A Need to Rethink the TDM exception in the wake of AI?

Written by Tanya Sara George, The author is a law student currently pursuing BA.LLB (Hons) from MNLU, Mumbai

Introduction 

Recently, ANI, a news organisation, filed a suit against Open AI in the Delhi HC, citing copyright infringement. While similar lawsuits have been filed across the world, this marks the first case of copyright infringement against OpenAI in India. The petition based its claim on concerns that OpenAI often uses its news reports when feeding information to users and inter alia also feeds wrong information, linking it back to the petitioner’s news channel. 

This case, being a first of its kind, has brought to light certain lacuna in the legislation governing the growth of AI and intellectual property in India. The court's refusal to grant an injunction against the AI shows its inclination to treat this use as an exception, however, this stride is without legislative support. Notably, albeit the exceptions to copyright are enumerated under Section 52, the section may not encompass the technological models used by AI. One such lacuna brought forth by the petition is rethinking the need for Text Data Mining, a process of computing vast amounts of texts to create structured information, as an exception to copyright in India. Thus, while this case is a milestone, it raises significant questions about the efficacy of the copyright regime in India, when confronted with AI. 

This article aims to address the ambiguities brought on by this recent petition. The author first analyses the present copyright regime, highlighting its drawbacks when referring to cases concerning AI. Secondly, the author analyses a crucial need of the hour - the necessity for a TDM exception in India to harmonize AI into the copyright framework. Lastly, the author compares international frameworks to suggest better domestic pathways. 

The Current Copyright Regime and AI 

Section 13 allows any person to gain a copyright for their original work. As per a later amendment in section 2(d)(v), this also includes computer-generated work. However, domestic courts have narrowly construed the definition of a ‘person,’ limiting the scope to only natural persons, indicating that an AI would not have the locus standi to gain copyright in India. This standing was also reaffirmed in the case of Raghav, an AI model who was refused from gaining copyright despite the co-author being a natural person. 

Now, the case for Open AI and ChatGPT is different, as these interfaces use copyrighted works and simply place them in a structured and compiled format. They do not necessarily result as the ‘author’ of the subsequent work. A fundamental question then emerges in the domestic landscape as to whether this would constitute infringement under Section 51. As per section 52(1)(a), the doctrine of fair use, an exception to copyright infringement, would, per se, not be applicable to computer programs. This leaves a major lacuna for users to always claim infringement under AI, as the doctrine of fair use is not squarely applicable to this scenario, tying up these programs in years of court fees. 

Moreover, courts have also held that the fair use doctrine would not be applicable to commercial cases and independent research cases, which creates further confusion as certain copyrighted works need to be used in the creation of an AI, which may fulfil the characteristics of being “commercial in nature.”  This becomes increasingly difficult in today’s world, where academicians are increasingly co-creating with businesses, blurring the lines of what constitutes a commercial activity. Thus, the lack of established guidelines on the scope of fair dealing with respect to AI allows for a can of worms to be opened by anyone with a copyrighted work. 

Additionally, this framework does not account for the training models of AI, which often uses several copyrighted works for effective optimization. Although these would be covered under the transformative use doctrine internationally, in the domestic sphere, it remains a grey area. Further, beyond the fair use doctrine, the other present exceptions to infringement enumerated under §52 do not account for technological models such as AI, rendering them open to a plethora of lawsuits. 

The TDM Exception 

‘TDM’ or Text data mining is essential to the functioning of AI. It refers to a process wherein extensive amounts of digital text are fed to an AI, which is then enabled to find patterns and valuable information from this corpus. India has yet to create a TDM exception. However, we may have to rethink this very soon. 

It has been internationally recognised, both in legislation and in cases such as Author’s Guild vs Google Inc., that using TDM, including copyrighted works for research purposes, would fall under the ambit of fair use and function as a legally valid exception to copyright infringement. The domestic recognition of the TDM exception even dates to the Puttaswamy case, wherein the court saw the legitimacy of data mining for valid purposes and upheld the same. Moreover, it has already been implicitly domestically observed that text data mining that does not unreasonably prejudice author rights would not be considered an infringement. Although, the court held that TDM would only be valid when within the boundaries of §52. This simply cannot be accomplished presently, as the section did not account for AI models. 

The substantial gap present in the copyright regime can be rectified to account for all known AI models through the introduction of a TDM exception or an adaptation of the fair use doctrine to include TDM in §52. This creates an exception for the information given by AI models, as well as the information fed to create the AI models. In the event of this exception, domestic safeguards such as the total concept and feel test can be applied to ensure that the data produced is not an exact representation of the copyrighted work and that the rights of the copyright holder are not violated.

Analysing International Frameworks 

The UK, under §29a of the Copyright, Designs and Patent Act, allows for the TDM exception for non-commercial uses. Although a stride forward, the inability of the exception concomitant to commercial purposes would cause major hindrances, when confronted with newer versions of AI down the line. 

Looking at the EU Directive, Article 3 allows for TDM in non-commercial research uses, and Article 4 of the directive allows for TDM in commercial use, as well as independent research. The article attempts to create a balance between the two rights as copyright holders are given an ‘opt-out right’, which would then make the exception void. This right, however, renders the risk of the exceptions once again being narrowly construed and essentially unusable. 

Section 243 of the Singapore Copyright Act provides for the TDM exception and extends it to the recording of protected works. Section 244(2) of the act allows for certain conditions for the exception to be applicable to ensure that the rights of the copyright holders will be protected. Thus, while restricting the scope of TDM to indiscriminately violate existing copyrights, the act allows for their judicious use without hampering the existing rights of holders. 

As per §244, inter alia, the AI creators or programmers would first require lawful access to the material only subsequent to which the copy can be created for computational data analysis, i.e., to feed it to the AI. This ensures that the copyright holders would still gain their rightful revenue for their innovation. It seems that following the Singaporean model would pose the least hurdles and ensure the best balance as the rights of both the copyright holders and the AI creators are considered. 

Alternative Path 

Alternatively, in the lack of the TDM exception, another way forward would be to follow the path of the USA. The doctrine of fair use in their jurisdiction is so broad as to cover TDM within its ambit, as there are no uses specified therein. The judge decides on each case of fair use using a set of factors such as the purpose, the impact on the market, and so on. This stance has been shown in cases such as Author’s Guild Inc. v. Hathitrust., where the court ruled that the mining of copyrighted works for the purpose of data analysis and database creation would be protected under the doctrine of fair use. 

Similarly, in the Google Books case, the court held that the entire corpus of human knowledge used to create a relational database would make it a transformative use, once again falling under the ambit of the fair use doctrine. Thus, unlike the narrowly construed nature of the domestic fair dealing premise, this thereby brings the fair use doctrine squarely applicable to all types of cases, including those concerning AI models. Additionally, this model allows the judiciary to adapt the doctrine in line with developmental needs without needing to adhere to legislative hurdles. 

Conclusion 

The advent of AI has revolutionized industries, but it has simultaneously exposed glaring inadequacies in India’s copyright framework. The ANI case underscores the need for a balanced legal approach to TDM and the broader interplay between AI and intellectual property. While Section 52 provides certain exceptions, it fails to account for the complexities introduced by AI models, leaving creators, users, and innovators vulnerable to prolonged litigation.

Drawing lessons from international frameworks, such as Singapore and the EU, India must enact a TDM exception that ensures lawful access to copyrighted works while safeguarding the rights of creators. Alternatively, expanding the scope of the fair use doctrine to encompass TDM could provide much-needed flexibility, as demonstrated by the U.S. model. These measures would strike a balance between fostering innovation in AI and upholding intellectual property rights, enabling India to remain competitive in the global AI landscape. Without these reforms, India risks stifling its AI ecosystem and missing out on a transformative era of innovation.