Right on the heels of Vedika’s earlier post, we are pleased to bring to you this guest post by Dr. Anson C J taking an in-depth look into the question of whether an AI-generated work is a “work” under the copyright law. Dr. Anson is an Assistant Professor at the Inter University Centre for IPR Studies, Cochin University of Science and Technology, Kochi.
‘AI Generated Work’, ‘Computer Generated’ and ‘Work’ in Copyright: Whether AI Generated Work is a ‘Work’?
by Dr. Anson C J
As technology advances, the emergence of synthetic creativity has raised intricate questions about copyright law and the concept of originality (Samuelson 2023). This post delves into the intersection of synthetic creativity and copyright, specifically focusing on the complex issue of establishing originality in creative works generated by artificial intelligence (AI) and other computational methods especially the question of whether an AI generated work is a ‘work’ under the Indian copyright system. In Europe, the Court of Justice of the European Union (CJEU) has affirmed multiple times, notably in its significant Infopaq ruling (C-5/08 Infopaq International A/S v Danske Dagbaldes Forening), that copyright is applicable exclusively to original creations. The concept of originality implies that a work must emanate from the “intellectual creation” of the author. Essentially, this entails that for a work to be eligible for copyright, it must bear the imprint of the author’s individuality, emphasizing the requisite involvement of a human author.
AI Generated ‘Work’ or Software Outcome?
AI generated works are not works as such, rather they are the outcomes of maximizing the likelihood of the prompted text or pixel in the command. They can be explained as the software outcomes rather than expressions of intellectual creativity to present something in the digital form. In other words the permutation and combinations of the prompt is controlled by the maximum likelihood algorithm of the next pixel, character, or byte determined by the highest predicted likelihood of the complex algorithm of the searches. Helena Vasconcelos (et. al.) suggest that highlighting tokens with the highest predicted likelihood of being edited by a programmer leads to faster task completion and more targeted edits. Greta R Bauer , Daniel J Lizotte discuss the prediction of action outcomes in robotics using semantic augmentation and physical simulation, aiming to enhance failure tolerance. These all reveals that AI generated works are outcomes of the programmed algorithms based on prompt, tokens and highest maximum likelihood. The above articles shows that technically AI generated works are software outcomes and their creative aspect relies on maximizing the likelihood of the prompt with pixel, text, byte etc. So this cannot be treated as ‘work’ and it is a synthetic creativity. For understanding, there are similarities in Google search and AI outcomes as major result in Google search gives ‘results’ whereas AI gives composite ‘result’ using maximum likelihood of the unsurpassed prompted query.
| Feature | Google Search | AI Results |
| Type of results | Links to websites | Maximum likelihood algorithm Generated content |
| Accuracy | Generally high, but can vary depending on the query | Can be high, but is still under development |
| Completeness | Comprehensive, but may not include all relevant information | Can be incomplete, but may include new and unique information |
| Personalization | Can be personalized based on user search history and preferences | Not yet widely personalized |
| Creativity | Can generate some creative results, such as lists and tables | Can generate more creative results, such as poems, code, scripts, and musical pieces |
| Use cases | Find information on the web, research topics, answer questions | Find information, research topics, answer questions, generate new content |
Overall, Google Search is a more reliable and comprehensive way to find information on the web. However, AI outcomes are becoming increasingly sophisticated and can be a valuable tool for generating new content and finding unique insights from its.
Computer Generated Works and Copyright
In this work, Prof. Pamela Samuelson discusses the legal challenges posed by the emergence of generative AI, or AI that can create original content. She argues that the current copyright law is not well-suited to address these challenges, and that it is in need of reform. One of the key challenges is determining whether generative AI-created works are eligible for copyright protection. Samuelson argues that, under current law, they are not. This is because copyright law requires that works be created by human authors, and AI is not considered to be a human author. Samuelson argues that this is a complex question, and that it will likely need to be decided on a case-by-case basis. For example, if a generative AI program is created by a team of people, then it is possible that all of the team members would be co-owners of the copyright to any works created by the program.
In their work, Senja Assinen concludes that AI-generated works are not currently protected under European Union copyright law, as they are not considered original works of authorship. Yurii Burylo argues that copyright law does not protect AI-generated works in most countries, but some countries have legislation that grants copyright to those who arrange for the AI system to produce the works. Yong Wan and Hongxuyang Lu examine the experience of China and highlights that some AI-generated outputs are eligible for copyright protection in China, depending on the specific circumstances. In her work, Jessica Gillotte focuses on the copyright infringement issues arising from AI-generated artwork and argues that under current copyright law, engineers may use copyrighted works to train AI programs without infringing copyright. Overall, these papers indicate that there is a need for further discussion and development of copyright laws to address the unique challenges posed by AI-generated works.
AI Generated Works and Indian Copyright Act of 1957
According to the Copyright Act of 1957 in India, copyright protection is granted to original works that are created by humans. This leaves a grey area when it comes to works generated solely by AI algorithms. In the context of copyright laws, AI-generated work poses a unique challenge in determining authorship and ownership. In the European Union, AI-generated work is generally considered to belong to the human creator if there is sufficient human oversight in the creation process. However, the definition of what constitutes “sufficient” human oversight remains ambiguous. The authorship of AI-generated work in India is not explicitly addressed in the Copyright Act of 1957. The Act defines the following
(o) “literary work” includes computer programmes, tables and compilations including computer databases;
In this section though computer program is regarded as a work, an AI outcome is not a program and would fall outside the scope of the provision.
(y) “work” means any of the following works, namely:—
(i) a literary, dramatic, musical or artistic work;
(ii) a cinematograph film;
(iii) a sound recording;
Considering AI outcome is a result of a software outcome or maximum likelihood result of the composite search based on prompt thus, it cannot be treated as a ‘work’ or “works”.
(z) “work of joint authorship” means a work produced by the collaboration of two or more authors in which the contribution of one author is not distinct from the contribution of the other author or authors; (za) “work of sculpture” includes casts and models.
And in author :-
(d) “author” means, —
(i) in relation to a literary or dramatic work, the author of the work;
(ii) in relation to a musical work, the composer;
(iii) in relation to an artistic work other than a photograph, the artist;
(iv) in relation to a photograph, the person taking the photograph;
(v) in relation to a cinematograph film or sound recording, the producer; and
(vi) in relation to any literary, dramatic, musical or artistic work which is computer-generated, the person who causes the work to be created;
This is the only place where the Act is using the word ‘computer generated’. The issue of the ‘Author’ can be solved legally or the existing copyright law permits that but the concept of work technically and its parameters to fulfil as ‘work’ in copyright legal regime is complex.
Conclusion
It must also be noted that the recent decision in LI v. LIU by the Beijing Internet Court appears to diverge from recent U.S. rulings on the copyrightability of Artificial Intelligence-Generated Content (AIGC) output, exemplified by cases such as “Zarya of the Dawn,” “A Recent Entrance to Paradise,” and “Theatre D’opera Spatial.” Notably, both the U.S. Copyright Office and U.S. courts in these instances have denied copyright protection to AIGC outputs lacking direct human authorship.
However, the disparity between the Chinese case and the U.S. decisions does not arise from the belief that non-humans can be considered “authors” or from an absence of a requirement for “human authorship” in copyrightable works within Chinese copyright law. Instead, the Beijing Internet Court, in LI v. LIU, seems to draw a distinction between two scenarios involving AIGC output.
The technical background of the Artificial Intelligence is very complex and it generates a convincible outcome largely resembling the functioning of the human brain . Searching for the maximum likelihood of the prompt and matching it with the existing results based on permutation and combinations are at the backend of AI generated outcomes. This gives rise to complex questions like- can these outcomes be treated as ‘works’ or are they ‘software outcomes’? The assessment of originality based on human intellect using prompts involving pixel, character, byte, etc., is a complex task. It’s challenging to determine conclusively whether this method is sufficient for estimating originality. The effectiveness of this approach depends on various factors, including the nature and diversity of prompts. Considering both positive and negative prompts is essential to comprehensively evaluate the presence of ideas and originality in responses. The number and quality of prompts play a significant role in this assessment. The complexity of the issue suggests that a conclusive determination may not be possible by focusing solely on technical aspects or delving into the layers of AI involved. A holistic approach that considers both human intellect and the AIGC framework is necessary for a comprehensive solution.
Acknowledgments
I would like to extend my thanks to Prof. Arul George Scaria, Prof. N. S Gopalakrishnan, and Mr. Jagdish Sagar. The three were the key speakers of the session on ‘Artificial Intelligence and Copyright’ which gave me clarity on the central issue of this post. The session was a part of the Round Table discussion conducted by IUCIPRS, CUSAT in the memory of Valasalakutty Ma’am on August 5, 2023.