Legal issues pertaining to the use of AI are growing exponentially, perhaps even faster than AI itself. Think about it, how does AI even work? How does it churn out answers faster than you blink? Let me give it to you straight from the horse’s mouth. As stated on OpenAI’s page- What is AI, AI tools such as “ChatGPT” are based on large language models (LLMs). It says that LLMs learn “patterns” from large amounts of text, from many sources, which they then use to generate and transform text. The models keep training and updating, for a better user experience. The ISO is of the view that AI is a very powerful tool. It also says that there is a need to ensure responsible use of the vast data that it collects. Another key concern that arises with AI tools is the possibility of copyright infringement. Copyright is the automatic legal right to exclusive ownership of any “original work”, that vests in a creator.
Derivative Works and Storage of Data for Training Purposes
The law on copyright, as it currently stands, in India, does not specifically talk about AI. Therefore, we can only interpret it on the basis of court judgments and government circulars. To understand whether any copyright vests in AI generated work, one should have a look at the Supreme Court’s landmark judgment, in the case of Eastern Book Company & Ors. Vs. D.B. Modak & Anr. (2008)1SCC1. It talks about “derivative works”, which are essentially new works created on the basis of, or derived from prior existing works owned by other creators or authors. The Supreme Court ruled that copyright vests in derivative work, if there’s some degree of creativity and application of mind. Essentially, the derivative work must carry a flavour different from the original. In the context of AI, this could mean that an AI generated work “derived”, from publicly available content would be infringing if it is replicated as is without the author’s consent. More recently, the Delhi High Court has rendered a judgment very relevant to the developing jurisprudence around AI. In the case of ANI Vs. Open AI, the court has held refused the grant of an interim injunction to the Plaintiff, holding that the storage of data for training purposes was not itself infringing copyright.
Authorship of AI Generated Content
Another interesting question that props up when we talk about the law around AI, is the “authorship” of AI generated content. Different countries view this issue with their own unique lens. For computer generated work, the Copyright Act in India defines the author as the person who causes the work to be created. Although it doesn’t specifically refer to AI generated work, a literal interpretation of the provision indicates that the human author initiating the prompts in AI would be the author. The US Supreme Court, in the case of Thaler Vs. Perlmutter (Case No. 25-449), held that “human authorship” is necessary, and there is no copyright in purely AI generated works. In the UK, the legal position is more clear. Section 9(3) of the Copyright, Designs and Patents Act, 1988 stipulates that the author of any computer generated work is the person who makes the “arrangements” necessary for the creation of the work. This provision existed long before AI ever came into the picture. A plain reading of this section indicates that the person initiating the prompts in AI would be the author. Chinese courts have in more than one ruling, held that content generated by AI is eligible for copyright. They’ve held that works created with the use of AI must contain elements of originality. However, they’ve granted copyright to derivative works created by minimal prompt engineering alone, even if they’re new works created by combining or modifying pre-existing works created by other authors. Chinese courts have also recognized the human initiating the prompts as the author.
Key Principles Emerging Across Jurisdictions
Across jurisdictions, some core principles emerge. The higher the level of human creativity and application of mind there is, in the creation of an AI generated work, the greater the probability of copyright vesting in it. Work that is generated through AI without much application of mind, or is merely a reproduction of prior existing works, is likely to be considered infringing. So far, only human beings have been recognized as authors, and prompt engineering is itself turning out to be an art. In the context of India, even thought the Hon’ble High Court of Delhi has held that storage of data by OpenAI, for training purposes, is not itself infringement of copyright, there are still many loose ends.
It is a known fact that AI uses data and prior existing works to create new works. What is the criteria for deciding whether a work created by AI shall be considered as a derivative work, and therefore an original, or a mere duplicate and a copyright infringement? What shall be the extent of liability of the AI platform being used, along with the human author initiating the prompt, for the content being generated? What happens when the human author trusts that a derivative work created by the use of a particular AI platform would not be infringing, yet the outcome turns out to the contrary? How can we have checks and balances on the unauthorized access and use of data by AI tools? The jurisprudence around these questions is still developing, in India and around the world.
A Peek at the Future
Since we’ve delved into some of the core legal concerns pertaining to the use of AI, at present, it’s only fair that we have a look at the future. AI is here to stay, and with its use, legal concerns are growing, exponentially. So what does the future look like?
Well, for starters, the EU AI Act isn’t going to be the only legislation for AI. Each country will have its own laws. There may also be international regulations framed by bodies such as the UN, or regional councils. These will serve as reference points, especially when there is a conflict of laws between countries. It’s also evident that authors will not be able to evade liability. The greater the usage of AI by an author, the higher shall be their threshold for liability, in the event of infringement, whether deliberate or inadvertent.
Entities offering AI services will also have to structure their contracts and operations very carefully. They will need to disclose the sources that they use to generate their output. This will keep a greater check on the unauthorized use of data and infringement of copyright. It will also become increasingly difficult for AI to claim “intermediary status”.
Conclusion
AI is one of the most powerful tools of the digital era. It’s transforming the digital economy in ways that one cannot fathom. It’s only natural that robust regulatory regimes need be instilled, to tame it. This will help ensure its constructive, rather than destructive use.