Justice Aedit Abdullah: Keynote address delivered at the CIArb – IPOS IP and Tech Dispute Resolution Conference

CIArb – IPOS IP and Tech Dispute Resolution Conference
29 August 2024

Keynote Address

1.          It is indeed a great honour to speak this morning. If I may say, CiArb and IPOS have had great foresight in dedicating time and space for discussion about developments in dispute resolution in intellectual property and technology. Change is ever constant; technology ever more so, and all of us, in all the professions, have to accept that constant change and adaptation is part of our working life, all our working lives.

2.          What I hope to do in this speech is to lay out what we, in the Singapore judiciary, see as the likely impact of Artificial Intelligence, or AI, on the courts; what we see as the main areas where AI could assist the work of the judges; and what may come to pass in perhaps the mid-term. The parallels between the courts and other tribunals are close, so I hope this sharing will help spur discussion as regards both arbitrations and IP tribunal hearings. There may be areas of interest to all of you primarily involved in arbitration or non-judicial IP work, and certainly I expect there would be areas which we in the courts may have overlooked, and have much to learn from everyone else.

3.          Firstly then, let me take the current state of play. Right now, there is a lot of bubbling of hype, with occasional bouts of scepticism and even cynicism, about AI. We, in the Singapore judiciary, aim to take a dispassionate and objective approach. We want to explore, to kick the tyres as it were, so that we can see for ourselves what works, what does not; what AI can do right now; what it may not be good at, for now; and what we need to keep an eye on for the future. Nothing is an unalloyed good; there will always be some risk, some trade-off that needs to be made. We need to be conscious of both the pros and cons. To do that, we need to experiment and be willing to explore.

4.          What I can say is that while it may be that AI will continue to surprise us in many ways, the advent of robot lawyers and judges is probably some way away. Nonetheless AI, especially generative AI, will have significant impact on our work. Parallels can be drawn with the early days of the internet. I remember in the mid-1990s, talk was of the potential of the internet, of the world-wide web, for uniting people, with new channels of knowledge, heralding the dawn of a new age of Aquarius. Who could have foreseen the dominance of social media, fast fashion and TikTok challenges? What this illustrates to me is that the change and the challenges associated with AI maybe both less eventful than the hype, but conversely more significant and pivotal than we can foresee or that we can anticipate. I am reluctant therefore to attempt to forecast or predict. What I hope to outline are present efforts and what may come to pass in the near term. I will however go out on a limb and hazard one guess about the future.

5.          Let me briefly lay out what AI, especially generative AI can do right now. What must be borne in mind, as you would be aware, is that Generative AI works on producing text or images responding to prompts, by identifying the most probably sequence of words or pixels or other data, based on what it has gleaned from trawling its data set. Generative AI does not use knowledge or information to deduce through reasoning what goes together; it uses what it has to create a product based on what the data predicts as a matter of probability should come next subject to the constraints of the prompt given or question asked. So that use of probabilities frees the AI to produce seemingly creative work, but often with a great risk of a departure from what is real or factual, hence the hallucinations that we encounter.

6.          With that limitation in mind, in the near term, the immediate impact of AI is, we believe, in being a force multiplier for the work we already do. That is, AI can function well as an assistant in certain functions, rather than displacing us. AI can realistically help, at either its current state or very soon, with the work of lawyers, arbitrators and judges.

7.          Many law firms and legal technology companies are already hard at work in developing, rolling out and using AI on daily basis, in various areas, including discovery, marshalling of evidence, legal research, and drafting. There is also considerable room for the use of AI to expand in IP work generally. As it is, AI has also proven itself useful in many areas of transactional work, helping out with drafting and precedent research. We of course, are monitoring these.

8.          For us in the judiciary, a good example of AI’s capabilities is in digesting information, making it more readily understood or remembered by us. Summarising, tabulating or making more readable, documentary evidence or submissions are prime activities where Generative AI’s facility with text comes to the fore. Similarly, asking AI specific questions of the document generally returns fairly decent answers; this allows the judge to essentially speak to the documents. In all of these activities, limiting the mining of text or data to a specific set of documents helps to reduce the potential for hallucination or falsity, while giving a real benefit. The Singapore judiciary has found from our experimentation that using Generative AI in this way holds great potential for helping judges understand better the legal arguments and evidence put to them, by allowing them to weigh, contrast, and test what is submitted against case law, affidavits and documents. It allows the judges to zoom in on what is essential in a case. As it is, the volume of material, in the form of submissions and evidence, is constantly increasing, running into hundreds of pages of submissions and thousands, if not tens of thousands, of pages of affidavits and documents. Better and faster grappling with the material is needed to ensure the judges master their briefs, and hopefully help with the timeliness and quality of decision making. We are exploring how best to institutionalise and deploy this.

9.          We are also exploring AI in helping us with transcription and translation. Transcription in particular is important, given that hearings are oral. The challenge is that the court setting is not ideal. Many things can happen in a courtroom. Not limited to communication between the lawyers and the court, we find that AI still has difficulties with the messiness, and the situation is perhaps testament to the old adage that garbage in, garbage out. From what we can see, some human element will still be required for the near future. Translation is another important area, where there is an immediate demand for assistance from AI. Human interpreters will be needed and will continue to be used in proceedings for some time; there are nuances that cannot yet be captured by the machine. However, greater availability of good machine interpretation will help in many situations, where human interpreters may be either unavailable or unaffordable. Documents, including contracts, notices or affidavits, sometimes need to be interpreted on the fly; machine translation provides a path for greater efficiency and accessibility.

10.          Court administration illustrates a potential use of AI analytics. Lawyers are already using analytics and AI for prediction of outcomes, as well as what appears to be some degree of forum shopping. That is of course less of an interest for us in the courts, though perhaps one day we may need to put in measures to defeat AI-aided forum shopping. What analytics would be useful for us, in the courts, and is suspect for arbitrators and tribunals is in case management, particularly of complex litigation. This could include forecasting of likely difficulties, estimation of hearing duration, fitting in hearing dates and working out possible common dates between often very busy counsel. Being able to trawl through historical data to identify probable exemplars is very useful in this area.

11.          We are also examining its use in assisting us in preparation of judgments, including red-teaming, i.e. identifying weaknesses in our drafts, and perhaps in helping to develop checklists. Similar trends are likely to be seen in arbitration and other dispute resolution work. The challenge for us is in ensuring that AI functions as an aid, and not a crutch. The judge must be the one making the decision and must exercise her or his mind about the various issues. Any use of AI should only go to the extent of assisting in drafting rather than substituting for the judge. We need to consider carefully how we move in this space.

12.          Our biggest effort is not however, in complex commercial or technical matters, but in trying our best to improve access to justice by assisting self-represented persons (“SRPs”). We have entered into a collaboration with Harvey AI to assist SRPs in Small Claims cases; these do not involve lawyers. We are trying to see what can be done to assist the SRPs in the preparation of their cases, and understanding the position of the other side, through summarisation and translation, as well as assisting the tribunal magistrates in examining and understanding the evidence presented by SRPs to them.

13.          We are also examining the direct use of AI in dispute resolution, such as through advanced online dispute resolution, by nudging or identifying possible settlement outcomes, giving indicative outcomes or even first level decisions (subject to human review) for certain categories of cases.

14.          We recognise that the use of AI poses risks and challenges. Hallucination and inaccuracy are obvious and well known. Output will also vary according to the prompts used. The threats of bias and information leakage are also well-known.

15.          Some of these risks are addressed by proper design, and scoping the use of AI. In addition, guidance is being worked on by the Singapore courts, taking in inputs from our stakeholders. We had considered a requirement for disclosure that AI has been used, but this may be overly onerous given that AI availability is burgeoning, with an increasing number of AI apps and AI assisted apps appearing. AI use is neutral, and par for the course, and the responsibility lies with the court user to ensure accuracy of what is submitted to us.

16.          Throughout all of our current efforts, international cooperation and sharing is important: we have a regional technology meeting drawing together various jurisdictions, and are aware of various efforts. This allows us to share experiences, keeping track of developments and identifying blind spots.

17.          Let me turn next to what I think the near future, the next 10 years or so, may hold. I believe by that time, AI assistance will ubiquitous. It would seem that in the next few years, there will still be a high possibility of further leaps in performance. Some of this may be through the deployment of multi-agent AI, with multiple separate AI systems working together on a single problem. Additionally, it would seem that there are efforts to combine generative and logic or knowledge based AI, allowing us to reap the benefits of both. Other yet unknown improvements may be in the offing as well.

18.          All this underlines that in the next decade or so, AI will probably be ubiquitous and commonplace in all our lives. It will be as much part of the every day as the internet and mobile phones are to us right now. Beyond what we are doing now, I think it likely that AI will be assisting us in assessment of cases, particularly settlement and negotiations. It will have a role in legal strategy such as forum selection, public relations, financing and funding planning. AI will probably be essential in helping lawyers prepare their advocacy, for instance in the preparation of questions, discovery, and methods of presentation of evidence, argument preparation in writing and collation of materials.

19.          All of this should lead to an increase in productivity. But no improvement in productivity ever goes unpunished. With greater productivity comes heavier work. Judges often refer to being on the judicial treadmill; AI will enable us to perhaps run faster on the treadmill, but  the treadmill speed and the incline will likely be increased. What I mean is that expectations from clients, stakeholders and the public will be that all of our work will have to be done faster, more efficiently, and more effectively. There will be greater demands on lawyers to strategize, to plan for the big picture while teams may get smaller. Tribunals and courts will face similar pressures to be as efficient and cost effective, with smaller support, smaller number of judges and panel  members. Thus, I foresee, and I am willing to bet on this, that ultimately, the workload on individual lawyers and judges will not decrease and likely increase. At a more macro level, the impact may include the possibly that cases may go down, through pressure to settle or decide, but with those that go on being more involved, perhaps with an AI arms race between counsel acting for each side.

20.          The next decade will bring to the fore the substantial challenges in the training and development of younger lawyers: much training comes currently from on-the-job exposure to routine work, which increasingly will be taken over by AI. No longer would one turn to a recently qualified lawyer to carry out basic legal research and write a simple memorandum capturing the results. Nor would one deploy new lawyers to plough through boxes of files. Such work would be mundane and often boring. But at the same time, these would have been necessary training exposing these acolytes to simple work, giving them the opportunity to get their teeth into work, making mistakes and learning from these mistakes. But these opportunities will be gone, for these mundane activities are the kinds of activities that AI already excels at with faster and better results. Crucially, AI will be cheaper. Even drafting or writing beyond the level of the beginner assistant may be taken over by AI. This displacement requires responses to be developed in training, perhaps with the use of AI. There may be alternatives including some form of simulation, though that will not be a complete replacement.

21.          Let me then come back to the point made at the start – we will not see Robot Judges or Robot Lawyers in the near or mid future. The more pertinent question is whether we in fact need legally trained people to function as either lawyers or judges. In arbitrations, IP disputes, in particular, technical expertise is perhaps more important. Perhaps not generalists, but deep non-legally trained experts or skilled professional specialists could function as deciders. Thus for engineers, scientific technical professionals, AI could conceivably supply the legal expertise, particularly, guidance on the law. So this may bring us back the medieval lex mercatoria, with merchant judges and advocates.

22.          Shakespear has Dick the Butcher say,

We do not actually need to kill the lawyers. A much worse fate is in the offing: we just lay them off.   

23.          Is that too dire a forecast? It is one plausible future. There are others. It is incumbent on all of us, as leaders in our profession, to think about the challenges that lie before us.

24.          All of us, in the judiciary, at the bar, in our professional organisations, in government, must work together, to identify likely challenges and to start planning now for the kind of world of dispute resolution that harnesses AI properly, while serving the broader needs of society. Hard decisions will be needed on manpower, on training, and development. These decisions must be confronted, lest we let an AI-centric world run away from us.

25.          Thank you.