This piece has been authored by Amitesh Mishra, Student, 3rd year, Hidayatullah National Law University
Key words: Artificial Intelligence, Machine Learning, Copyright, Ownership, Authorship
INTRODUCTION
In 1927, Metropolis became the first Hollywood film with a depiction of machines having human-like intelligence, but mankind has dreamt of machines with anthropomorphic cognition for decades, if not for millenniums. This dream was manifested in reality for the first time after the [i]Dartmouth conference of 1956, after which the [ii]momentum for the progress of artificial intelligence went unhindered. Contemporarily, Artificial Intelligence has almost identical (if not better in some cases) cognition to that of the human intellect, and with breakthroughs in machine learning models and the emergence of complex neural network models, Artificial intelligence has become capable of performing intricate activities. Artificial intelligence can perform a plethora of activities conventionally idiosyncratic to humans, such as that of generating complex pieces of Art, music, literature, and even images.
Hitherto the ability to create art and literature has been attributed solely to humans, consequentially, the majority of legal frameworks across the globe attribute copyright protection to human authors. Artificial intelligence’s newfound ability of creative content generation provokes the issue of authorship across various jurisdictions, questions with regards to the ownership of the AI-generated content and the attribution of liability in case of an infringement are still matters of murky waters.
This article is an attempt to address the aforementioned ambiguity concerning the authorship of AI-generated content. The article will briefly address the process of content generation by Artificial Intelligence to pinpoint the possibility of an infringement, the article will also analyze whether training of AI is considered a copyright infringement; various plausible models elucidating upon the authorship of the AI-generated content, such as public domain, AI ownership, AI agent, and AI employee would be analyzed, to conclude that it is the person who publishes the content shall be held liable for the infringing content generated by the AI.
ANALYZING THE [iii]BLACK BOX: THE PROCESS OF AI CONTENT GENERATION
The process of AI content generation can be divided into 4 parts namely Input, Learning algorithm, Trained Algorithm, and Output.
Similar to humans, to generate novel content AI relies on pre-existent data sets which are fed into the AI system. For example, a language-based AI such as ChatGPT is subjected to gigantic text sets called Large Language Models to generate content. After such datasets are arranged, [iv]they are subjected to a Learning Algorithm which analyzes the data set for future art generations, these learning algorithms might be Neural Network Models such as a Generative Adversarial Network or Autoencoders. The learning algorithm analyzes and identifies task-specific data patterns, this pattern recognition results in the genesis of a trained algorithm, Generative Pre-trained [v]Transformers algorithms (like [vi]chatGPT) are such trained algorithms. The last stage is generating an output, an AI generates the same based on a stimulus which might be in the form of a prompt or a grammar inserted by the user. The trained algorithm utilizing the data patterns generates the output based on the user-supplied prompt.
Since the AIs are trained on a pre-existent data set, the output generated may sometimes resemble closely the pre-existing copyrighted work, and thus, AI possesses the potential to generate infringing content. If the AI does produce infringing material, the ownership of such work, accountability, and liability belong to legal gray areas.
AI-GENERATED CONTENT: WHO IS THE OWNER
AI as the owner
As per the Indian Copyright Act, 1957, the [vii]Author of a work is the first owner of the copyright. Section 2(d) provides for the definition of an author, as per the same, an author of a work is an [viii]author, [ix]composer, [x]artist, a person taking a photograph, [xi]producer, and a person using a computer to [xii]generate some work. Though the word ‘person’ is explicitly used only in connection with `photograph, cinematograph film, and computer-assisted work, if the same was to be interpreted through the lens of doctrine of [xiii]modicum of creativity as set up in the case of [xiv]Feist Publication Inc v. Rural Telephone Services Co Ltd, it can be conclusively stated that section 2(d) is inapplicable to AI. As to have an original work under [xv]section 13, the work shall have some modicum of creativity to it and contemporarily, the quality of creativity cannot be attributed to AI. Therefore, AI cannot be an author under section 2(d) and to extend the same no owner.
AI as a joint owner
A joint author is a person who collaborates with another author for the production of a work. [xvi]To establish joint authorship there shall be a significant creative input to the work and the contribution should be non-distinctive from the other author. [xvii]To establish joint authorship there shall be an active intellectual collaboration and contribution. AI contemporarily is incapable of providing intellectual collaboration and as established before is not even capable of authorship. Therefore, AI cannot be held as a joint owner of the content it generates.
Internationally, in the celebrated case of [xviii]Naruto v. Slater, the courts in the US have held that a non-anthropological entity shall not be provided with the copyright of any sort.
AI as an employee
A mere plain reading of the proviso clause of section 17 of the copyright act makes it clear that for any work generated by an employee in his course of employment, [xix]the copyright for the same is granted to the employer. The pertinent [xx]factor for the ascertainment of employment is ‘the degree of control’. The relationship between AI and the user displays a high degree of control, the AI does everything precisely in the way as directed by the user, therefore, the relationship between both may be perceived as that of employer-employee.
Though this seems like sound logic, it doesn’t account for other characteristics of an employer-employee relationship. Employer-employee relationships are most often contractual in nature, of which [xxi]consent is the most essential feature. Since AI is incapable of giving consent, it cannot be a part of such relation. Remuneration also forms an essential of an employer-employee relationship, renumerating AI belongs to the pages of a fiction novel. Therefore, AI cannot be an employee and its generated content cannot be attributed to the users.
Corporation creating the AI as the owner
As explained under the head ‘AI as the owner’, it is the author of the actual piece who is granted the first ownership. A corporation might have created an AI but it never engages with the actual piece and therefore, cannot be given its first ownership. As explained above AI being an employee of the corporation is not a possibility.
Public Domain
This probably is the most flawed approach regarding the determination of the ownership of AI-generated content. As per this approach contents generated by the AI shall be put directly in the public domain. This would be disastrous to the copyright structure as even in the case of an infringement, the infringed material would be in the public domain for the public to use freely. By this, we would be putting the copyrighted material in the public domain, even with copyright still intact.
CONCLUSION: NEED FOR NEW MODELS
The first thing that can be conclusively stated at this juncture is that the fate of authorship of content generated by AI is yet undecided. As generative AI proliferates much deeper into our daily lives the ambiguity will haunt us more, as the cases regarding AI generating infringed content will increase. Therefore, the crucial thing to do now is, to devise a model to address the person of liability in case of an infringement.
The models that are prevalent in the intellectual milieu, such as attributing the ownership to AI, to the user, to the corporation owning AI, and putting it in the public domain all have flaws to them. The major problem with all of them is the non-applicability of most laws on AI. Therefore, the best possible solution right now is to attribute it to a person.
What precisely is being suggested here is that the person, whosoever [xxii]publishes the infringed material shall be held liable. It accepted that AI might generate infringing content, but the utilization of the same doesn’t happen immediately after generation, the content is supplied to the user on a user-AI interface, after which it is upon the user to make use of the content.
The user knowing the possibility of infringement shall investigate with regards to the same, and if infringement is found refrain from using the same. If the user chooses to utilize the infringed content, then the user is to be held liable. The corporation shall take appropriate steps to facilitate such investigation.
It is accepted that this also doesn’t address the problem of the ownership, but it at least suggests a route for remedy in case of an infringement, and thus, protects the right of the author, which is the most essential part of any kind of intellectual property right.
[i] The Dartmouth Summer Research Project on Artificial Intelligence was a 1956 summer workshop widely considered to be the founding event of artificial intelligence as a field.
[ii] Stephan De Spiegeleire, Matthijs Maas, Tim Sweijs, ‘WHAT IS ARTIFICIAL INTELLIGENCE’, Hague Centre for Strategic Studies (2017).
[iii] The black box is phenomena wherein only the product of generative AI is visible and the function of the neural networks cannot be accessed.
[iv] This is referred to as Training of an AI.
[v] A transformer model is a neural network that learns context and thus meaning by tracking relationships in sequential data like the words in this sentence.
[vi] T. Y. Wu, S. Z. He, J. P. Liu, S. Q. Sun, K. Liu, Q.-L. Han, and Y. Tang, ‘A brief overview of ChatGPT: The history, status quo and potential future development’ 10 IEEE/CAA J. Autom. Sinica1122–1136 (May 2023).
[vii] The Copyright Act,1957, s 17
[viii] In case of literary work
[ix] In case of musical work
[x] relation to an artistic work other than a photograph
[xi] in relation to a cinematograph film or sound recording, producer is further defined as in Section 2(uu) as in relation to a cinematograph film or sound recording, means a person who takes the initiative and responsibility for making the work.
[xii] This sub-section gives an impression that work of generative AI is covered herein but keeping in mind that AI was inexistent when the act was promulgated, this refers to work generated through computer assistance and not by computer itself.
[xiii] This doctrine stipulates that originality subsists in a work where a sufficient amount of intellectual creativity and judgment has gone into the creation of that work. The standard of creativity need not be high but a minimum level of creativity should be there for copyright protection.
[xiv] Feist Publication Inc v. Rural Telephone Services Co Ltd, 499 U.S. 340, 342 (1991)
[xv] The works in which copyright subsists.
[xvi] Elizabeth Verkey, Jithin Issac, Intellectual Property, (2nded Eastern Book Company 2021).”
[xvii] Najma Heptulla v. Orient Longman Ltd, 499 U.S. 340, 342 (1991)
[xviii] Naruto v. Slater 888 f.3d 418.
[xix] Burlington Home Shoppping Ltd v. Rajnish Chibber, AIR 1989 Del 63.
[xx] University of London Press Ltd v. University of Tutorial Press Ltd, (1916) 2 Ch. 601.
[xxi] Indian Contract Act, 1882, s 10.
[xxii] Whether it be the user, the corporation or any other third party.