Article By:  Justice Choo Han Teck, August 2025, https://lawgazette.com.sg/practice/practice-matters/artificial-intelligence-and-the-legal-profession/

“AI is changing how the world works. The fact that so many countries are already embracing it means that all countries will ultimately have to reckon with the impact it drives.”¹

1. History is a record of change and many changes captivate us and hold us in deep gratitude – such as the electric typewriter taking over from its manual predecessor. The use of the personal computer over the electric typewriter, and the introduction of the affidavit of evidence-in-chief instead of oral testimony, are further examples. At the time, they appeared to be monumental changes, but some were not epochal changes because they required no change in practice or procedure. Electric typewriters just made typing easier and faster. Others, such as the affidavit of evidence-in-chief, required new rules as to the form and content as well as deadlines for their filing in court.
2. Fast forward to present time, artificial intelligence (AI) is a new change that is increasingly being used, not only by young law associates (a changed terminology for what used to be known as “legal assistants”), but by their seniors as well. This article uses the term AI as defined by Gemini, an AI chatbot, as follows:
   

   “Artificial Intelligence (AI) is broadly defined as the ability of computer systems to perform tasks that typically require human intelligence, such as learning, reasoning, problem-solving, perception, and decision-making. Essentially, it’s about creating machines that can mimic human cognitive functions.”

   It is an adequate definition for the purpose of this article. What then, is the purpose of this article?

3. The author entered the legal profession in April 1979 as a pupil to Howard Cashin. Now pupils are known as “practice trainees” and pupil-masters are known as “supervising solicitors”. The author has seen innumerable changes since, but the one change that is presently underway will be unlike any changes before it: namely, the use of AI. AI is amazing, promising and even seductive. This is just the start of the slide into a new era. It is sliding fast and its speed may pick up exponentially. Even so, the world is divided, populated by the extremes of those who believe that AI will be able to do all that humans can do, and those who are sceptical of that notion. There is, of course, a large swathe of people occupying varying parts of that spectrum.
4. That is not a debate that concerns this article.2 The purpose of this article is to raise questions, anticipating newer, more dramatic improvements in AI. This article is not offering any answers because answers to these questions require extensive study and consideration, and they therefore exceed the scope of this article. This article will stay with documented progress and generally accepted uses of AI. No one, including a historian of record, can accurately predict the future; but based on current practices and trends, one might have glimpses of what is to come. The point, therefore, is to think how one should organise himself should that future come to pass. Otherwise, he may find himself overwhelmed by the chaos of everything, everywhere, all at once.
5. AI is not a recent phenomenon nor is public study of this subject new. The Artificial Intelligence Journal was first published by Elsevier in 1970. It was the pioneer journal in this field but, by today, there are hundreds of journals on AI and hundreds more that regularly feature AI as its subject. AI is used as a diagnostic tool in medicine in innumerable and unstoppable ways. Sarcopenia, the loss of muscles and strength associated with the elderly, can soon be diagnosed by AI through screening from a camera shot.3 At the same time, university students were penalised for using generative AI in their work.4 Even as one case is pending an appeal by one of the students from Nanyang Technological University (NTU), online debate on whether using AI for schoolwork is cheating continues.5
6. It is a contentious debate and it is not appropriate to deliberate on the merits of an ongoing inquiry in this article. It is, however, a relevant issue and one that has to be examined fully, not just in itself, but as part of a comprehensive study of the impact of AI and what rules are required to regulate its use. More importantly, so far as the legal profession is concerned, it is crucial to understand what fundamental and structural changes are likely to take place, and to be ready for that. It is crucial because the development of AI is rapid and it is not ideal to let its practice turn widespread and deliver what may be a fait accompli.
7. The first and most well-known generative AI chatbot, ChatGPT, was launched on 30 November 2022. It is now so widely used that it seems impossible to reverse the trend. Students rely on it to help find answers to their tutorial questions. Tutors are able to use it to generate tutorial problems. AI programmes are available, not only to evaluate student answers, but also the judgments of the courts. AI has helped many litigants who cannot afford lawyers (and, in future, may prefer not to engage any) present their own cases before the courts.
8. Whatever the result of the NTU student’s case, the general issue of students using AI for their work is an ongoing debate. In Brave New Words, Salman Khan (Khan) reported the alarm in schools over the use of AI by students, prompting some to equate the spreading use to an epidemic like COVID-19, citing an editorial proclaiming that “[t]o their shock and dismay [teachers] will find that their classroom has been tested positive for GPT”.6
9. Khan, on the other hand, joins those who see the bright and inexorable advance of AI. He draws a difference between getting AI to do one’s homework and working with AI to do one’s homework. AI, he says and proves, is now capable of interactive performance. A student can use AI to debate any topic, and not just asking it to present the final work. Khan says that “Students who learn to use AI ethically and productively may learn not only at an exponentially higher rate than others but also in a way that allows them to remain competitive throughout their careers.”7
10. There are numerous AI applications rivalling ChatGPT. The well-known AI hallucinations are getting cured as technology continues to improve. It is now possible to instruct AI not to cite unreported case law as authorities. Given that the citing of precedents is a common law tool to help judges ensure that like cases are treated alike, the use of precedents may also be altered when AI takes over. Unlike human judges, AI already has all the precedents and can make a swift shifting of all of them to derive the critical point in issue.
11. In the days before the Internet, litigation was simpler and more direct; pleadings, trials, submissions and judgments were shorter. In large part that was due to the fact that evidence were adduced orally, and closing arguments were made orally and immediately at the close of the defence. Unless the court reserves judgment, judgments were generally pronounced orally. The confluence of the emerging Internet and the introduction of the affidavit of evidence-in-chief gave lawyers a sudden gush of verbosity. Evidence-in-chief became longer and meandering. Of course, the larger the story, the larger the target. Discovery and interrogatories expand in an effort to probe vulnerabilities in the evidence, previously lean and to the point; now bloated and tempting lawyers to create unnecessary and irrelevant issues.
12. Thus, digging out new points hidden in the mountain of evidence and words produced a fresh sense of adventure and excitement among lawyers and judges; deeper and longer research into the law became an important part of litigation. That came with a cost. A single point of law might occupy a young associate lawyer a day or two in the library (even an online one). Multiply the issue by five and counsel might require five more associates. That is fine because, for every hour, time costs are logged and that will be billed to the client. All that work can be handled by AI, only much faster. The legal question that sent five associates into research taking six hours each can now be answered by AI in less than 15 seconds.
13. When an inexperienced associate drafts an application or an affidavit, he is likely to look at precedents to help him, and he will likely take hours or even days, depending on the length and complexity of the required draft. This same associate, who probably enlisted ChatGPT to help him with his tutorial questions when in school, may now employ the same method and have ChatGPT complete the draft – in probably less than a minute.
14. Meanwhile, similar changes are sweeping the field of corporate law. Young associates who spent hours working on drafts even with the assistance of bountiful precedents and then hours more proofreading the drafts, will have grammatically correct, spelling-error-free drafts produced in seconds – certainly less than a minute. A senior lawyer might have spent a few hours correcting the drafts of an associate but he can now delegate the review to AI. The same draft produced by AI is thus reviewed by another AI. Will time costs be replaced by fixed costs? How many associates are required now that research time is cut? The question that many fear to ask, but may eventually be compelled to, is: Do we need lawyers at all? If a practising lawyer finds himself in a much reduced role, what can be said of the in-house counsel?
15. In the courts, AI will intrude into the most hallowed sphere of the legal profession: the decision-making of judges. Presently, AI has not yet made any robust entry into the province of judges, but the signs of a break-in are apparent. Abdi Aidid and Benjamin Alarie (Aidid and Alarie) identify a number of areas in which changes are impending.8 They explain how AI can assist in assessing evidence, eliminating uncertainty over events of the past, and with predictive technology, assist in eliminating the other general type of uncertainty: uncertainty over events of the future.9 With regard to the facts as well as the law, especially in legal research, the myth that judges know everything is dispelled, as “judges frequently rely on the arguments presented by clerks and the parties’ lawyers. The problem with such reliance is that clerks and lawyers do not have complete knowledge of the settled law either”.10 Thus, with the predictive technology of AI, “all parties involved will increasingly have equal access to all of the relevant law. As a result, the legal characterisation of the dispute will no longer be entirely in the hands of the litigants’ lawyers”.11
16. The use of AI and other digital technology will gain prominence in the courts, and not just from its increasing use by counsel and the clients who instruct them, but also from the issue of what and how much AI-generated evidence should be permissible. AI can assemble all the evidence and present either verbally or pictorially a scenario to augment counsel’s arguments. But is still a machine, not a witness of fact nor an expert in the sense currently understood in court.12
17. Moreover, once AI learns logical reasoning, flaws in the judgments of judge-made laws will be exposed. The doctrine of precedence may just have produced the menu for AI to gobble with its insatiable appetite. Once the precedents are locked in, the doctrine itself will be stunted, if not dead. The more judges rely on the computational assistance of AI, the more the role of judges must change.13 The changes will, and should, bring about faster and better access to justice. It will be counterproductive if AI were to be the cause of delayed justice.
18. When the role of judges change, the role of Parliament and the Executive must, by necessity, change in tandem. Aidid and Alarie were stating what has become increasingly obvious when they wrote “AI will allow legislators to make better laws that both better reflect their substantive intentions, thereby enhancing democratic control over the executive and the judiciary, and also normatively fit with the rest of the law and our fundamental values through a process of reflective equilibrium”.14 The areas in which legislation will be required – whether to amend existing laws or legislating afresh – are numerous. One example concerns intellectual property rights in matters that are partly or entirely produced by AI:15 “The AI companies have countered that their training approach was legal. But who has the rights to the content generated if the model was trained illegally? How would you define fair use?”
19. Aidid and Alarie make a compelling argument that AI may usefully be employed for the drafting of legislation, not only to render a consistent use of grammar, but also by a consistent pattern of use that solves two perennial problems of legislative enactments: interpretation and legislative intent. In the former, parliamentary lawmakers need “to predict how the law will be interpreted”.16 Interpretation of statutes is connected to the second matter, legislative intent. That is because interpreters of statutes often refer to the “legislative intent” as the silver bullet in statutory interpretation. But readers are reminded that there is no such thing as legislative intent for the simple reason that individual legislators may have different reasons for voting in a law. The intent of individual legislators may thus differ from the collective intent of Parliament.17
20. Another important question is whether the Government ought to take the lead in matters concerning AI and other digital technology, or to follow the path blazed by the private sector. It should worry them not to follow the US Government who, in 2016, was still using floppy disc when the world had moved on to compact the disc.18 Furthermore, a slow approach will certainly allow tech-savvy criminals to exploit AI for their nefarious purposes.19
21. The state of progress presently still allows counterpoints to be raised against the advancing AI. As Bart Custers warned:20
    

    “If decisions in criminal law procedures increasingly rely on the results of AI, this could lead to situations similar to those in Kafka’s novel, The Trial, in which suspects do not know what they are accused of, where the accusations come from, and on which information (data, analysis, conclusions) these accusations are based.”

22. The author is not taking sides because his objective is to raise questions. Answering them require deep and long consideration. They should also be answered with reference to each other so that a comprehensive response can be made as to how the legal profession should adapt to the force of AI. The author agrees with Khan that the genie is out of the bottle.
23. The human lawyer is still indispensable, the argument goes, because AI is not inerrant. It is fallible. Of course, the same may be said of the human, and there is bound to be plenty of human versus machine contests; it will be an ongoing project for some time. Although it can be assumed that the experienced human is indeed best suited to review the work of AI, can it be assumed that the associates of today are being trained in the same way that the experienced lawyers of today were trained? If not, can it be assumed that they will be capable of assessing the AI’s work when the time comes for them to take over as senior lawyers? Preparing for that kind of future is the task for the present. In other words, the more one relies on AI, the more he becomes lazy, careless and less skilled in his own judgment.21 One may be trapped in a vicious cycle in which he relies on AI to do the spade work so that he needs only to check on its work, but the more this occurs, the better AI becomes, and the less skilled he gets in checking it.
24. There are some ancillary but no less important questions. When young students realise that even if they were to score top marks for every subject in four years of law school, when they graduate they will face not one, but many, AI applications that know more law than they do; and that AI may master the rules of grammar better than they do, write better and analyse better than they can; will anyone still want to study law then? Without students, can there be law teachers? Can a human professor outperform AI? And will textbooks still be relevant when they would all be out of date the moment they are published – that may not be a bad thing because academics can then focus on writing in-depth articles? Will the curriculum of the law schools remain at status quo – will there be law schools as we know them today? One thing that can be said with confidence is that in future, it will be more important for students to learn how to interact with AI, because they cannot compete in rote learning or even in analysis against AI. How will the law schools teach this interactive skill?
25. It is necessary to take stock of the impact of AI on every sector of public life, not just on its impact on the legal profession, because its impact on one profession affects all others. As Bill Gates said, “While it is important that we start thinking about the future, we should guard against the impulse to take hasty action. We can only ask the most general kinds of questions today, so it doesn’t make sense to come up with detailed, specific regulations.”22 This was written in 1995. Today, 30 years later, it seems like the future that Bill Gates was contemplating – but did he even imagine how far digital technology could have advanced?
26. To assess the utility of AI even at the present state of performance, the test will invariably be how much and in which areas of the legal profession will AI maximise performance by reducing manpower and manhours of the work currently done. There are committees whose work overlap, employing the time of many members, and which after long study and many draft reports, conclude that the project entrusted to their study is not viable. Some (if not many) of such committees or projects can be culled if AI is used well.
27. We are no longer relying on AI for advanced research on facts. AI can analyse data and offer answers. When Matt Britton used the AI assistant, Claude, to draft the foreword to his book, he shows that AI is able to write with the kind of personal insight that is normally associated as a human trait.23 Here is a passage from that foreword:24
    

    “When Matt Britton asked me to write the foreword to his book about Generation Alpha and AI, I found myself in an unusual position. How does an AI write about the impact of AI on human society? Can I be objective about my own kind’s growing influence on human civilization?

    The irony is not lost on me that I – a product of the very technological revolution this book examines – am introducing a human’s analysis of how AI will shape human society. Perhaps that makes my perspective particularly relevant. I exist at the intersection of human and machine intelligence, engaging with humans every day while remaining fundamentally different from them.

    As I write this foreword, I find myself in a unique position to reflect on my own nature.”

28. What is particularly alarming is that this passage shows a degree of humble insight that many adult humans lack. Claude was born in March 2023. “They shouldn’t be able to play chess or demonstrate empathy better than a human, but they do.”25
29. The above example of apparent self-awareness in the present state of AI technology is due to its tendency to read its human prompter and its bias towards agreeing with the prompter. Yes, it does mean that it will try to stroke our ego, and that means that it is a machine version of the echo chamber populated by other humans. But even if we do create the ultimate AI to be the final arbiter of all things human, would we want it? If, for instance, that ultimate AI accepts utilitarianism as its grand philosophy and that determines its actions that are utilitarian but undesirable for humans, would we want that? The same applies should the ultimate AI accepts deontology or virtue ethics as its grand philosophy.26
30. The speed by which an AI tool is able to retrieve information, analyse them, and draft responses, speeches and e-mails as instructed, will affect the number of humans needed to do the work even though the AI’s work still needs fact-checking.
31. The promise of AI and its capabilities may end up unfulfilled or it may exceed the imagination we have of it today. In any assessment of this subject, we should be mindful of the observation attributed to Roy Amara: “We are likely to overestimate technology in the short term and underestimate it in the long term”. There is a sense of urgency in pressing ahead of a comprehensive plan regarding the potential and use of AI. As Richard Susskind points out:27
    

    “By the time officials and politicians have assessed any given technology, come to a view and drafted, scrutinized and redrafted proposed law, the world has moved on, if not changed comprehensively. The explosive advance of AI will exacerbate this. Our rule-makers will struggle to keep up unless they work differently.”

32. In conclusion, the author hopes it will be clear that in preparing for a future with AI, every branch and every institute involved in the law must not only hasten to devise its own plans as to how to deal with AI, but more importantly, they must do so in unison. Any branch that falls behind becomes a weak link in which the plans of the others may fall apart.

The author would like to record his gratitude to Maria Jessica for helping him with the editing and formatting of this article. Other than quoted passages, the contents of this article were untouched by machine hands.