Published on: 26th September, 2024
Authored By: Karthika R
His Highness Maharajas Government Law College Ernakulam
Introduction
Creativity was hitherto attributed as a manifestly human quality that knew no bounds. People like writers, artists, musicians, and journalists thrived and generated income based on this quality. Even after the advancement of technology in almost all spheres of life, creativity as a dynamic human quality remained on a pedestal that technology could not effectively mimic. But through the recent developments in Artificial Intelligence, this has been proved wrong. Nowadays, generative AI applications like ChatGpt can create new content in the form of text, images, or audio which typically requires immense knowledge and creativity. In a recent event, an award-winning Japanese writer admits using ChatGpt to write her novel. This points out the potential of Artificial Intelligence in upheaving the creator economy.
Another aspect of the use of Artificial Intelligence in producing creative work is the evident conflict it poses against existing copyright laws. Copyright is an intellectual property right that protects the original literary and artistic works of creators. The conflict of Artificial Intelligence concerning copyright laws is twofold. The first one is with respect to the granting of copyrights for the creative works produced by generative AI models and the second one is the possible copyright infringements that occur during the training of various algorithmic models which uses large amounts of data from across the internet, which include copyrighted works of writers, artists, and journalists. In this article, we are going to explore how copyright law, intersects with Artificial Intelligence by examining the scope and developments in the existing legal framework.
Artificial Intelligence
Technology known as artificial intelligence, or AI, makes it possible for computers and other devices to mimic human intelligence and problem-solving skills.[1] An artificial intelligence (AI) system is a machine-based system that can function at different levels of autonomy, can adapt once it is in place, and can infer how to produce outputs—like recommendations, content, predictions, or decisions—that can affect real-world or virtual environments based on the input it receives. These systems can be used for explicit or implicit purposes[2]. The ability of AI systems to infer is one of its primary features. The ability of AI systems to infer models, algorithms, or both from inputs or data is referred to as this ability to infer. It also describes the process of acquiring the outputs, such as predictions, content, recommendations, or judgements, which can influence real and virtual surroundings. When developing an AI system, inference is made possible by methods such as machine learning, which learns from data how to accomplish certain goals, and logic- and knowledge-based systems, which draw conclusions from encoded information or symbolic representations of the tasks at hand.[3]
A subset of artificial intelligence known as “generative AI” is concerned with how computers may produce text, graphics, code, and synthetic data by using models. One kind of generative AI is LLM. To perform well in language processing tasks, deep learning models known as LLMs use enormous datasets for training and consumption. They use its training data to generate new text combinations that resemble real language.
Copyright and Artificial Intelligence.
Copyright is considered to be a sui generis right which means that a person who is a creator of a thing using his intellect is the prime owner of that thing and has an immediate right over it. Original Work of Authorship (OWA) is the state in which a work is entirely produced without replication by the independent intellect of its creator. Anyone who is the original creator of any work has the automatic right to use, copy, and reproduce it for their own purposes without permission. [4] Copyright covers works of literature, art, and technology, including computer programs and electronic databases, as well as films, books, music, paintings, and sculptures. Copyright is sometimes referred to as writers’ rights in several languages.[5]
The extensive use of Artificial Intelligence for the purpose of producing creative content poses the question, “ Who holds the copyright with respect to the content generated by Artificial Intelligence tools? Another conflict that exists is copyright infringements that happen during the usage of publicly available data for the training of algorithmic models.
Is the granting of copyright possible for AI Generated Content?
Earlier the granting of copyright to computer-generated work was not much of a topic of debate for global Intellectual Property Law enthusiasts as in most such works computer was only used as a mere tool for creating the content. The granting of Copyright becomes a problem when the creative work is completely produced using an AI tool without any human intervention. Because in most legal jurisdictions across the globe, Copyright is only granted to natural persons.
Section 2(d)(vi) of the Copyright Act, 1957 delineates the scope of computer-generated works for copyright for the purpose of which, the person who causes the work to be created shall be considered as the “Author”. Therefore in India, the input and involvement of a person in the process of generating the AI Content is necessitated for the granting of copyright. Recently, the copyright office of India recognized an AI tool named Raghav as the Co-author of an artistic work.
The UK [7]and Irish[8] Copyright laws also take a similar stance wherein the author of the computer-generated work is the person by whom the arrangements for the creation of the work are undertaken. In the famous case of Thaler v. Perlmutter the principal issue was whether, in the U.S., AI-generated content could be granted a copyright. In this case, a person named Stephen Thaler owned a computer system called “ Creativity Machine” which as per the declaration given by Mr. Thaler created a visual art on its own. He approached the US Copyrights office to grant the copyright to him as the owner of the machine by registering the work created by the machine for copyright. In this case, the district court affirmed the decision of the copyright office in rejecting the claim of the petitioner stating that copyright cannot be granted to work created without any human involvement.
In Europe, the Court of Justice European Union (CJEU) in its landmark judgement Infopaq International A/S v Danske Dagblades Forening held that copyright could be only granted to original works that reflect the author’s own intellectual creation thus necessitating the existence of a human author behind the creation of the work. [10]
The Commercial implications due to the non-granting of copyrights.
Granting of Copyright to creative works prohibits the unauthorized use and commercial exploitation of the creative works by anyone other than the original author of the work. The creation of AI tools demands millions of dollars as an investment. When creative content generated by AI is not granted copyright, such works could be used and reused by anyone which would significantly affect the commercial viability of the content, which in turn would create huge loss to the owners of the AI tools using which the content was created. This in turn could dishearten the potential investors in the realm of automated systems. One possible way to mitigate this commercial impact is by granting copyright to the provider who produced the AI model or if the output generated by the AI tool necessitates the use of a skilfully developed prompt, in that case, the creator of the prompt.
Copyright Infringement
When artificial intelligence (AI) systems are exposed to vast volumes of data—which could include text, photos, and other works obtained from the internet—they are “trained” to produce literary, visual, and other artistic works.[11]. The data for training such Large Language Models include personal and non-personal information as well as copyrighted creative works of artists that are scrapped from publicly available sources using text and data mining techniques which may lead to copyright infringement and unfair competition. In general, copyright infringement happens when someone reproduces, distributes, performs, exhibits their work in public, or transforms another person’s creation into a derivative work without that person’s consent.[12]
AI firms may contend that their training procedures meet fair use requirements and are hence non-infringing[13]. J.L v Alphabet Inc [14]a class action was taken against Google for using copyrighted content of various producers for training its AI tools, Bard, Gemini, Imagen, MusicLM, etc. The major consideration that was weighed during this case was whether the ground of fair use can be used to mitigate the copyright infringement associated with data usage for training AI tools. In a similar case in December 2023, the New York Times sued Open AI and Microsoft for the unauthorized use of millions of published articles in the Times for the training of AI models.
Another type of copyright infringement that could occur is with respect to the outputs generated by AI tools which could appear similar to the creative works of other artists. For instance, if given the correct prompt AI could generate creative works similar to that of known writers and artists. In the case of Andersen V. Stability AI [15], Plaintiffs Sarah Andersen, Kelly McKernan, and Karla Ortiz, who are visual artists, sued Stability AI, Midjourney, and DeviantArt alleging infringement of the artists’ copyrighted works in connection with the companies’ generative artificial intelligence systems and products. However, the case was dismissed by the court with leave to amend the artists’ copyright claims stating that the artists failed to prove sufficient similarities between the AI-generated outputs and their copyrighted works.
Many lawsuits are pending across various jurisdictions challenging the copyright infringements caused by generative AI systems. Once these are settled by the courts we can expect more clarity on this issue.
Copyright protection under the EU AI Act.
The EU AI Act is the first comprehensive legislation enacted for the regulation of AI systems. The EU AI Act regulations are majorly characterized by a product-centric risk-based approach where as the AI systems are categorized based on the impact it creates on the fundamental rights of the individuals. Some of the significant provisions included in the AI Act aim at regulating and curbing copyright infringements caused by AI models.
The transparency principle prescribed to be followed by the providers of general-purpose AI models concerning the training data is one such provision. Most of the general-purpose AI models were not hitherto transparent enough as to the data used for the training of their respective AI models. Recital 107 [16]of the EU AI Act mandates to release to the public a summary of all the data, including public and private sources that has been used by the provider for the pre-training and training of general-purpose AI models. The summary thus released would enable the concerned rights holders to know whether or not their copyrighted contents are being used for AI training.
The EU AI Act also mandates the providers to acquire the authorization of the concerned rightsholder prior to the use of any copyright-protected works for the purpose of development and training of their general-purpose AI models. [17]
These provisions enabled through the EU AI Act may help curb the future copyright infringements associated with the AI model training and thereby balance the interests of the rightsholders against the innovative and commercial interests of the AI System providers.
Conclusion
The development of more sophisticated AI models and their widespread use is unpreventable. But what could be done is the building of robust legal frameworks to address the potential risks associated with it. At the same time, it is pertinent to encourage innovation and protect the commercial interests of the AI system providers who have invested millions in the development of cutting-edge AI tools. Although many jurisdictions do not grant copyright to creative content solely by AI systems, granting copyrights to the person who enabled the operation of the AI system can be considered a rational approach to be taken by authorities. As the demand for huge data sets for the purpose of training AI models is increasing, being transparent and reliable concerning the data set used for training, can enable AI system providers to escape the challenges emanating from Copyright infringements. Let’s hope that more jurisdictions would come up with regulations for AI systems, catering to the interests of the AI system providers and the original authors and thereby balance innovation and creativity.
References
[1] ‘What is Artificial Intelligence?’ IBM https://www.ibm.com/topics/artificial-intelligence accessed August 8, 2024
[2] EU AI Act 2024 Article 3
[3] EU AI Act 2024 Recital 12
[4] Mariya Paliwala,’ What is Copyrights under Intellectual Property Rights (2019) Ipleaders https://blog.ipleaders.in/copyright-intellectual-property-rights/ accessed August 3, 2024
[5] ‘Understanding Copyrights and Related Rights’ WIPO https://www.wipo.int/edocs/pubdocs/en/wipo_pub_909_2016.pdf accessed Aug 4, 2024
[7] United Kingdom Copyright, Designs and Patents Act 1988, s.9(3)
[8] Irish Copyright and Related Rights Act, 2000, s.21
[10] Infopaq International A/S v Danske Dagblades Forening [2009] ECJ 1
[11] ‘Generative Artificial Intelligence and Copyright Law’ (September 29, 2023) Congressional Research Service https://crsreports.congress.gov/product/pdf/LSB/LSB10922 accessed August 4, 2024
[12] US Copyright Office https://www.copyright.gov/help/faq/faq-definitions.html#:~:text=As%20a%20general%20matter%2C%20copyright,permission%20of%20the%20copyright%20owner. Accessed August 4 2024
[13] ‘Generative Artificial Intelligence and Copyright Law’ (September 29, 2023) Congressional Research Service https://crsreports.congress.gov/product/pdf/LSB/LSB10922 accessed August 4, 2024
[14] J. L. v. Alphabet Inc., 3:23-cv-03440, (N.D. Cal.)
[15] Andersen v. Stability AI Ltd., 23-cv-00201-WHO (N.D. Cal. Oct. 30, 2023)
[16] EU AI Act Recital 107
[17] EU AI ACT Recital 105
