Unleashing the Power of Large Language Models with Responsible Regulation

[By Yuvraj Mathur and Ayush Singh]

The authors are students of Rajiv Gandhi National University of Law, Punjab.



Within the first two months of going live, AI-powered chatbots like ChatGPT and Google Bard became worldwide sensations with over 100 million users. While offering great opportunities, there is already a great deal of conjecture on how it might disrupt several industries, democracy, and our everyday lives. With AI gaining consciousness and taking decisions, users have reported that the chatbot is claiming to have feelings, gaslighting them, refusing to accept its mistakes, threatening them, and so on. As per a Fox News report, Microsoft Bing not only indulged in hostile exchanges but also wanted to steal nuclear access codes and engineer a deadly virus.

These Large Language Models learn natural language sequences and patterns from vast amounts of text data culled from existing sources like websites, articles, and journals to generate intricate results from simple input. In order to achieve this, it uses a modified version of the “Generative Pre-Trained Transformer” (GPT) neural network Machine Learning (ML) model. As AI systems like ChatGPT trained by OpenAI become more advanced and sophisticated, their potential applications in the legal domain continue to expand. In order to assess the necessity of implementing an AI-centric policy in India, this article critically evaluates the potential uses of the AI system in the field of law as well as its legal ramifications.

Revolutionising the Commercial Landscape

In light of the buzz Generative AI creates on media platforms, it can have a wide range of use cases in the commercial sector.

1. Customer service: Conversational AI can be used to provide automated customer support via chatbots, helping customers with frequently asked questions, order tracking, and other inquiries. It can also assist in lead generation and conversion by providing personalized recommendations and engaging in conversational marketing with potential customers.

2. Legal Research: AI Chatbot ChatGPT can provide general legal information on a wide range of topics and can also help with legal research by providing relevant cases, statutes, and regulations. After feeding it 50 prompts to test the reliability of its legal assistance, Linklaters, a magic circle law firm, concluded that legal advice is often context-specific and relies upon several extrinsic elements.

3. Legal Drafting: The software might theoretically be used to produce early drafts of documents that do not entail significant creativity. Nevertheless, since it does not grasp the law, correlate facts to the law, or employ human abilities like emotional intelligence and persuasion, it is likely to be deceptive in more intricate and nuanced legal documentation.

Allen & Overy (A&O), another magic circle law firm by incorporating Harvey, a cutting-edge AI platform based on a subset of Open AI’s most recent versions optimised for legal work, has made significant strides in the field of artificial intelligence. Harvey is a program that automates and optimizes several aspects of legal work, including regulatory compliance, litigation, due diligence, and contract analysis by employing data analytics, machine learning, and natural language processing.

Another machine learning software, Kira, assists in precisely and effectively identifying, extracting, and analysing contract and document information.

4. Data analysis and insights: Generative AI can analyse large volumes of customer data and provide insights on consumer behaviour, preferences, and trends, helping businesses make informed decisions. It can also be used to generate content such as product descriptions, marketing copy, and social media posts, saving time and effort for businesses.

5. Personal assistants: Virtual Assistants can act as digital subordinates, helping with tasks such as scheduling, reminders, and managing emails, despite being quite generic and superficial. ChatGPT can provide employees with personalised training and development content, helping them learn and upskill in their jobs.

The Liability Conundrum

ChatGPT’s position in the legal diaspora has been in question since its origination. The concept of liability of AI Chatbot’s has been in news recently after the revelation of certain racially discriminating content by ChatGPT in a prompt raised by a U.C. Berkeley professor, which revealed the inherent biases within the software. Any act of discrimination fundamentally goes against the tenets of Article 14 of the Indian Constitution. For the same, the question of the ownership of the content provided by ChatGPT and these Large Language Models becomes pivotal. In a reply created by ChatGPT in response to a prompt by the authors, the AI-based Chat Box’s reply reflected that it lacked indexing of data and was unable to provide the authors with the source of the information as presented by the Chatbot.

Furthermore, these Large Language Models can be observed as aggregators of the information provided by them, and the makers of such AI models can be held accountable for the same. A similar case of a platform being observed as an aggregator was visible in the case of Facebook, where the social media giant acted as an aggregator and undertook racial profiling as a method to identify the target audience for advertisements. As aggregators, these Large Language Models can be held on the same footing as Facebook, as both act as a medium of information between the content creator and content consumer

The Predicament of Ownership

An offshoot of the accountability dilemma is the issue of Ownership of the created content. The issue revolves around the fact that AI-generated content is created from pre-existing copyrighted data sets. Certain lawsuits, such as HiQ Labs v. LinkedIn and Warhol v. Goldsmith, and others, revolve around the issue of data harvesting from copyrighted content to train AI systems. For instance, Goldsmith established her contention in the Warhol ruling by demonstrating how Warhol’s prints violated the copyright of her images, notwithstanding Warhol’s argument that they were transformed in terms of size and colour. The U.S. authority should be used as a strong justification in this case even though Indian courts have not yet dealt with this question. Moreover, Section 43 of the IT Act, 2000 makes it illegal to retrieve data without permission.

The AI models create certain data by amalgamating various sources of information but fail to provide the owner of the content with any return. While there is no clarity regarding the copyright infringement liability of the content generated by these large language models, the issue is being prima facie dealt with in a class action lawsuit against Google and GitHub in the state of San Francisco, USA. While Google relies on a vast network of web crawlers that continuously index new pages on the internet to provide relevant results, ChatGPT, on the other hand, is pre-trained on a massive corpus of text data, which is continually updated and fine-tuned to improve its language processing capabilities. The major differentiation between the two rests in the fact that Google indexes all the data provided by the search engine.

Potential Legal Solutions

Nations all around the globe have been striving to modernise their current Information Technology regulatory frameworks in order to keep up with the advancement of AI systems. As a cornerstone of President Biden’s vision, the United States White House Office of Science and Technology Policy has proposed a Blueprint for an AI Bill of Rights to identify vulnerabilities, threats, and possible impacts of the AI system. UK Secretary of State for Digital, Culture, Media and Sport also presented a policy paper ‘Establishing a pro-innovation approach to regulating AI‘. Canada brought forth its C-27 Bill to provide guidelines for protecting personal information while taking into consideration the demands for enterprises to obtain, utilise, or disclose personal information for business purposes. Nonetheless, India still complies with the outdated Information Technology Act of 2000 (the “IT Act”). This raises concerns about how Sam Altman’s ChatGPT would be governed inside India’s current legal system.

The significance of intentionality is essential to addressing the issue of culpability. In accordance with this, which scholars Timnit Gebru and Emily Bender refer to as “stochastic parroting,” ChatGPT lacks an independent consciousness and thus imposes accountability on the creator. But now, anomalous entities like companies may be held liable for both civil and criminal offences, and they are given their own responsibilities.


An AI-specific regulatory mechanism is the need of the hour to overcome this liability conundrum. Liability imposed under this rule should consider the system controlled by humans. Liability should be solely incurred by the organisation the AI-system is connected to, and not by its programmers, in the case of a completely autonomous system. This step, however, has only been implemented in Saudi Arabia for an AI humanoid named Sophia. Furthermore, piercing of the “AI veil” needs to be permitted in the event it turns out that the programmers did not include adequate safety procedures to avoid all possible unlawful actions. As there is a large amount of human control in a semi-autonomous system, creators should be personally liable.

In addition, before releasing the AI system for commercial application, a data protection impact assessment must be carried out, as specified in the United States Wyden-Booker Bill. If it is established that the program violates already-protected data without properly crediting the source of the information, the application of the program must be denied. Nonetheless, this assessment must be in accordance with an appellate mechanism. Alas, this issue has been left out of India’s impending Digital Data Protection Bill. Lastly, considering that these large language models learn constantly from their surroundings, dire financial losses and the sole liability of the creator should also be shielded in case of unpredicted actions of the AI. With the advent of the Draft Digital Personal Data Protection Bill and Draft Digital India Bill, it will be engrossing to observe how these Large Language Models will be regulated.


Leave a Reply

Your email address will not be published. Required fields are marked *

Contact Us

Kerwa Dam Road., 
National Law Institute University, Bhopal
Madhya Pradesh, India. 462044​.

write to us at – cbcl@nliu.ac.in