I. ABSTRACT
This piece addresses the 21st-century challenge of balancing artists’ rights to their original works with the use of copyrighted works by AI models for self-improvement. It begins by untangling complex issues surrounding authorship, questioning whether AI itself can be considered an author or if the developer of the AI model holds that distinction. The article explores various global legal frameworks for this matter and suggests a range of solutions. Rather than advocating a singular solution, the authors propose a combination of identified approaches, emphasizing that countries can adopt a mix of these solutions to strike a balance between artists' rights and AI model development through their works.
II. INTRODUCTION
“The art challenges the technology, and the technology inspires the art.”
~ John Lasseter[ii]
December 20th, 2023 marked a significant day for the Artist community. A lawsuit was filed in Manhattan federal court by a group of 11 nonfiction authors (namely Pulitzer Prize winners Stacy Schiff, Taylor Branch, and Kai Bird - who had co-written the J. Robert Oppenheimer biography "American Prometheus")[iii]. The authors allege that OpenAI and Microsoft have improperly utilized their written works to train the models behind OpenAI's ChatGPT and other AI-based software.[iv] This Incident makes us ponder over the extent to which technology affects the rights of artists.[v]
The introduction of innovations such as Chat Generative Pre-training Transformer (Chat GPT) and various other generative AI technologies has already begun to revolutionize the digital landscape. Some perceive these advancements as the initial strides toward achieving Artificial General Intelligence (AGI).[vi] This raises concerns regarding potential copyright infringements of the work of Artists who had dedicated decades to honing their skills.[vii]
III. UNTANGLING THE DILEMMA
The dilemma persists because humans are advancing at such a pace that they are very close to giving human consciousness to machines.[viii] These unprecedented conceptions raise a series of complicated issues that must be simplified before their solutions can be discussed. The Dilemma here is not the fair use of Artificial Intelligence (AI) (examples include Midjourney, Chat GPT, and Stable Diffusion among other things) but how the AI technology is misusing the copyrighted works available on the Internet to train itself and suggest outputs.[ix]
These generative AI models are feeding on the travails of Artists to design tailored outputs.[x] The work of Artists that they could master after years of struggle, education, and practice is open to misuse and modification without any credit, consent, or compensation to them.[xi] Additionally, questions arise as to whether this prompt-generated work can be given a Copyright, if yes, then who is to be the author of such work? Is it the AI itself or the AI developer? Furthermore, if violations occur, they are often justified by Corporate Giants under the garb of the Fair use doctrine. Whether the use of AI for such purposes fall under the Fair Use Doctrine?
To unravel this quandary let’s dive deep and deal with each issue separately and build a balanced approach to tackle this issue.
Whether prompt-generated work can be given a Copyright, if yes then who is to be the author of such work? Is it the AI itself or the AI developer?
The definition of the Author has seen three divergences. Firstly, only humans have been allowed to hold copyright for their works (2.1). Secondly, some legislation has provided for the computer software inventor to hold the copyright for the works generated by the software (2.2). Thirdly, in the evolving definition of author, certain legal systems have allowed for AI to be co-authors in copyrighted works. (2.3)
2.1 The essential human element
Firstly, in the case of the United States of America, the copyright Office states that it will “register an original work of authorship, provided that the work was created by a human being.” This understanding flows from the case of Fiest Publications[xii] and Stephen Thaler[xiii] which clarifies that copyright law only protects “the fruits of intellectual labor” that “are founded in the creative powers of the mind.”[xiv] Additionally, Australia is seen holding a similar stand.[xv]
However, the US Patent Office has shown a unique transgression in the case of Kristina Kashtanova[xvi], wherein an AI-generated comic book was granted registration[xvii]. However, the subsequence revocation of registration confirms the US’s stand on the need for a human author.
According to USA Copyright Office policy[xviii] in evaluating AI-generated content, the determining factor lies in distinguishing whether the contributions stem from "mechanical reproduction" or if they represent the author's "own original mental conception, to which (the author) gave visible form."[xix] This inquiry is inherently case-specific, as the unique characteristics of each situation will influence the attribution of authorship and the nature of creative input.
2.2 Developer is given the Authorship
The second option of giving authorship to the programmer, is evident in a few countries such as India[xx], Ireland[xxi], New Zealand, Hong Kong (SAR), and the UK (whose approach was adopted by the former countries). This approach is best depicted in UK copyright law[xxii], The Copyright, Designs and Patents Act 1988(CDPA), which states that:
“In the case of a literary, dramatic, musical or artistic work which is computer-generated, the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken.[xxiii]”
The significance of this approach is the inherent distinction between “computer-assisted works and computer-generated works”. According to Indian legislation, the author of computer-generated literary, dramatic, musical, or artistic works is deemed to be the "person who causes the work to be created," as introduced through the 1994 amendments.[xxiv]
Joint Committee Report on Copyright (Second Amendment bill1992)[xxv] India, recognized the transformative potential of artificial intelligence, acknowledging that computers, through this technology, could generate new ideas beyond what is explicitly programmed into the system. The Committee drew inspiration from the Copyright, Designs, and Patents Act, 1988 of the UK,[xxvi] which extends protection to "computer-generated works" where there is no identifiable human author.
Crucially, the Committee underscored the necessity of distinguishing between computer-generated works, where there is no discernible human authorship, and computer-assisted works, where human contributions are readily identifiable.
2.3 The AI model as a co-author
The third option, that of recognizing AIs as co-authors can be understood through Stephen Thaler's attempts to secure patents for inventions generated by his AI system (Co-Pilot), known as the Device for the Autonomous Bootstrapping of Unified Sentience (DABUS),[xxvii] which faced rejection in various countries, including Australia, the United Kingdom, the United States, New Zealand, and the European Patent Office.[xxviii] The rejections were based on the rationale that patent laws in these jurisdictions do not permit the attribution of inventorship to an AI system.[xxix]
Similar to DABUS, the copyright office in India, though in error, acknowledged an AI system named Robust Artificially Intelligent Graphics and Art Visualizer (RAGHAV), [xxx] as a co-author of artistic work and approved the application for copyright protection. Subsequently, the copyright office issued a withdrawal notice for the registration, acknowledging its inadvertent approval. Mr. Sahni, the human co-author, was requested to consider the legal status of the AI system RAGHAV. Although the copyright office's website still indicates a 'registered' status for the application, a court decision on this matter is pending.[xxxi]
Current status of AI-generated work in other countries
Country | Definition of Author | Status of AI-generated work being given copyright | Reasoning. |
USA | Must be Human. | Human authorship is a bedrock requirement for copyright protection.[xxxii] | The Copyright Office iterated authorship requires human input and that merely mechanical or machine-produced works devoid of human creativity are uncopyrightable material. |
EU | Must be Human.[xxxiii] | The human element must be evident in the work. | Copyright grants protection to only original works, and that originality must reflect the “author’s intellectual creation.” Work must reflect the author’s personality, which means that a human author is essential for a copyrighted work to exist.[xxxiv] |
China | Can be AI. | AI-generated work can be granted copyright if the author has made a certain amount of intellectual investment.[xxxv] | It was stated by China’s court that the more technology grows, the smarter the tools become, and the less time people will need to invest in creative work, but this should not hinder the continued application of copyright laws to promote creative activities.[xxxvi] |
India | Follows the EU model after 1994 amendments to the Copyright Act, India.[xxxvii] | In a unique case, RAGHAV was given co-authorship of copyright-ed work along with a human but sole authorship was denied. Presently the registration is under review. | Jurisprudence is still evolving. |
IV. THE SOLUTIONS
The dilemma highlighted is complicated hence no straightjacket solution will work. We must find a series of solutions that work in tandem to balance the rights of artists with innovation and creation using AI.
3.1 Fair Use Doctrine (a Curative Approach)
The current issue with fair use is that many big-tech companies justify data theft through the doctrine of fair use to train their generative AI models. This fair use defense was taken by the mid-journey AI against the infringement case file by Getty Image (Offers professional, licensed photos)[xxxviii] by relying on the judgment of Author Guild vs Google[xxxix] in which the US Court stated that copying and display of snippets from copyrighted books fell under the "fair use" doctrine, which allows limited use of copyrighted material for criticism, commentary, reporting, academics, and other purposes.[xl] The issue with this is that it sets a precedent for the use of ‘fair use’ in the digital age, expanding its scope to include large-scale digitization projects of big tech companies like Google Book Search and training of generative AI models.[xli]
The current fair use doctrine, designed for older technologies like criticism and commentary, may not adequately address the specific nuances of training large language models. Allowing the wholesale use of copyrighted works for training AI models could negatively impact creators by potentially diminishing the market for their original works.[xlii] The amount and nature of copyrighted material used in training can be vast and intricate, making it difficult to apply the four-factor fair use test[xliii] clearly, and consistent new criteria must be established.
An article written by Ariel Soiffer (partner) and Aric Jain (associate) at the law firm WilmerHale[xliv] laid down interesting approaches to analyzing the use of the Fair Use Doctrine in justifying the AI using copyrighted material to train itself. These approaches are discussed below-
3.1.1 Fair Use Minimalism-
The preachers of this approach believe in the minimum use of the fair use doctrine and believe that there should be a very limited number of cases wherein this doctrine can be relied upon. However, they do rely on the fact that how much of the work has been relied on by the AI.
3.1.2 Fair Use Maximisation-
At the opposite end of the spectrum, Fair use maximization preaches the doctrine to cover almost every case believing that the medium of AI delivers the requisite transformation to the derivative work. Naturally, the big tech companies today as vehemently advocating this approach.
3.1.3 Conditional Fair Use Maximisation-
Sitting between the extreme ends of the aforementioned approaches is the Conditional Fair use maximization method which preaches that each alleged infringement must be judged on a case-by-case basis. To put it simply, it asks-
Whether heart of the work is used?
How specific was the prompt engineering?
Whether the AI is using the idea behind the copyrighted work or the expression through which it was depicted.[xlv]
This approach would introduce a requirement for AI companies; they must document the prompt generating process of the users extensively to fall under the virtue of fair use doctrine and steer clear of the vice of copyright infringement.
These approaches are not proposed as solutions individually, but a combination of any of these together might solve the complicated issue at hand.
3.2 The 3C Solution (Credit, Consent, Compensation)
The 3C Solution underscores the importance of the generative AI developer giving recognition and respect to the creative contributions of the artist. To uphold this principle, it is imperative to ensure that proper and fair credit is attributed to the artist for their work. Any breach of a paid wall,[xlvi] particularly one safeguarding the artist's creativity, must be handled with utmost sensitivity, and fair compensation should be awarded to the artist in such instances.[xlvii]
Furthermore, when considering additional training in the Generative AI model using the artist's data, it is essential to adhere to ethical standards. Seeking explicit consent (through a license agreement) from the artist through a transparent and well-defined channel is crucial. This process involves clear communication and a comprehensive explanation of how the data will be used, maintaining transparency and fostering a relationship built on trust between the LLM model developer and the artist.
3.3 Ensuring Transparency Source Disclosure in Privacy Policy for Data Training Mechanisms
Transparency in tracing the origins of training data enables users to discern potential biases and limitations within the AI model. This knowledge empowers users to make informed decisions, fostering prudent use while respecting copyright considerations. Moreover, in alignment with ethical principles, artists must be given the explicit right to opt-out[xlviii] if they choose not to continue contributing their data to the training of AI models, ensuring both autonomy and collaborative integrity.
3.4 The Inculcation of Ethical Elements in Technology
Developers must embed ethical considerations directly within generative models. This approach not only serves as a proactive measure, notifying developers in the event of a breach but also contributes to preserving the originality of the artist's work.[xlix]
As an exemplary case, consider the implementation of glaze technology[l]. It enables artists to apply “style cloaks” to their art before sharing it online. The cloaks “apply barely perceptible perturbations to images, and when used as training data, mislead generative models that try to mimic a specific artist[li]”. By using the Glaze app, artists can protect not only their work from being used by AI models without their permission but also the process through which they arrive at their product.
3.5 Regulatory Framework for LLM Models
Addressing liability and the changes in the current IPR regime across the country through establishing clear frameworks to attribute responsibility to AI owners, creators, or operators. Clear criteria should be established for determining liability in copyright infringement cases involving AI-generated content. These guidelines should differentiate between intentional human infringement and unintentional or algorithmic infringement.[lii]
3.6 Recognizing Creative Input (Sufficient mental use)
Two types of works would come before the copyright office, work created by AI with human interference and secondly, work created by AI without human interference. In the first category, the owner of the work must be acknowledged for providing creative inputs (the amount of mental exercise put into the prompt) to the AI as the owner of the work.[liii] This reflects their contribution and control over the creative process. Additionally, AI can be recognized as the author, acknowledging its role in generating the work based on human inputs and its capabilities. In the work created under the second category, the person who owns the AI software or the AI system itself should be recognized as the owner, as they hold the rights to the AI’s output.[liv]
V. CONCLUDING REMARKS
As we stand on the precipice of a transformative era marked by technological advancements, the narrative extends beyond mere progress. Balancing the rights of the artist without stifling innovation, demands a nuanced and detailed approach to tackle this recent challenge in the rapidly evolving tech world. Various factors must be considered while establishing a regulatory framework, ensuring due credit, fair compensation, and ethical considerations are adequately addressed. Striking this delicate balance not only safeguards artistic integrity but also fosters an environment where technological advancements and creative expression can coexist harmoniously.
The changes in the International IPR regime will further be impacted as the regulatory mechanism and understanding around Artificial technology grows. The EU is taking the lead with the introduction of its European Guild of Artificial Intelligence Regulation aims to "protect artists and copyright holders from predatory AI companies through the creation of laws and regulations”.[lv] India is introducing sweeping changes in its IT infrastructure with the upcoming Digital India Act.[lvi] All these changes will impact the use of AI, however, how the impact will take place is yet to be seen.
[i] The author is a student at the Dharmashastra National University, Jabalpur.
[ii] John Lasseter, the art challenges the technology, and the technology challenges the art, All Author (Dec. 20, 2023). https://allauthor.com/quotes/55481/.
[iii]Blake Brittain, Pulitzer-winning authors join OpenAI, Microsoft copyright lawsuit, Reuters (Dec. 24, 2023) https://www.reuters.com/legal/pulitzer-winning-authors-join-openai-microsoft-copyright-lawsuit-2023-12-20/.
[iv] Id
[v] Global X Etf, A decade of change, how tech evolves in the 2010 and what’s in the store for the 2020s, Nasdaq (Nov. 28, 2023), https://www.nasdaq.com/articles/a-decade-of-change%3A-how-tech-evolved-in-the-2010s-and-whats-in-store-for-the-2020s.
[vi] Sam Altman, Planning for AGI and Beyond, OpenAI (Nov. 28, 2023), https://openai.com/blog/planning-for-agi-and-beyond.
[vii]Christiopher T. Zirpoli, Generative Artificial Intelligence and Copyright, Congressional Research Service (Nov. 28, 2023),https://crsreports.congress.gov/product/pdf/LSB/LSB10922#:~:text=Generative%20AI%20also%20raises%20questions,that%20resemble%20those%20existing%20works.
[viii] Supra note 4.
[ix] Gordon Rees and Scully Mansukhani, AI-generated Content and Copyright Ownership Right in the U.S, (Dec. 21, 2023) https://www.lexology.com/library/detail.aspx?g=93cf3f01-5e64-4088 9af237fd38165e9#:~:text=The%20Copyright%20Office%20guidance%20indicatesOffice%20will%20not%20register%20it.
[x] Patrick Wheeler & Zoe Deckker, Does AI Image Generation Infringe artist copyright, Arts Professional (Dec.21, 2023) https://www.artsprofessional.co.uk/magazine/article/does-ai-image-generation-infringe-artists-copyright#:~:text=Artists%20are%20increasingly%20concerned%20that,rise%20of%20AI%2Dgenerated%20artworks.&text=Copyright%20is%20a%20property%20right%20which%20recognises%20an%20artist’s%20creativity%20and%20originality.
[xi] Id.
[xii] Feist Publications v Rural Telephone Service Company, Inc. 499 U.S. 340 (1991).
[xiii] Stephen Thaler v. Shira Perlmutter, Register of Copyrights and Director of the United States Copyright Office Civil Ac on No. 22-1564 (BAH).
[xiv] Id.
[xv] Acohs Pty Ltd v Ucorp Pty Ltd. [2012] FCAFC 16.
[xvi]Benjamin Dehant, Made by AI, Medium (Dec. 24, 2023), https://bootcamp.uxdesign.cc/i-interviewed-kristina-kashtanova-the-first-person-who-copyrighted-a-genai-creation-b8be4778654c.
[xvii]Monit Khanna, Comic book made by AI loses Copyright Protection, It (Dec. 24, 2023) https://www.indiatimes.com/technology/news/comic-book-made-by-ai-loses-copyright-protection-588388.html.
[xviii] U.S Copyright Office, Copyright registration guideline, Federal Register (Dec. 24, 2023), https://www.federalregister.gov/documents/2023/03/16/2023-05321/copyright-registration-guidance-works-containing-material-generated-by-artificial-intelligence.
[xix] Id.
[xx]The Copyright Act, 1957, § 2(d), No. 14, Acts of Parliament, 1957.
[xxi] Irish Copyright and Related Rights Act, 2000, § 21, S.I. No. 157/2007.
[xxii] The Copyright, Designs and Patents Act, UK, § 9(3)
[xxiii] Id.
[xxiv]Pallavi Sondhi, A brief look at the copyright issues raised by generative AI, Ikigai Law (Dec.23, 2023), https://www.ikigailaw.com/article/11/a-brief-look-at-the-copyright-issues-raised-by-generative-ai.
[xxv]Joint Committee Report on Copyright (Second Amendment Bill 1992, Rajya Sabha (Dec.24, 2023) https://rsdebate.nic.in/handle/123456789/215197?viewItem=browse.
[xxvi] UK Copyright, Design, and Patent Act, 1988, CDPA; c. 48.
[xxvii] Renu Bala Rampal and Swaraj Singh Raghuwanshi, Demystifying Rights Of AI-Generated Inventions, Livelaw (Dec.22, 2023), https://www.livelaw.in/law-firms/law-firm-articles-/ai-generated-inventions-chatgpt-indian-patent-act-dabus-united-states-patent-trademark-office-european-patent-office-226394.
[xxviii] Saiko Hidaka, Court of Appeal - AI Generated Inventions Denied UK Patent in Dabus Case, Gowling WLG (Dec.22, 2023) https://gowlingwlg.com/en/insights-resources/articles/2021/updated-ai-invention-denied-patent-in-dabus-case/.
[xxix] Id.
[xxx] Sukanya Sarkar, Exclusive: India recognizes AI as co-author of copyrighted artwork, Managing IP (Dec.22, 2023) https://www.managingip.com/article/2a5czmpwixyj23wyqct1c/exclusive-india-recognises-ai-as-co-author-of-copyrighted-artwork.
[xxxi] Rajiv Sharma and Ninad Mittal, Artificial Intelligence Lacks Personhood To Become The Author Of An Intellectual Property, Livelaw (Dec.22, 2023), https://www.livelaw.in/law-firms/law-firm-articles-/artificial-intelligence-intellectual-property-indian-copyright-act-singhania-co-llp-238401.
[xxxii] Thaler v. Perlmutter, No. CV 22-1564 (BAH), 2023 WL 5333236, at *3.
[xxxiii] C-5/08 Infopaq International A/S v Danske Dagbaldes Forening [C-5/08 Infopaq International A/S v Danske Dagbaldes Forening]
[xxxiv] Id.
[xxxv] Aris Teon, AI-Generated Images: Who Owns the Copyright? Landmark Ruling From a Chinese Court, China Journal (Dec.22, 2023), https://china-journal.org/2023/11/29/ai-generated-images-who-owns-the-copyright-landmark-ruling-from-a-chinese-court/.
[xxxvi] Id.
[xxxvii]Rajiv Sharma and Ninad Mittal, Artificial Intelligence Lacks Personhood to Become The Author Of An Intellectual Property, Livelaw (Dec.21, 2023) https://www.livelaw.in/law-firms/law-firm-articles-/artificial-intelligence-intellectual-property-indian-copyright-act-singhania-co-llp-238401.
[xxxviii]Emilia David, Getty lawsuit against Stability AI to go to trial in the UK, The Verge (Jan. 10, 2024) https://www.theverge.com/2023/12/4/23988403/getty-lawsuit-stability-ai-copyright-.
[xxxix] Authors Guild, Inc. v. Google Inc., 828 F.3d 1109 (2d Cir. 2015), cert. denied, 137 S. Ct. 613 (2016).
[xl] Id.
[xli] Cala Coffman, Does the Use of Copyrighted Works to Train AI Qualify as a Fair Use? Copyright Alliance (Jan.10, 2024) https://copyrightalliance.org/copyrighted-works-training-ai-fair-use/.
[xlii] Gregory Barber, The Generative AI Copyright Fight Is Just Getting Started, The Wire (Jan. 10, 2024) https://www.wired.com/story/livewired-generative-ai-copyright/.
[xliii] 1. Purpose and character of the Work.
2. Nature of the Copyrighted work.
3. Amount in sustainability of the portion used in relation to the copyrighted work as a whole.
4. The Impact on the value and market for the copyrighted work.
[xliv]Ariel Soiffer & Aric Jain, Copyright Fair Use Regulatory Approaches in AI Content Generation, Tech Policy. Press (Dec.28, 2023) https://www.techpolicy.press/copyright-fair-use-regulatory-approaches-in-ai-content-generation/.
[xlv] Id.
[xlvi]Artificial intelligence is reaching behind newspaper paywalls, The Economist (Jan.10, 2024) https://www.economist.com/business/2023/03/02/artificial-intelligence-is-reaching-behind-newspaper-paywalls.
[xlvii] Gil Appel, Juliana Neelbauer & David, Generative AI Has an Intellectual Property Problem, Harvard Business Review (Jan.10, 2024), https://hbr.org/2023/04/generative-ai-has-an-intellectual-property-problem.
[xlviii] Stephen Titcombe, Opt-out right under, privacy law, (Jan.10, 2024) Terms Feed https://www.termsfeed.com/blog/opt-out-rights-privacy-laws/.
[xlix] Molly K. Land, Human Rights, and Technology: New Challenges for Justice and Accountability, Annual Review (Jan. 10, 2024), https://www.annualreviews.org/doi/10.1146/annurev-lawsocsci-060220-081955.
[l] Glaze, what is Glaze, Glaze (Jan.10, 2024), https://glaze.cs.uchicago.edu/.
[li] Id.
[lii] Shambawi Sharma & Vivek Kathpalia, Generative AI and its interplay with law, LiveMint (Jan.11, 2024), https://www.livemint.com/opinion/first-person/generative-ai-and-its-interplay-with-law-11693300089628.html.
[liii] Copyright Office, USA, Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, Federal Register (Jan.11, 2024), https://www.federalregister.gov/documents/2023/03/16/2023-05321/copyright-registration-guidance-works-containing-material-generated-by-artificial-intelligence.
[liv] Manish Jindal, Artificial Intelligence and Copyright in India, Bytes Care Blog (Jan.11, 2024), https://bytescare.com/blog/artificial-intelligence-and-copyright-in-india.
[lv] Suzanne Bearne, New AI system collides with Copyright law, BBC (Jan.11, 2024) https://www.bbc.com/news/business-66231268.
[lvi] Proposed Digital India Act, 2023, Ministry of Information and Technology (Jan. 11, 2024), https://www.meity.gov.in/writereaddata/files/DIA_Presentation%2009.03.2023%20Final.pdf.
Comments