Generative AI and Authorship in Copyright Law

I.  AI as tools or AI as authors?

Today, rapid advancements in AI capabilities to create literary, dramatic, musical and artistic works continue to redefine the human role in the creative process. Most of these works generated by computers rely heavily on the underlying algorithm and creative input of the programmers; the computers are arguably akin to paintbrushes or chisels – they are tools used in the creation of the artworks.[1] Jane Ginsburg and Luke Ali Budiardjo referred to the “amanuensis” – who acts as an agent by faithfully carrying out the subordinate task assigned by the principal – as distinct from the author in copyright law, and therefore to whom the attribution of authorship should not be accorded.[2]

While “AI art” has been referred to as “neural network art”, many online commentaries do not make a clear distinction between whether the AI is used as a tool by a human individual (e.g. how internationally renowned artists Sougwen Chung uses hand-drawn and computer-generated marks in her drawings, sculptures and installation works, and Scott Eaton who creates and trains AI to translate his drawings and animation into photographic, figurative representations as well as abstracted sculptural forms[3]) or the AI independently and autonomously produces a work without supervision or significant human intervention. For the services provided by OpenAI, which includes ChatGPT and DALL·E, the terms of use state that “OpenAI hereby assigns to you all its right, title and interest in and to Output” but cautions that “[d]ue to the nature of machine learning, Output may not be unique across users and the Services may generate the same or similar output for OpenAI or a third party.”[4] What this means is that, assuming that the work generated by ChatGPT or DALL·E is capable to attracting copyright protection, the copyright owner is the user who inputs the text prompts. Under the DALL·E Content Policy help section of OpenAI’s website, it is stated that “Subject to the Content Policy and Terms, you own the images you create with DALL·E, including the right to reprint, sell, and merchandise – regardless of whether an image was generated through a free or paid credit.”[5]

II.  Human Authorship

But the assignment of copyright to the user who provides the text prompts is valid if and only if the AI-generated output may be attributed to a human author in the first place; the maxim nemo dat quod non habet applies. Ownership is contingent on authorship as recognised according to the law. Much already has been written in this area with a preponderance of views against the extension of copyright to computer-generated works as a result of autonomous decisionmaking by AI systems, and this article would not be addressing each specific argument canvassed by different scholars.[6] Due to our shared Commonwealth common law heritage, the prevailing legislation and case law that govern the recognition of copyright in works in the UK and Australia may be relevant to Singapore law. The UK’s Copyright, Designs and Patents Act 1988 (“CDPA”), Australia’s Copyright Act 1968 and Singapore’s Copyright Act 2006 divide copyright subject matter into two categories – original authors’ works and “subject-matter other than works.”[7] In Singapore’s revamped Copyright Act 2021, the literary, dramatic, musical and artistic (LDMA) works are known as “authorial works”[8] within the broad definition of “works” which comprise LDMA works and what used to be “subject-matter other than works” (e.g. sound recording, film and broadcast).[9]

The Commonwealth common law jurisdictions have consistently premised their copyright regimes on requiring human authorship. In Asia Pacific Publishing Pte Ltd v Pioneers & Leaders (Publishers) Pte Ltd, the Singapore Court of Appeal cited the UK’s Copyright Act 1911 and CDPA as examples of copyright legislation implying human authorship, since they afforded copyright protection to authors for their lifetime plus 50 years. While section 9(3) of the CDPA appears to afford copyright protection to computer-generated LDMA works even in the absence of a human author,[10] it has been interpreted by English courts to require the identification of a “causal link” between the computer-generated work and a human author.[11] The English High Court affirmed the human authorship requirement in Nova Productions Ltd v Mazooma Games Ltd when it applied the computer-generated work sections of the CDPA to the computer-generated composite frames, and identified the human programmer in that case as the author of these artistic works.[12] In addition, section 32 of the Australian Copyright Act 1968 clarifies that qualified persons refer to natural persons.[13] Complementing the statutory approach, Australian case law, such as in Telstra Corporation Ltd v Phone Directories Company Pty Ltd and more recently in Acohs Pty Ltd v Ucorp Pty Ltd, requires that the “author” be an “actual person” and a “human author.”[14]

In Singapore, the Court of Appeal in a 2017 decision in Global Yellow Pages Ltd v Promedia Directories Pte Ltd affirmed its earlier comment in Asia Pacific Publishing Pte Ltd v Pioneers & Leaders (Publishers) Pte Ltd regarding the ‘natural persons’ requirement,[15] where the same court held that for copyright to subsist in any literary work, the authorial creation must causally connect with the “engagement of the human intellect.”[16] The Court of Appeal then proceeded to define human intellect as “the application of intellectual effort … or the exercise of mental labour,” which a non-human author is deemed to be unable to provide.[17] Furthermore, in Singapore’s new Copyright Act 2021, a suite of statutory provisions when read together indicate that only a human individual or natural person may be an “author”: (i) the duration provisions (s 114) – where duration is pegged to the death of a person (i.e. 70 years after death unless in the case of anonymous/pseudonymous works); (ii) the “qualified individual” provision (s 77) – where copyright in an authorial work subsists only if the author is a qualified individual;[18] (iii) the connecting factors provisions (ss 109, 110) that articulate the conditions for copyright to subsist in unpublished and published authorial works; and (iv) the moral rights provisions (ss 370, 386, 387) – which refer to rights being personal in nature, and devolution of rights on death.

In 2018 the US Copyright Office (USCO) received an application for a visual work, “A Recent Entrance to Paradise” that the applicant described as “autonomously created by a computer algorithm running on a machine.” The application was denied by the USCO because, based on the applicant’s representations in the application, the examiner found that the work contained no human authorship. After a series of administrative appeals, the Office’s Review Board issued a final determination affirming that the work could not be registered because it was made without any creative contribution from a human actor.[19] In 2023, the USCO reviewed a registration for a work containing human-authored elements combined with AI-generated images, and concluded that the graphic novel “Zarya Of The Dawn”, comprised of human-authored text combined with images generated by the AI service Midjourney, constituted a copyrightable work, but that the individual images themselves could not be protected by copyright.[20] While there is much debate recently about whether AI that autonomously creates works can be an “author” in copyright law, the USCO has issued a strong and comprehensive statement in March 2023 that it does not recognise the AI as author and would not be accepting registration of copyright for such works.[21]

III. Conclusion – Finding that Causal Nexus

In summary, the human authorship requirement would not be abandoned even in the face of technological developments, as the collective wisdom of case law across numerous jurisdictions clearly demand the identification of a human author. For ChatGPT and DALL·E operated by OpenAI, as well as other generative AI systems like Stable Diffusion and Midjourney, it would appear that OpenAI asserts presumptive primary authorship over the output and then assigns its ownership rights to the user who provided the text prompts. However, it is inconclusive that in every scenario, the output is attributable to a human author be it the programmer of the algorithms that drive the generative AI system or the human individual who supplies the text prompts or a combination of the two. What is clear today is that when the human input lacks a “sufficient causal nexus with the final work”,[22] then the human author, from whom a work originates, cannot be identified.[23] As a result, what we have is an authorless work – no matter how aesthetic, useful or valuable. This was pointed out in Payen Components South Africa Ltd v Bovic Gaskets CC: “There may be cases where the real work has been done by the computer, the human contribution being too trivial or not sufficiently related to the work that has emerged.”[24]

In the generative AI systems like ChatGPT or DALL·E, it would really depend on how much human input in terms of specific text prompts that may be seen to satisfy the causal nexus requirement. A general command like “Create a painting of the Singapore Merlion in the style of Picasso” on DALL·E would not suffice as sufficiently engaging the human intellect for the creation of the work. Ultimately, the line which delineates human-authored AI-aided works from autonomous AI-authored works is unclear.[25]

 

AUTHOR INFORMATION

Professor David Tan is the Co-Director of the Centre for Technology, Robotics, Artificial Intelligence & the Law (TRAIL) and Head (Intellectual Property) of the EW Barker Centre for Law & Business at NUS Law.

Email:  david.tan@nus.edu.sg

 

REFERENCES

[1]    Andres Guadamuz, “Artificial Intelligence and copyright”, WIPO Magazine, October 2017 <https://www.wipo.int/wipo_magazine/en/2017/05/article_0003.html>.

[2]    Jane C. Ginsburg and Luke Ali Budiardjo, “Authors and Machines” (2018) 34 Berkeley Tech. L.J. 344 at 355. See also ibid at 360 (internal citations omitted) (“Attribution of authorship effectively follows general rules of agency: ‘the physical acts of the agent are attributed wholly to the author’ under whose control and direction the amanuensis acts.”).

[3]    Christopher McFadden, “7 of the Most Important AI Artists That Are Defining the Genre”, Interesting Engineering, 10 November 2019 <https://interestingengineering.com/7-of-the-most-important-ai-artists-that-are-defining-the-genre>; Sougwen Chung, AIArtists <https://aiartists.org/sougwen-chung>; “Scott Eaton Artist + AI: Figures & Form in the Age of Intelligent Machines, Lux Review <https://www.lux-review.com/scott-eaton-artistai-figures-form-in-the-age-of-intelligent-machines/>.

[4]    Terms of Use, OpenAI (14 March 2023): https://openai.com/policies/terms-of-use.

[5]    DALL·E Content Policy, OpenAI (15 May 2023): https://help.openai.com/en/articles/6425277-can-i-sell-images-i-create-with-DALL·E

[6]    E.g. Mauritz Kop, “AI & Intellectual Property: Towards an Articulated Public Domain” (2020) 28 Texas Intellectual Property Law Journal 297; Megan Svedman, “Artificial Creativity: A Case Against Copyright for AI-Created Visual Artwork” (2020) 9:4 IP Theory 1; Pratap Devarapalli, “Machine Learning to Machine Owning: Redefining the Copyright Ownership from the Perspective of Australian, US, UK and EU Law” (2018) 40 EIPR 72; Massimo Maggiore, “Artificial Intelligence, computer generated works and copyright” in Enrico Bonadio and Nicola Lucchi (eds), Non-Conventional Copyright: Do New and Atypical Works Deserve Protection? (2018) 382; James Grimmelmann, “There’s No Such Thing as a Computer-Authored Work – and It’s a Good Thing, Too” (2016) 39 Columbia Journal of Law & the Arts 403.

[7]    The key reason for distinguishing original LDMA works from “subject-matter other than works” is because only LDMA works require originality in the sense of originating from a human author. In the earlier Singapore Copyright Act 2006, “qualified person” for LDMA works is restricted to natural persons (s 27(4)) while “qualified person” for “subject-matter other than works” includes “a body corporate incorporated under any written law in Singapore” (s 81(1)(b)). See Copyright Act (Cap 63, 2006 Rev Ed Sing).

[8]    Copyright Act 2021 s9 (an “authorial work” is a literary, dramatic, musical or artistic work).

[9]    Copyright Act 2021 s8.

[10]   Copyright, Designs and Patents Act 1988 (c. 48; U.K.) (“CDPA”). Section 178 defines “computer-generated” as work that is “generated by computer in circumstances such that there is no human author of the work.”

[11]   Jacob Turner, Robot Rules Regulating Artificial Intelligence (Palgrave Macmillan, 2019) at 125.

[12]   Nova Productions Ltd v Mazooma Games Ltd [2006] EWHC 24 (Ch) at [12]-[18] and [104] (“Nova Productions”).

[13]   Asia Pacific Publishing Pte Ltd v Pioneers & Leaders (Publishers) Pte Ltd [2011] 4 SLR 381 at [48] (“Asia Pacific Publishing”).

[14]   Telstra Corporation Ltd v Phone Directories Company Pty Ltd [2010] FCAFC 149 at [100] and [134] (“Telstra”); Acohs Pty Ltd v Ucorp Pty Ltd [2012] 201 FCR 173 at [57].

[15]   Asia Pacific Publishing [2011] 4 SLR 381 at [82] (“without the identification of a human author from whom the work originates, there can be no ‘original work’ capable of copyright protection.”).

[16]   Global Yellow Pages Ltd v Promedia Directories Pte Ltd [2017] 2 SLR 185 at [24] (“Global Yellow Pages”).

[17]   Global Yellow Pages [2017] 2 SLR 185 at [24].

[18]   According to this provision, an individual is a “qualified individual” only if “he or she” is (a) a Singapore Citizen; or (b) a Singapore resident; or (c) an individual who, if he or she had been alive on 1 November 1957, would have qualified for Singapore citizenship under the repealed Singapore Citizenship Ordinance 1957. Copyright Act 2021, s 77 (emphasis added).

[19]   U.S. Copyright Office Review Board, Decision Affirming Refusal of Registration of a Recent Entrance to Paradise at 2 (14 February 2022): https://www.copyright.gov/rulings-filings/review-board/docs/a-recent-entrance-to-paradise.pdf.

[20]   U.S. Copyright Office, Cancellation Decision re: Zarya of the Dawn (VAu001480196) at 2 (21 February 2023), https:// www.copyright.gov/docs/zarya-of-the-dawn.pdf.

[21]   US Copyright Office, ‘Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence’ 37 CFR Part 202 (16 March 2023): https://www.copyright.gov/ai/ai_policy_guidance.pdf.

[22]   Global Yellow Pages [2017] 2 SLR 185 at [24].

[23]   Asia Pacific Publishing [2011] 4 SLR 381 at [82] (where the court held that “without the identification of a human author from whom the work originates, there can be no ‘original work’ capable of copyright protection.”).

[24]   Payen Components South Africa Ltd v Bovic Gaskets CC and Others (448/93) [1995] ZASCA 57 at [15].

[25]   Toby Bond and Sarah Blair, “Artificial Intelligence & copyright: Section 9(3) or authorship without an author” (2019) 14 Journal of Intellectual Property Law & Practice 423, 423. See also a more comprehensive analysis in Wee Liang Tan and David Tan, ‘AI, Author, Amanuensis’ (2022) 5(2) Journal of Intellectual Property Studies 1.