By- Varsha Sharma
INTRODUCTION
The proliferation of Artificial Intelligence (hereinafter ‘AI’) in varied fields has increased over the years ranging from its use in medical sciences, technology, transportation, journalism and so on. The 2019 Report of the World Intellectual Property Organization also indicated the same. It was suggested by Niti Aayog in their conversation paper of 2018 that the usage of AI had been increasing in all areas, such as schooling, medical care and so on.
The research questions explored in the article are: (i) Is it beneficial to make a difference between computational creativity and human creativity while analysing AI work? (ii) What is the scope of granting legal personhood to AI work? (iii) Considering there are various candidates who could claim copyright in AI work, how should the tussle be solved? (iv) Who should probably be conferred liability for damage or be held responsible for copyright infringement?
In the end, a proposal is suggested to make a distinction between an AI –made works and AI-suggested works, along some amendments to the Copyright Act, 1957.
WHAT IS ARTIFICIAL INTELLIGENCE?
The term artificial intelligence was first coined by McCarthy in the year 1956. Over the years, no legal definition per se has been ascribed to the use of the term, but broadly it is being perceived as the ability of machines to perform work requiring human intelligence. It involves the ability of the machines to perform tasks involving the use of biological cognition involving learning, problem-solving, decision-making and so on. The three categories of Artificial Intelligence as identified by the World Intellectual Property Organization include (i) expert (knowledge-based) system, (ii) perception system (iii) natural language system.
For the purpose of this paper, it is sufficient to understand the use of AI as a system of originating original work with minimal human intervention.
UNDERSTANDING THE REQUIREMENTS OF COPYRIGHTABILITY IN THE INDIAN CONTEXT
Copyright is a legal term which is used to describe the rights which a creator has over their artistic or literary work, which might range from books, paintings, advertisements, technical drawings, maps etc.
The twin- requirement for the attainment of copyright includes (i) originality and (ii) fixation in a tangible medium of expression.
What Entails Originality?
The word originality does not necessarily require that the work must be a product of inventive or original thought. In this regard, in the UK, the test followed was “sweat of the brow” as used in the case of University of London Press Ltd. v. University Tutorial Press Ltd, where the court said that the word literary work covered works expressed in print or writing, irrespective of whether the style or the quality of the work was high. The doctrine was further used in the case of Walter vs Lane and Ladbroke Limited vs William Hill where the court said that it was immaterial whether the work had any literary merit and reiterated the requirement of “labour, skill and judgement” and the requirement that the work should originate from the author.
In USA, the courts have adopted a broader outlook towards what entails copyright, thus not just limiting it to ‘labour, skill and judgement’ but including a certain level of creativity. Thus, in the case of Fiest Publications Inc. vs Rural Telephone Service, the United States Supreme Court negated the concept of labour, skill and judgement. The court held that in order for a work to be able to get copyright, it should not only be independently created but also showcase a minimum level of creativity. Thus, the aspect of ‘creative originality’ was propounded in the case in order to protect the work with minimum creativity. Here, the standard of creativity required is not high, but a minimum level of creativity is considered sufficient.
The Indian Supreme Court rejected the approach in the UK and has instead adopted the approach as taken in the case of Fiest Publications. In the case of Eastern Book Company vs DB Modak, the court shifted to an approach of ‘Modicum of Creativity’, thus introducing the idea of ‘flavour of minimum requirement of creativity’. The court held that in order to establish copyright, the creativity standard does not require involving something which is novel or non-obvious, rather it shall mean to include ‘some amount of creativity’ so as to claim copyright.
Further, a very key element in terms of copyright is ‘idea/ expression dichotomy’, i.e., an idea is not copyrightable, only the expression of the idea is copyrightable.
IMPORTANT QUESTIONS TO BE ANSWERED WHILE CONSIDERING GRANTING OF COPYRIGHT TO AI
Requirement of Human Element in AI
A common argument for denying copyrightability to AI is that it does not involve the human element. It shall be analysed to what extent this argument holds true.
In the case of Naruto vs Slater (known as the monkey-selfie case), the question before the Ninth Circuit Court of Appeals was whether the animals could be regarded as the author of the work created by them. In this case, a seven-year-old Macaque (Naruto) clicked photographs by using the camera of the respondent. The respondent published a book encompassing the photographs clicked by Naruto. Post this, the People for Ethical Treatment of Animals (PETA) moved to the Court against Slater, claiming copyright infringement. The Ninth Circuit held that Naruto lacked the required understanding in the case.
From this, it is clear that the requirement of a human element is required in order to grant copyright to a work.
Difference between Human Creativity and Computational Creativity, “Is there a human element in AI.”
Looking at AI in the above context, firstly, it becomes imperative to differentiate between human creativity and computational creativity. At present many AI, processes mimic human capacities. Therefore, it is crucial to understand what the state of the art in AI is and whether AI could be considered to demonstrate the role played by humans in the context of creative machines and creative output.
In this regard, it must be seen that even though AI involves very less human intervention and uses its own framework for the purpose of generating the output, the same would not have been possible if the process wasn’t initiated by a human being from the starting till the end, i.e., beginning from the creating various parts of the machine, programming, using, training the system and so on. Thus, it can be seen that there is a presence of human element at various stages in the production of final output by the AI system.
Further, Peter Norwig and Stuart Russel, two prominent scholars in the field of AI, attempted to define AI in the following approaches: (i) The Turing Test Approach, (ii) The cognitive Modelling Approach (iii) The laws of the Thought approach (iv) The rational agent approach
The first approach, i.e., the ‘Turing Test’ approach was formulated by Alan Turing in the year 1950. Turing was an English computer scientist, mathematician, theorist, biologist and a cryptanalyst. The Turing Test approach is a technique of inspection in order to decide whether a machine is capable of mimicking human intelligence or not.
The second approach i.e., the ‘cognitive modelling approach’ involves a new approach so as to emulate the computation occurring within the human brain so as to discover new knowledge. On the basis of this phenomenon, an intelligent system known as the Knowledge Growing System was developed. This approach is the basis for the purpose of determining an agent which has the ability to act and think rationally similar to a human being, known as the cognitive agent.
The third approach towards AI is ‘the laws of the thought approach’. Thinking rationally is again an element attributed to human beings that is human beings would act rationally. According to the book, “Artificial Intelligence a Modern Approach”, the “right-thinking reasoning process was first codified by the Greek Philosopher Aristotle. Aristotle in his syllogisms, provided patterns for argument structure which always made correct premises. For instance, Socrates is a man, and that all men are mortal would lead to a conclusion that Socrates is mortal.
The fourth approach to ‘AI is Acting Rationally: Rational Agent Approach.’ Acting rationally implies acting in order to achieve one’s goals, in pursuance of one’s beliefs system. An agent is someone who perceives and acts. Under this approach, AI is considered a construction and study of rational agents. In the third approach, i.e., the laws of the thought approach, to AI, the entire emphasis was on making correct inferences. Being able to make correct inferences is sometimes considered as part of the rational agent approach, because one way to act rationally is to reason one’s action so as to be able to achieve a particular conclusion. However, making correct inferences is not the only element which encompasses rationality, as there are situations where there is no proven correct thing as such which should be done however something must be done. Further, there are ways of acting rationally which might not necessarily involve inference, for example, pulling off one’s hand from a hot stove, which is a reflex action and which might be more correct as compared to a deliberate thought of action in such a circumstance.
A common factor which can be inferred in all the approaches to AI is their requirement to act akin to a human being either by mimicking them, or acting rationally in a similar way to humans. Thus, it can be concluded that AI performs function similar to human beings, thereby broadening the scope for AI to be considered as qualifying the requirement of human element for a copyright claim, and there might not be much scope to make a differentiation between human creativity and computational creativity.
CAN AI BE ACCORDED LEGAL PERSONHOOD?
While asserting legal personhood to AI, one possibility is that the AI system is treated as a trustee. This draws from the discussion of the idea of legal personhood in the work of Chipman’s Gray, “The Nature and the Sources of Law.” He talked about the attribution of legal rights and duties while talking about “legal personhood”, instead of the person being considered synonymous to human beings. Thus, the question of whether an entity should be granted legal personhood or not should ultimately rest on the fact whether it should or it should not be granted legal rights and duties.
In the context of AI this will be analysed under two prongs- (i) Whether Artificial Intelligence could serve as a trustee? (ii) Does it qualify the moral grounds to be ascertained with legal rights and duties?
Moving on with the first inquiry, here, two scenarios shall be looked into, (i) the first being what would be the legal consequences of developing a system which possess the capability of performing things a human person can. For instance, developing a program which is capable of investing in stock.
Such a system could be further subdivided into further stages. The first is, the program is acting as an aid to a human trustee for the administration of work. For instance, the program places investment order through electronic or modem mail, investing in public traded securities. Post this, via an electronic checking program the program disburses the funds to the beneficiaries of the trust. In an instance where the program is intimated of an event, such as, the death of the beneficiary, the program would follow the instruction as provided in the trust agreement, which might be terminating the trust or a change of beneficiary. Apart from this, the program also has the capacity to file the ITR returns, but, the ultimate decision of investing lies with the human trustee. The human trustee also reviews the activities of the program so as to ensure that the terms as provided under the agreement of the trust are satisfied.
The second stage might involve a scenario wherein there is a greater role accorded to AI. This involves the development of systems which actually outperform the humans. This includes a scenario where the ultimate decision with respect to investing lies with the program. This might be a case scenario where the administration program of the trust is really well proficient in analysing and then implementing the terms of the instruments. Here, it can be seen that AI system has a much greater role and responsibility, as compared to stage 1.
Thus, there might be a requirement to make a distinction between AI made systems and AI- supported systems such as in explained in the above two stages, and accordingly their legal position with respect to legal personhood might vary.
The second scenario consists of an enquiry whether AI system justifies a claim of moral rights or they are mere property. Section 57 of the Indian Copyrights Act provides for the moral rights to the author. According to it, an author shall have the right to claim authorship over their work even after the rights have been assigned to someone else, and further to claim damages with respect to any mutilation, distortion or modification of the work. A reference should be made to the case of Amarnath Sehgal vs Union of India, where the mural created by an artist and sculptor was torn down, there the Delhi High Court had held that the moral rights of the artist are soul of his works, and thus the author has a right to nurture, protect and preserve his creations using his moral rights.
Now, the question here to consider is whether AI can be said to have moral rights? Whether it can be said to have a personality? According to the theory given by Kant, “all rational beings and not just humans are persons.” Here, one might argue that the theory given by Kant lacks the idea of consciousness, intention and emotion, however these claims are highly contentious. Further, there exists the idea of AI experiencing human emotion, for instance if emotion is a facet of human mentality, then it might be fed into a computational model resulting in a computational process. As argued by Araon Salmon a control system is required for a system having multiple goals, and human emotion is one such system.
Hence, the idea that AI lacks human element or it cannot understand human elements such as emotions etc. could be negated to a significant extent, therefore making a scope for granting them legal personhood.
ANALYSING THE DIFFERENT CANDIDATES/ STAKEHOLDERS FOR AUTHORSHIP IN AI WORK
When it comes to AI different candidates/stakeholders could be considered to be qualifying the requirements of copyrightability ranging from the programmer, trainer, user, data proprietor and so on. Thus, it shall be analysed as to what extent they qualify the requirements of ownership.
Programmer
In AI the major contribution with respect to the usage of the AI system comes from the programmer. One of the main arguments for granting ownership in AI to programmers is that if the programmer, had not trained, designed or created the user software, the work would not have come into existence at all. Thereby creating an independent, creative and sophisticated system consisting of mentally challenging processes, involving significant costs, and hence the same should be awarded.
However, as also argued by Pamela Samuelson, one of the possible reasons why such approach should not be taken is that the authorship of all the AI generated work should not be given to the programmer, as the user in such a case might not even disclose the end-product. Further, the incentive of the user for the production of different outputs from the AI system would go down. Thus, a similar approach to the first sale doctrine should be adopted.
It is further contended by Samuelson that granting authorship of all the works produced by the AI to the programmer is problematic with respect to enforceability as it would result in user being incentivised not to report the output to the programmer, and it would become difficult to determine whether a particular work was created by even using the AI system. Therefore, once the programmer decides to exploit the AI system by distributing it to the third parties, it shall no longer lay a claim over the output.
Further, since the idea of copyrightability rests on the idea/ expression dichotomy, it is contended that the programmer is not the person who is responsible for the actual output generation. Therefore granting ownership to programmers in every case, would be in certain sense a departure from the legal doctrine.
In this regard the case of Burrow Gills might be referred to, where the lightning, location and pose of the subjects were considered by the Court. In the case the court had recognised the author on the ground that when the photographic picture was being taken, she was the one closest, as she had made the arrangements and thereafter, the designing of the picture was done by her, as she had decided the position of the subjects and the location of the shoot. It was declared by the court that the photographer’s author should be the one who generates, represents or embodies the idea, or imagines or fancies the ideas associated with it.
In AI, the programmer could be considered to be in an analogous position to the photographer, in terms of setting the framework and the subsequent output.
Therefore, programmer on the basis of the above reasoning might not be best suited person for the purpose of granting ownership, especially where user has attributed significant creativity in terms of arrangement etc. at its own end.
Trainer
The trainer is the person responsible for training the neural network. In certain situations, the programmer and the trainer might be one and the same person.
Firstly, after the network is introduced to a data for developing an output, it requires training. Here, training is extremely crucial, if the wrong data is selected for the purpose of training it would result in deviation from the output, and would result in a failure of the system.
Machine learning depends on the quality and the amount of input data by the trainer. Therefore, the trainer does qualify for a competing claim towards ownership of AI. In most situations the programmer and the trainer, are the same. However, they might be different as well.
In this context another possibility which exists is that of work for hire. In such a situation the company who had hired a trainer for designing the system, the company should have the ownership and the liability in cases of infringement.
Further, even though the trainer satisfies a good claim for authorship, but again if the user had attributed sufficient creativity, skill and judgment at its end, the trainer might not be the best person to have the authorship of the product.
User
As argued by Samuelson, the user of a computer-generated work selects, arranges, polishes, edits the raw material, into a fixed product which could be exploited commercially. Further, in certain situations the user might even have more knowledge as compared to a programmer.
However, there might be situations where the role of the user is almost as minimal as clicking of a button, and further a more sophisticated AI would have very less or almost minimal human intervention. In such a situation the user having the ownership and the user being made responsible for damage or infringement might not be the correct solution. Hence, the programmer/ trainer might be given authorship in such a situation where the end product rests on sufficient creativity at the user-end.
In this regard, a differentiation might be between AI made work and AI supported work. This would also ensure that none of the stakeholders involved in the process performs their tasks negligently and no one is accorded responsibility unnecessarily.
The AI System
Another possibility of according ownership in the case of AI is the AI system itself. However, before doing the same, one needs to look into the factor whether legal personality could be accorded to the system.
In this context, it must be noted that various entities have been accorded legal personhood over the years. A common example of this is company where the company is being accorded a separate legal entity irrespective of the people who runs or owns it.
Further, what could be included within the ambit of legal personhood has been evolving over the years, which is often based on the public reasoning and suasion. For example, Africans and American women were earlier excluded from being granted legal recognition, but were recognised as legal persons over time. Thus, as also argued by Lawrence Solum if over the years the public considers it reasonable the AI could be granted legal personality based on cognitive sign which affirms that operation of AI system is similar to that of human mind.
Further, what is granted legal personhood is often based on political, legal and moral considerations, thereby implying that law is not always just a simple reflection of the reasoning process. Therefore, accepting artificial intelligence as legal person is a decision which requires practical considerations, including societal impacts.
Another consideration for granting authorship to AI itself is the “work-for-hire” model, which could be applied in the case of an AI generated work.
In the case of Community for Creative Non-Violence vs Reid, it was held that a contractor who is working in his independent capacity should not be considered an “employee” under the Indian Copyright Law. Thus, various factors identified by the Supreme Court as constituting employer-employee relationship included: (i) control exerted by the employer over the work (ii) control exerted by the employer over the employee (iii) the status and conduct of the employer. In the case of Marco vs Accent Publishing Company, the court held that a freelancer photographer was an independent contractor. Important factors taken into consideration while deciding whether a work is ‘work for hire’ or not includes ‘local custom, the occupation of the hired party, and the control which the person had over the details of the work.
In the context of AI, the work-for-hire model would not necessarily work except where there is an express contract for the same.
Joint Authorship
A common suggestion for allocating authorship in cases of AI is joint authorship between the user, the programmer, and the AI.
In this context, the meaning of joint authorship in the Copyright Act, 1957 must be looked at. Section 2(z) defines joint authorship, as the work which is produced in joint collaboration between two or more authors, wherein the contribution of the authors is not distinct in nature. In this regard, a famous case of Najma Heptulla vs M/s Orient Longman Ltd, might be referred to. The issue before the court was the authorship of the book India Wins Freedom. In the book the experiences of Maulana Abul Kalam Azad, were noted by Humayun Kabir. Kabir had prepared a draft in English and given it to Azad to record his feelings. Kabir’s function was to record Azad’s feelings. Thus, Azad communicated his thought to Kabir in Urdu and the latter strived to note it down in English.
The court in Najma did not follow the approach taken in the case of Donoghue vs Allied Newspapers. In the latter case, Donoghue supplied the experiences of his life to a journalist, Felstead. The court held that the person who is responsible for clothing the idea, whether by means of a book or a picture owns the copyright, and thus Donoghue was not given the copyright.
However, in the case of Najma Heptulla, court held that the book was the product of the “close intellectual collaboration” between Azad and Kabir, and the work could not be considered as the sole work of either Kabir or Azad.
In the case of Artificial Intelligence even though work done by one stakeholder might be based on the work done by other, there is not a necessary intellectual collaboration between the different candidates. For instance, assuming there is an AI system made by a programmer/ trainer. Now, the user might make various arrangements such as arranging, filing, editing and so on, on which the output would depend. Thus, even though the output was dependent on add-on value by each party, there isn’t a back-forth collaboration between the parties so as to qualify the requirement of joint authorship.
Public Domain
Another argument regarding granting authorship in AI is that the computers do not require to be incentivised in order to create the work, and the work resulting from AI should be placed within the public domain, as doing so would be beneficial for the purpose of science, thereby proposing a kind of utilitarian approach.
In this approach, it is wrongly assumed that the AI systems could generate work out-of nothing (ex-nihilo). Thus, even though the AI system do not require any incentive as such, the human elements linked to it such as the programmer/ user etc. definitely require incentive in order to produce the work.
INTRODUCING AI INTO THE STATUTE: POSSIBLE AMENDMENTS
According to section 16 of The Copyright Act, 1957 copyright should not be granted unless the same is provided under the Act, even though a claim for breach of confidence or trust might be made. Based on this, if AI is accorded copyright, then it must be specifically included within the text of the statute. Some of the suggestions in the form of amendments to be made in the Copyright Act: The phrases which should be added in different sections are:
In Chapter 1, in section 2(d), the following phrase should be added, ‘in relation to any AI work, the work made by the programmer/user.’ Under section 2(m) infringing copy with respect to an AI work should also include, ‘the copying of the specific programming set within the AI system/ production of an output by a user which is a copy of an earlier work.’ Section 14(a), should also include, in the case of an AI work, ‘to reproduce a similar work by any means/ to make any translation/ adoption of the work made using AI/ to reproduce the work made by AI in any material form/ to issue copy of the work made by AI to others in circulation/ to perform the work produced by AI in public or communicating it to the public.’ In section 17 (a), ‘in the case of literary, dramatic, artistic work, AI work made by the author ….’
The amendments suggested would possibly include AI within the framework of The Copyrights Act, 1957.
Further, some discretion should be left to the judicial authorities in order to decide whether in a particular claim the user/ programmer should be granted copyright or not based on a particular factual scenario.
Similarly, who should be held liable for copyright infringement should also be a fact-based inquiry based on the level of contribution in terms of creativity exerted. With respect to conferring damage for responsibility there might be few possibilities such joint liability or who had the knowledge/possibility/last chance to prevent the accident. For instance in 1981, in Japan, an incident occurred where the robot had tracked down the worker as a danger towards its main goal and drove him to a working machine nearby so as to dispose of the danger. In such an incident it might be relevant to look at whether the incident could have been avoided by taking precautionary measures by the user, or whether such an incident could have been predicted or not by the programmer, or in some cases of damage the responsibility might be required to be conferred on both the programmer/ trainer and the user, where the liability for damage cannot be attributed to any one of them particularly.
CONCLUSION:
Firstly, it was shown that AI work cannot be said to necessarily exclude the presence of ‘human element’ which is an essential requirement for claiming copyright. Further, an argument was made that it might not be necessary to make a differentiation between human creativity and computational creativity, as the same might not provide a useful inquiry in ascertaining whether AI work should be granted copyright or not. Further, an argument was made that AI work might be granted legal personhood in future on the basis of its perception in society, and it might not be said to necessarily exclude elements such as moral rights and a personality of its own. Further, various candidates for granting copyright over AI work were looked into and it was concluded that the answer in each case should be rather a fact-based inquiry based on the level of creativity exerted by each candidate at every stage, which also goes in line with the requirement of “flavour of minimum requirement of creativity” as put forth by Supreme Court in the case of DB Modak. Similarly, it was suggested the conferring of responsibility for copyright infringement should also be a fact-based inquiry. Along with that some tests to be used in cases of damage caused by an AI work were also proposed. Lastly, few amendments were suggested which could be incorporated into the Copyright Act, 1957 and which would bring the AI work within the ambit of copyright law in India.
(Varshsa Sharma is a law undergraduate at O.P. Jindal Global University, Haryana. The author may be contacted via email at 19jgls-varsha.s@jgu.edu.in).
Cite as: Varsha Sharma, ‘Interaction Between Intellectual Property Rights and Artificial Intelligence: A Tussle Between Completing Claims for Copyright in AI Work’ (The RMLNLU Law Review Blog, 5 May 2023) <https://rmlnlulawreview.com/interaction-between-intellectual-property-rights-and-artificial-intelligence-a-tussle-between-completing-claims-for-copyright-in-ai-work/>date of access.
The RMLNLU Law Review Blog 