A (Cautious) Case For AI In Legal Research

“I doubt if such an animal is worth preserving,” he says. “He must eventually go under in the cosmic struggle when pitted against well-armoured and warmly protected species, who have wings and trunks and spires and scales and horns and shaggy hair. If Man cannot live without these luxuries, you had better abolish Man.”

Curiously enough, the so-called modern intellectual overlooked the most distinctive feature attributable to a Man – his natural intelligence. And, as history bears (for good or for bad), the Man valiantly survived in the cosmic struggle without 'fur' on his body but surely with a 'feather' in his cap!

Introduction to AI

In The Coming Wave, Mustafa Suleyman, the co-founder of the London-based AI company DeepMind takes us through the transformational moments of human civilisation from discovery of fire to invention of wheel to harnessing of electricity. Homo sapiens, he believes, can more aptly be described as homo technologicus. The early nomadic way of life was replaced by a life of domestication of plants and animals in the Agricultural Revolution around 9000 BC. The next prominent wave came around 1700s with the use of steam and advent of the Industrial Revolution – building up of factory, railways, telegraphs, steamships, cars, etc.

However, the Industrial Revolution resulted in a crisis of excess information being generated. This crisis was harnessed by use of technology. The Information Revolution, thus, came to life. This change is captured by James R. Beniger in The Control Revolution: Technological and Economic Origins of the Information. Earlier stages revolved around proliferation of telegraphs and telephones for effective communication of information. Later, with the use of internet, online portals started collecting terabytes of data on a methodical basis. As more and more user-based information is collected by the online systems across the globe, faster and accurate processing and analysis of the data became the need of the day. A system which would absorb all the data and intelligibly respond to a query. The wave of Artificial Intelligence (AI) was born.

The use of computers in our daily lives is primarily employment of technology in processes to reduce human discretion and smoothen the processes. AI on the other hand tends to reduce human intervention all together by replacing the massive human-hours labour with split-second results. This technology is run by brute processing power, remarkable algorithm and large bodies of data. It excels in processing and understanding massive datasets and generating intelligible results based upon the query. AI has two facets - supervised learning (based on past conduct of humans) and reinforcement learning (self-learning based on self-created data and constant additions).

With supervised and reinforcement learning comes the capability of AI to do something which is likely to go beyond the human imagination and working. Tests in the gaming arena has pretty much proved this theory. Go is an ancient East Asian game played on a 19-by-19 board with black and white stones. One is required to surround the opponent's stones with your own, which would result in removal of the stone. AlphaGo was created by the AI enthusiasts, which was initially trained by watching 1,50,000 games. Over time, multiple AlphaGo systems played against each other and continued their self-learning. When AI-stimulated AlphaGo came to oppose a world champion, the result was astonishing. AlphaGo chose to skip all the earlier 'human' moves learnt by watching but rather 'created' its new moves. The era of artificial intelligence in the gaming zone had truly arrived! The game had gone beyond the human imagination of thousands of years and evolved the strategies to a whole new level. One is now witnessing similar strides by AI in almost all arenas known to a human mind.

Plethoric Availability of Judicial Decisions

Legal research through precedents is one of the most important skillsets a legal professional is expected to acquire. Official law reports in Indian courts found way through the Indian Law Reports Act, 1875 which authorised publication of reports of cases decided by the High Courts. Over time, the publication of Indian Law Reports was picked up by different High Courts. Post-independence, official reporting came under heavy criticism, particularly by the Law Commission in 1957, on account of them being expensive and very slow. This led to the growth of private reports, which continues till date. What has however now changed is the online publication of the judgments of all the Courts by various portals. No more is one dependent upon printing and bindings of law reports. Instant online reporting of judgments has become the norm. Supreme Court's official report – the Supreme Court Reports – have evolved through quality head notes and through introduction of Digi-SCR. Private portals not only make a judgment available online barely within a few hours but also provide a summary with appropriate links. Availability of immediate court updates through social media platforms has added to the pile of data available for access by a legal researcher. With multiple online reporting of precedents, even the line between 'reported' and 'unreported' judgments has evidently blurred.

In Legal Research and Methodology published by the Indian Law Institute (2001), M.P. Jain contributed a piece on Law Reporting in India. He criticised paper publication of multiple law reports regardless of their value as precedents. This issue seems to have raised its ugly head again in the present digital age with too many case laws being available for consumption, analysis and citing. The practical outcome of online law reporting is mind-boggling on account of its enormous quantity and uncertain quality. Overloading the system with countless precedents is likely to result in a conundrum for a researcher.

In a recent decision rendered in NBCC (India) Ltd. v State of West Bengal, 2025 LiveLaw (SC) 46, the Supreme Court considered the difference between decision-making and precedent-making and held that not every judgment is intended to be a binding precedent under Article 141. At the same time, the Court recognised the difficulties faced by the High Courts and the district judiciary in ascertaining whether a judgment of the Supreme Court is in the process of decision-making or precedent-making. With heavy volume of case laws being made available daily, the need for Control Revolution is being felt once again. We are today faced with a situation of excessive information (case laws) rather than lack thereof.

The author of the book 'Using a Law Library – A Student's Guide to Legal Research Skills' (first published in 1992), Peter Clinch (himself a law librarian) quotes Samual Johnson in the Introduction to the book who said, 'Knowledge is of two kinds, we know a subject ourselves, or we know where we can find information upon it.' The book offers a legal researcher ways to find the information in a physical library with a basic knowledge of the research resources and techniques relevant to the literature of law. It is fascinating to now go through Appendix 6 of the book, which dealt with LEXIS, an on-line full-text retrieval service, which was in 1992 freshly introduced in the legal research market. It offered a tool, which helped the researcher 'search the actual language used in the original legislation and judgments which form the database'. Even then, the offer to use this online research tool came with a caution: 'As with any computer system the information can be retrieved only in response to requests structured in a way the computer can understand.'

In the legal field, presently, technology can intelligibly appreciate the keywords provided for research and based upon the vastness of its own database, provide answers with corresponding legal literature. The use of legal research tools offer a support system to a legal researcher by speeding up the process of research. They provide the researcher with answers after scanning through volumes of judicial precedents and law literature based upon the keywords provided by the researcher. Hours of pouring over digests like Supreme Court Yearly Digest, AIR Manuals, Quinquennial Digest, pre-independence 50-years digests, etc. to locate a relevant case law hardly a couple of decades back stands firmly replaced today by instantaneous results through the legal research software. But one cannot underestimate the fact that the system can respond intelligibly only if asked intelligibly. The quality (and surely not the quantity) of output is directly proportional to the quality of input. The better the clarity in input, the more precise is the response.

Legal Research and AI

It cannot be denied that AI is rapidly changing the way legal research is done. It is already quite common to see legal researchers relying upon AI tools for legal research. Lawyers are often found raising queries and seeking precedents by using AI. The issue however remains of an intelligible question would expect a useful answer. To ensure insightful results through AI-based legal research, old-school reading and appreciation of legal principles is a must. After all, AI is primarily based on human knowledge created and gathered so far. It has the tendency of absorbing knowledge on its way and then, responding to queries in an intelligent manner. It also tends to artificially evolve its own intelligence by interplay of algorithms and its knowledge base (and, of course, its endless memory and capacity of instant extraction of relevant data).

With the ever-exploding data in the legal field (through legislations, judgments, writing, etc), a legal researcher has no option but to rely on all the tools available to him, be it printed material, web-based legal research tools or AI. But, depending on any one of them alone, and particularly AI alone, would be detrimental and self-destructive. A lawyer's legal mind and imagination is extremely crucial for his own professional growth. His professional art – for instance, drafting of pleadings, his articulations, his court craft, on-the-legs thinking, and his flow of thoughts – can be cultivated by constantly reading books (offline or online) and using the knowledge while using online legal tools or AI search modules. One should, therefore, adopt a multi-dimensional approach in legal research. Well-researched legal commentaries are as relevant as AI research tools. One assimilates multiple dimensions of a legal issue while flipping through the pages and could beneficially correlate with the factual matrix. A bare reading of the commentaries often offers new ideas and legal points to articulate.

One must countenance the fact that online research is limited to the legal jargon and principles of law one has learnt through research and reading. The principle of GIGO (garbage-in-garbage-out) has found much application in computer systems through the decades. GIGO refers to an idea that in any system, the quality of output is determined by the quality of the input. The GIGO concept applies equally to any online research including legal research. The need for constant reading, therefore, is more relevant in today's era to fully exploit the AI's potential in legal research. Only by a higher quality of questioning can one expect a better quality of output. Or else, a legal researcher is likely to land in the GIGO syndrome.

There is another dimension which too requires deeper introspection. Unlike the field of gaming, graduation of AI from supervised learning to reinforcement learning in the legal research is something one is yet to assess. Can one expect AI to evolve new legal theories and jurisprudence in future? Can it evolve a doctrine like the “basic structure” or new rights under Article 21, as the Supreme Court has been evolving over the decades? In short, can it, in view of contemporary social needs, provide such solutions within the constitutional scheme, which otherwise is not found in the legal literature generated so far. Like AlphaGo, can AI create its own novel legal moves and articulations and offer better solutions to a dispute? Is it possible for the AI to supplant rather than merely supplement the dispute resolution mechanism? Can it be an alternative to courts, arbitration, mediation and other ADRs present today?

We cannot overlook that law tends to regulate human behaviour and constantly attempts to keep up with the present-day ever-changing societal needs. While regulating such human behaviour, frictions arise require a dispute resolution process. This process is not a board game. A 'human' is at the core of any dispute resolution process, and he cannot be turned into a computer algorithm. Aharon Barack, former President of the Supreme Court of Israel, in his celebrated book The Judge in a Democracy, has stated, 'The life of law is not just logic or experience. The life of law is renewal based on experience and logic, which adapt law to the new social reality.' These words of wisdom re-affirm that both human behaviour and societal expectations undergo a change over the course of years and decades and law must be abreast to such changes as well.

AI and the Cautious Approach

The promise of AI can never be overestimated. As Rebecca Brendel, Director of Harvard Medical School's Center for Bioethics, says, “I'd be hard-pressed to say that we'd ever want to give away our humanity asking decision of high consequence.” This, of course, was commented upon in the context of taking critical decisions in the field of medicine. But why is it not equally important in the field of law? One should remember the words of Justice Rohinton Nariman, 'One should have a heart to enforce Article 21'. Would AI have the contextual 'heart' while appreciating liberty applications which so often come before the Courts? Would it be able to assist the bar and the bench based on large volumes of data in matters like matrimonial disputes where personal experience, societal norms, contextual balance, and equitable considerations are likely to play a role as much as the applicable law. It may suggest answers to yesterday's troubles, but would it effectively offer solutions to today's problems.

Innovative concepts are invariably the result of diligent and meticulous effort. The author of a ground-breaking idea has as much of a gift of original thinking as a sound research technique. It is a combination of these qualities of a legal researcher, which would help enrich the Indian jurisprudence. Use of AI in legal research is a welcome step. However, its use as a 'tool' for research ought not to be confused with the 'end result' of such research. The degree and depth of AI-based research might enormously vary from the traditional book-based or software-based legal research. Yet, it must be remembered that it still remains a research tool. An AI tool needs to be used as a part of one's legal research weaponry but cannot be the sole arrow in one's quiver.

In this context, AI hallucination in legal research is something one needs to be wary about. The system, being heavily reliant on large data, is known to 'hallucinate' and throw research results, which might not even exist. Experience in AI-based legal research shows that ghost case-laws with fake party names and non-existent precedents are shown as a perfect match to a research question. Blind 'copy-pasting' of these outcomes in one's pleadings or submissions is bound to lead to disastrous results.

This issue came to be flagged recently by Justice B.R. Gavai, Judge, Supreme Court of India at a conference organised by the Supreme Court of Kenya when he advocated caution while using AI. While acknowledging that AI can be a beneficial tool, he warned about the risks inherent in over dependence on AI and said, “Relying on AI for legal research comes with significant risks, as there have been instances where platforms like ChatGPT have generated fake case citations and fabricated legal facts. While AI can process vast amounts of legal data and provide quick summaries, it lacks the ability to verify sources with human-level discernment. This has led to situations where lawyers and researchers, trusting AI-generated information, have unknowingly cited non-existent cases or misleading legal precedents, resulting in professional embarrassment and potential legal consequences.”

Instances of AI-generated fake cases and their use in courts leading to embarrassing moments for the lawyer and sanctions being imposed by the courts have been found in USA (see here, here and here), UK, Canada and Australia. Recently, the Karnataka High Court ordered probe and action against a Trial Judge for citing non-existing Supreme Court judgments in his judgment. The Court noted, 'What is more disturbing is the fact that the learned judge of City Civil Court has cited two decisions which were never decided by the Apex Court or any other Court.' A recent study – AI on Trial (2024) – by Stanford University revealed that legal models hallucinate in 1 out of 6 (or more) benchmarking queries. The report emphasises the need for lawyers to carefully double-check each citation provided by AI research. These instances have intensified the need to be particularly watchful while using AI as a research tool.

Another unwarranted outcome of AI-dependency is 'informational obesity' leaving the researcher confused and irritated. This refers to excessive and unmindful skimming through the AI-based research at the end of which the researcher is generally left clueless of what he has actually read. Dr. Swapnil Shenvi of Narsee Monjee Institute of Management Studies, Mumbai, advises mindful information-consumption habits, like reading books, research papers, etc. Foundational knowledge from the original source is, therefore, indispensable. To be able to harness the best of what AI has to offer, one should be able to know what to ask of it.

If a legal researcher does not equip himself with reading law literature and does not indulge in original thinking and rather, relies habitually on AI alone for legal research, then he is likely to permanently mortgage his legal reasoning and understanding to the vagaries of AI. One must remember that AI can be a delightful companion for legal research if accompanied by (natural) intelligible questioning but can easily turn into an errant and irritable assistant if blindly followed.

The author is a Senior Advocate, Arbitrator and author of the book, The Indian Makeover: From Being Ruled to Being Governed. Views are personal.