[vi] But can it legally and independently be an author of these works? This question is relevant to understand to whom and in what manner copyright is vested. To examine this, the US example is particularly instructive since it was one of the first IP regimes to lay down guidelines for the copyrightability of human-AI work. Thus, this section looks at the Indian and US regimes to explore whether AI can be an ‘author’ of copyrightable work.
In the US, an author is generally defined as a “party” who “creates a work”, which involves “translating an idea into a work that is eligible for copyright protection.”[vii] While AI can create such a work, is it a “party”? Recently, US courts have answered this by excluding non-humans from the definition of an author. This began with the intriguing “monkey-selfie” case, where the Ninth Circuit Court held that a monkey who photographed himself was not entitled to the copyright in the photo.[viii] This position has since been solidified. In Thaler, the Columbia District Court rejected the copyrightability of an image created by an AI image generator.[ix] The court held that “human creativity is... at the core of copyrightability” and thus authorship cannot be extended “absent any guiding human hand.”[x]
The Court’s rationale is that copyright serves to promote human creativity. On this basis, it concludes that non-human creators cannot be “authors.” But does this necessarily follow? What about situations where non-human creators (such as AI) serve to facilitate human creativity? If human creativity is really at the core of copyrightability, then the Court should have no issue with AI tools that are used in creative processes. However, what emerges from the Court’s holding is that US law would still resist the classification of AI as an “author”. Instead, AI would likely be considered a “tool”; just as humans use cameras as tools to create copyrightable photos.[xi] Thus, the takeaway from the US position is that AI may not be an author on its own but merely a tool to facilitate human creativity.
This position seems to be reflected in Indian law. Section 2(d) of the Indian Copyright Act defines an author as a “person” who “causes” the copyrightable work to be generated.[xii] But is AI a “person”? While no Indian Court has ruled directly on the personhood of AI, the Delhi High Court has repeatedly held that juristic persons are incapable of being authors of copyrightable works.[xiii] Applying this logic, AI would fall outside the ambit of an “author” in the Indian Copyright Act. This is also clear from the scheme of the Act, which contemplates the “death” and “citizenship” of an author,[xiv] concepts which make little sense for machines and programmes. Even the existence of moral rights confirms that the Copyright Act does not contemplate non-human authors.[xv]
However, if we pick up on the US position, AI could be considered a facilitative tool for human authorship. The Copyright Act allows human authors to copyright “computer-generated” work, which includes any device with “information processing capabilities.”[xvi] Seen this way, a human could use the information processing abilities of AI to “cause” the creation of copyrightable work.
But in such AI-generated work, what does it mean for a human to “cause” the creation of copyrightable work?[xvii] Taken too far, this could mean that any AI output in response to a human prompt would be copyrightable. On the other extreme, this could mean that any AI involvement would make the work uncopyrightable. Indeed, even a “typewriter ‘facilitates’ the creation of a manuscript; but this does not make the manuscript uncopyrightable.”[xviii] The ultimate question then is - what do we mean by ‘facilitate’? In other words, where should we draw the line between the human contribution and the AI’s?[xix] In the next section, I explore this line-drawing question by examining the US minimum standards of human input in AI-assisted copyrightable work.
I. The US’ de-minimis standard: stifling creativity?
To determine the copyrightability of human-AI work, the US focuses on the nature and extent of human input involved. In March 2023, the US Copyright Office issued guidelines for the registration of works that contained AI input. According to the guidelines, any work consisting of appreciable AI contributions is not copyrightable. Works with only de minimis AI contributions are copyrightable.[xx] The USCO clarified these standards by providing illustrations: “appreciable” contributions are those where AI generates text, translations, images and even background elements.[xxi] “De minimis” contributions are those where AI is used for names/headings, checking grammar or sharpening images.[xxii] Thus, the threshold for AI involvement is quite low; if the work has even slightly expressive elements generated by AI, it will likely not be copyrightable. In other words, the standard is designed to ensure that all, if not most, expression is done by humans themselves.
However, there seems to be a doctrinal tension between this standard of creativity in human-AI copyrightable work and the general standard of creativity in copyrightable work in the US. In its landmark case on the standard of ‘originality’ for a copyrightable work, the US Supreme Court held that a copyrightable work must demonstrate a “modicum of creativity.”[xxiii] This means that the work must involve more than a mechanical application of mind.[xxiv] The appreciable/de-minimis standard misses a crucial point: Is it not possible for Human work with an appreciable amount of AI input to also have human input with a “modicum of creativity”?
This logic was appreciated in the Spring Breeze case in China, where a Beijing court allowed an entirely AI-generated image to receive copyright.[xxv] This is because the human prompts that created the image were a “significant creative and intellectual achievement”,[xxvi] which was “more than merely mechanical”.[xxvii] Indeed, the author used a selective set of 30-50 words to create the image of a “beautiful woman for a love poem”. He then refined the image in seven different steps, using a variety of different prompts to modify the image’s colour, background and the woman’s pose in each iteration.[xxviii] If USCO’s standard is applied to this case, the image will not receive protection given the “appreciable” amount of AI input in the image; after all, the work was generated by the AI. However, few would disagree that the human prompts of selecting words and refining each iteration showed at least a “modicum of creativity.”[xxix]
Rather than creativity, the policy underlying the de minimis standard seems to hark back to the old English “sweat of the brow doctrine”,[xxx] that copyright vests in those who expend labour, material and effort on the final product.[xxxi] Since most of the heavy lifting in creating the work is done by the AI, it follows from this doctrine that the human (despite his prompts) should not be allowed any copyright in the work. However, this approach to copyright has been explicitly rejected in US law.[xxxii] The Supreme Court has held that the purpose of copyright law is not to protect the “sweat of the brow”, but to promote the constitutionally enshrined goal of “the progress of science and arts.”[xxxiii] Arguably, the use of AI provides impetus to creative expression. Digital art can now take on new forms of expression, with artists bringing their creative ideas to life using AI.[xxxiv] However, USCO’s handling of AI’s copyright challenges indicates that it has shifted course away from the creativity track. At least for Human-AI work, copyright has become about labour instead of expression.
What lessons can the Indian copyright regime draw from this? In the next section, I examine the underlying policy objectives of Indian copyright and explore how India’s copyright laws can accommodate Human-AI work within these objectives.
II. Human-AI Work in India: Going Back to Copyright’s Roots
In 2021, Rajya Sabha’s Parliamentary Standing Committee examined India’s IP regime.[xxxv] After detailing AI’s potential impact on IP,[xxxvi] it observed that there is a “need to review” the Copyright Act on “a priority basis.”[xxxvii] However, Indian courts have not yet directly faced the issue of copyright in AI.[xxxviii] When an Indian court is eventually faced with the question of the copyrightability of Human-AI work, should it even allow copyright? If so, using what standards?
As we have explored, it is unlikely that an AI is recognised as an ‘independent’ author on its own. Not only does this have no basis in the legislation, but it also opens a Pandora’s Box. Who would be responsible for cases where the AI infringes others’ copyright? Would it be the AI developer or the human who entered the prompt?[xxxix] What happens when the AI refuses to be an author? For instance, OpenAI and Midjourney’s terms explicitly grant all ownership in the AI output to the human who entered the prompt.[xl] Thus, a viable approach is to examine whether Human-AI is copyrightable with the human as the author.
But what kinds of AI input are acceptable? For this, we must turn to a fundamental question: what does our copyright intend to serve? If our answer is the “protection of labour and effort”, then the USCO’s de minimis standard is the route to take. However, India’s copyright underpinnings are different. Copyright in India seeks to foster creativity, innovativeness, and original expression. In the landmark EBC case, the Indian Supreme Court reminds us that “the sweat of the brow doctrine flout(s) basic copyright principles” since “the Copyright Act is concerned ... with the expression of thought....having a flavour of creativity.”[xli] If protecting creativity and expression is the end goal, then there is no reason to prohibit works with appreciable AI input as long as the human prompts show sufficient skill, judgement and a “flavour” of creativity.[xlii]
The Indian Copyright Office initially endorsed this approach when it granted copyright registration to an AI-generated image that stylized a human’s original picture like Van Gogh’s Starry Night portrait.[xliii] Arguably, the final image was a result of substantial creative human input since the author not only clicked the photo, which formed the base of the image but also ideated the “starry night” stylization. However, the Indian Copyright Office subsequently issued a withdrawal notice without providing any reasons.[xliv] Similarly, the USCO also rejected copyright for the same image using the de minimis standard.[xlv] The fact that the USCO and Indian Copyright Office’s approach align on this issue tells us that they both have deviated from the ‘creative and original expression’ roots of copyright.
In any case, India’s copyright regime’s encounter with AI is still fresh. If the goal is the protection of creative input, then copyright offices and courts should focus on examining whether the human input in the AI work demonstrates the standard of creativity and originality laid down in Indian law. In other words, does the human input in the work show sufficient “skill and judgement” to be copyrightable? This is an important question to ask since there is surely a marked difference in skill and judgement used in the seven-step prompts in the Spring Breeze case, as opposed to a prompt that simply says “create an image of a beautiful woman.”
Nevertheless, recognising the extent of AI input in a work is a challenging task. However, there are ways to address this. For instance, in the US, every copyright registration application requires disclosures of the manner and process in which the AI was used in the final work.[xlvi] Further, the office also seeks additional information on the process and inputs in the creation of the work.[xlvii] An understanding of the operation and training of AI models can also inform this analysis. In this regard, the US has introduced reporting and disclosure requirements for all AI developers.[xlviii] If done right, India’s copyright regime can harness AI’s significant potential to promote creativity and industry.[xlix] To do so, it must revisit its roots.
IV. CONCLUSION
In this paper, I have comparatively analysed the US and Indian copyright regimes to explore how and what kinds of Human-AI work can be copyrighted. While AI systems cannot be independent ‘authors’, they may be used to facilitate the creation of Human-AI works. However, the US does not grant copyright to any Human-AI work with an ‘appreciable’ amount of AI input. I argued that this standard may stifle innovation and creativity. Given India’s different landscape and underlying policy goals, I further argued that India can allow Human-AI works to receive a copyright if they meet the standards of creativity and originality in Indian law. In this manner, India’s IP regime can harness AI’s potential instead of combating it.
[i] The author is a student at the National Law School of India University, Bangalore.
[ii] Jane C. Ginsburg & Luke Ali Budiardjo, Authors and Machines, SSRN Journal, 346–347 (2018), https://www.ssrn.com/abstract=3233885
[iii] Department Related Parliamentary Standing Committee on Commerce, One Hundred and Sixty-First Report on the Review of the Intellectual Property Rights Regime in India 8.3 (2021).
[iv] United States Copyright Office, Copyright Registration Guidance (2023).
[vi] Ginsburg and Budiardjo, Supra Note 1
[vii] Community for Creative Non-Violence v. Reid, 490 U.S. 730, 737 (1989).
[viii] Naruto v. David John Slater, No.3:15-cv-04324-WHO 17 (2018).
[ix] Stephen Thaler v. Shira Perlmutter, No. 22-1564 BAH 15 (2023).
[x] Id. at 8.
[xi] Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 59 (1884)
[xii] Copyright Act, 1957, §2(d).
[xiii] Tech Plus Media Ltd. v. Jyoti Janda, SCCOnline Del 1819, ¶ 20 (2014); Delhi Public School Society v. DPS World Foundation and Anr., SCCOnline Del 2302, 30, 64 (2016).
[xiv] Copyright Act, 1957, §§13,20.
[xv] Id., §57; Ritvik M. Kulkarni, Of Artificial Intelligence and Authorship, Spicyip (2016), https://spicyip.com/2016/12/of-artificial-intelligence-and-authorship.html (last visited Jan 15, 2024).
[xvi] The Copyright Act, 1957, §§2(d)(vi), 2(ffb).
[xvii] Aparajita Lath, AI and Copyright: More Developments – Human Prompts Are Not ‘Direct Instructions,’ Spicyip (2023), https://spicyip.com/2023/09/ai-and-copyright-more-developments-human-prompts-are-not-direct-instructions.html
[xviii] United States Copyright Office, Artificial Intelligence and Copyright 3 (2023).
[xix] Ginsburg and Budiardjo, Supra note 1 at 424.
[xx] Copyright Registration Guidance, Supra 2 at 5.
[xxi] Appreciable or De Minimis: That Is the AI Question, https://www.dwt.com/blogs/artificial-intelligence-law-advisor/2023/12/copyright-appreciable-de-minimis-ai-generated-work (last visited Jan 15, 2024).
[xxii] Id.
[xxiii] Feist Publications, Inc., v. Rural Telephone Service Co., 499 U.S. 340 55.
[xxiv] Id., 53.
[xxv] Li v. Liu Jing 0491 Min Chu No. 11279 (2023)
[xxvi] Id., 13.
[xxvii] Id., 14.
[xxviii] Tian Lu, Chinese Court Deems AI-Generated Image Has Copyright – Assessing the Possibly over-Hasty ‘Spring Breeze’ Case, The IPKat, https://ipkitten.blogspot.com/2023/12/chinese-court-deems-ai-generated-image.html
[xxix] For more applications of USCO’s “appreciable” standard to work with creative human input, see Copyright Review Board of United States Copyright Office, Second Request for Reconsideration for Refusal to Register A Recent Entrance to Paradise, (2022); United States Copyright Office, Zarya of the Dawn (Registration # VAu001480196), (2023)
[xxx] Andreas Rahmatian, Originality in UK Copyright Law: The Old “Skill and Labour” Doctrine Under Pressure, 44 IIC 4, 4–5 (2013).
[xxxi] Feist, supra 20 at para 31-para 32.
[xxxii] Id., para 33.
[xxxiii] Id., para 19.
[xxxiv] Jani Ihalainen, Computer Creativity: Artificial Intelligence and Copyright, 13 Journal of Intellectual Property Law & Practice 724, 725 (2018).
[xxxv] Parliamentary Committee Report, Supra note 2.
[xxxvi] Id., para 8.1 – para 8.2.
[xxxvii] Id., para 8.3.
[xxxviii] The only time that AI and copyright have been an issue is the Delhi HC case of Anil Kapoor v. Simply Life India, SCC OnLine Del (2023). However, the suit concerned only defamation, and the court did not need to pass on the copyrightability of AI work. On the other hand, the Indian Copyright Office has rejected an application for an AI-generated work, which is explored in this paper below.
[xxxix] NYT litigation against OpenAI and Microsoft in Boom in A.I. Prompts a Test of Copyright Law, The New York Times, https://www.nytimes.com/2023/12/30/business/media/copyright-law-ai-media.html
[xl] Terms of use, https://openai.com/policies/terms-of-use (last visited Jan 15, 2024); Can I use my images commercially? | Midjourney Help, https://help.midjourney.com/en/articles/8150363-can-i-use-my-images-commercially (last visited Jan 15, 2024).
[xli] Eastern Book Company v. D.B. Modak 1 SCC 1 ¶50 (2008).
[xlii] Id., para 57.
[xliii] Lath, Supra note 15.
[xliv] Id.
[xlv] Copyright Review Board of the United States Copyright Office, Second Request for Reconsideration for Refusal to Register SURYAST, (2023).
[xlvi] Copyright Office Guidelines, Supra note 3 at 5-6.
[xlvii] United States Copyright Office, Zarya of the Dawn (Registration # VAu001480196), (2023).
[xlviii] Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence (October 30, 2023).
[xlix] Parliamentary Committee Report, supra 2 at para 8.2.
Comments