This article discusses the impact of Generative AI on intellectual property rights, focusing on challenges in copyright ownership and the need for updated regulations.
This article is written by Namah Bose a 5th Year student of Rajiv Gandhi National University of Law
By now, students, academicians and professionals all over the world have heard of ChatGPT. ChatGPT is one of the many platforms/tools providing Generative AI services. Generative AI is capable of creating content like images, essays, emails, audio based on prompts given by a human.
It creates this content through its language model which works on a process called
DeepLearning. DeepLearning helps produce human-like text through a neural network which predicts the next word and sentence based on its training. However, the ability to generate content depends on the content on which such language model is trained. The content on which such systems are trained include large data archives.
While the technical aspect of Generative AI tools is developing everyday, the content based on which such outputs are produced as well as the outputs itself run into intellectual property rights (IP Rights) issues. There are two major IP concerns arising out of Generative AI - One, that generative AI affects the copyright of third parties whose content has been used in the training process of the AI tool. Second, whether AI generated content can be afforded copyright protection and if yes, who would be considered as the owner of the output of AI? The big question with AI generated content is who is considered the creator of such content?
Generative AI helps thousands complete college essays, write articles, create animations for companies, thus, logically invoking the intellectual property (IP) challenge regarding ownership. Traditionally, the person who created the content owned the intellectual property rights of their creation. However, with generative AI, is it the company that owns the IP or the person who used the AI tool to generate content? Further, sometimes AI generates content excessively similar to an already existing piece of work, thus, affecting third party rights.
There is no simple answer to the question of who owns the output of Generative AI. Different jurisdictions approach the issue differently. Certain jurisdictions require the involvement of human creativity for the work to qualify for copyright protection. The Indian regime identifies only humans as capable of creating work.
Sec. 2 (d) (vi) of the Copyright Act, states that in the case of computer generated work the person who created the same would be considered the owner. It does not however, account for works created completely through Artificial Intelligence and only identifies humans under the definition of ‘Author’ in Sec. 2 (d). Indian laws have not explicitly recognised AI as an author. Further, as per copyright laws, the ‘work’ created has to be an original creation and since AI uses data to create its output, the ‘work’ may not meet the requirement of originality under the laws. While Indian laws do not have a clear answer to the question of who owns copyright of the content generated by AI, OpenAI (the company that owns (ChatGPT) has given an answer within their policy which transfers copyright to the person who has given the specific prompt. Thus, as far as ChatGPT is concerned, the person generating the output owns the content.
JURISDICTIONAL DIFFERENCES IN APPROACH TOWARDS IPR PROTECTION FOR AI
GENERATED CONTENT
The legal position internationally is varied and dependent on the laws within the jurisdiction of which the dispute lies, as well as the policies of the company owning the generative tool. The UK position is rooted in the importance of human creativity, and does not afford copyright ownership to an AI tool. The United States Copyright office also exclusively considers ‘content created by humans’ for copyright protection. However, if it is decided that certain AI generated content can be afforded copyright protection, the question of ownership follows.
Certain arguments can be put forward to analyze the probable position to be taken by the judiciary. In 1884, landmark case Burrow Giles Lithographic company v Sarony stated that the photographer receives rights to photographs if he makes decisions with regards to creative elements. In this way, the camera acts as a tool, the photographer as the owner of IP and photographs as the creation to be afforded IP protection. If this argument were to be applied to content generated by AI, the AI application can be treated as the tool, the output generated can be considered as the creation and the user as the owner of IP. The application of this argument depends on the usage of creativity by the user of the generative AI tool.
Since AI is trained on data, concerns of third-party copyright getting violated by the content generated is beginning to crop up before the courts. Mona Awad and Paul Tremblay are suing OpenAI (Parent company and owner of ChatGPT) claiming their books were used to train ChatGPT without their consent. It also brings to light that if AI is trained in such a manner, they may be able to replicate a particular author's style. It highlights that copyright laws universally are under-developed to tackle the challenge of AI violating copyright laws.
Andersen v. Stability AI et al is another example of a class action lawsuit filed against Stability AI (developer of Stable Diffusion) & other companies stating that they have acquired over five billion copies of copyrighted works of artists and used the same to train their AI which in turn produces high quality fakes. Among other allegations, such as storing and using copyrighted material without consent, the artists have alleged vicarious copyright infringement as Stable Diffusion enables third parties to generate works that can pass as originals of the copyright holder. The success of the suit depends on various factors, however, it would be hard to prove a copyright claim as a particular case of infringement between two products has not been made in the claim.
The European Union has recently proposed the AI Act as well as amendments to laws related to AI to ensure that providers of AI models publish data regarding usage of copyrighted material for training purposes. It is a comprehensive AI law which intends to ensure transparency on behalf of the companies that provide generative AI tools like DALLE.E and ChatGPT. It includes an inclusion of disclosure that the content is AI generated. The question on who owns the copyright is still left unanswered and depends on terms and conditions of the AI tools. It is this myriad of complexities which make a strong case for regulation of AI, especially generative AI.
The story of Hollie Mengert further highlights the dark side of generative AI. Hollie Mengert was a Disney illustrator who found that her works were used to train AI tool Stable Diffusion and produce content similar to her style. The copyright for some of the works belonged to clients like Disney and thus, it was a clear case of copyright violation. While Hollie decided against suing and expressed her displeasure, the incident acts as evidence of the possible copyright implications of generative AI.
The US copyright office denied copyright to an artwork titled ‘Suryast’ created by Ankit Sahni through use of RAGHAV AI. The attempt by Mr Sahni was to gain recognition for the inputs by him which contributed to the creation of the artwork. He argued that it was a creation through combined efforts by RAGHAV and his own inputs. The decision was largely based on arguments in the Thaler judgment which necessitated human origin for recognition through grant of IP rights. The Indian Copyright office and Canadian Intellectual property office aimed to recognise both Sahni and RAGHAV as co-creators and granted copyright protection. These decisions highlight differential positions in various jurisdictions and their interpretation of copyright itself.
Various tests have been developed in the traditional sense for distinguishing between
copyrightable and non-copyrightable content. Sweat of Brow doctrine which isn’t applicable in India for setting low standards aims to grant ownership of copyright to a creator who has put in ‘efforts’ into the creation. ‘Modicum of creativity’ is the American standard which requires a certain degree of creativity to make content copyrightable. The McLachlin test as used in Canada is followed in India which necessitates the work to be a product of authors skill and judgment.
The different tests have been drafted taking into account traditional creations and do not account for novel concepts like Artificial Intelligence. Generative AI, while being a great tool to help humans, can affect the copyright of artists who have worked for years to create their work and form their own techniques. Years of work can be replicated similarly within minutes using AI and therefore, it is the need of the hour to regulate AI generated content. Laws should be put in place requiring permission from artists and owners of copyright before using their work to train AI tools. Further, clear guidelines of ownership would give an understanding as to whether copyright is owned by the AI tool, the user of the AI tool who generates output or third parties whose content has been used to generate the output.
Differences in the policies of different Generative AI can lead to confusion with regards to
ownership. Recognizing the need for laws in India, the minister of state of information Technology, Rajeev Chandrashekar announced in June that the center intends to regulate AI to protect digital citizens. It is yet to be seen, what the new laws will entail for AI regulation in India. The Union Minister of State for Commerce and Industry, Shri. Som Parkash has recently submitted a written reply in Rajya Sabha stating that the current IPR regime is capable of dealing with AI generated works. However, seeing the spate of cases, it is evident that Artificial intelligence generated works can have an impact on upcoming copyright claims.
Recommendations include understanding that proliferation for AI generated content requires protection under IPR ambit. A standardized test requiring recognition of generative AI and contribution by the author through inputs could provide a solution to the decision making process of whether a creation is copyrightable .
Comments