Where the Machine Stops: Technology and Justice in Singapore
It grows harder each day to begin a presentation or article on technology, especially AI, without sounding like AI. That rapid ubiquity of AI, and indeed, of technology, makes it difficult to see how human clients, lawyers and judges can make a difference. Yet that human contribution lies at the heart of what we seek to preserve in the courts. Technology, including AI, is a tool: it can help solve some problems, but it is never the only answer and is, more frequently than one would expect, the wrong one.
Technology is often shiny, enjoyable to use and expensive. These qualities make it dangerous: we can become blinkered in rushing to adopt the latest technology as the solution to problems that may not exist. It draws headlines, captures the boss’s attention, advances careers and gives us a seeming edge over competitors. The temptation to keep pressing the technology button is palpable and persistent. That is why the courts seek to remain grounded when considering technology. We ask whether it will improve access to justice, make us more effective and maintain public confidence.
Confidence is perhaps the most important of these aims because it underpins the other two. The justice system works only if people trust that decisions are reached fairly, according to law and without fear or favour. Courts must avoid measures or systems that erode that trust. Any technological solution must therefore be robust, reliable and sufficiently transparent to sustain confidence.
Effectiveness requires us to remember that technology should never be employed for its own sake. It should make our work easier and help produce better outcomes, whether in the courtroom or in back-end administration. Efficiency is part of this: tasks should be completed faster and at lower cost. But we must also reach the right answer as reliably as possible.
Finally, as part of our systemic role, we must consider whether technology can facilitate access to justice. Again, this does not mean deploying technology for its own sake. Technology can improve access, but sometimes clearer instructions or simply a person available to answer questions will do more.
Confidence
Confidence in the justice system is the keystone that holds everything together. The justice system works because people generally accept its authority. They may disagree with a judge’s decision, but there are established procedures for dealing with such concerns. But there must remain sufficient confidence that the decision was reached fairly, according to law, by someone who considered the case without fear or favour. Technology must fit within that framework.
A person who comes to court expects the judge to apply his or her mind to what is presented. The judge cannot abdicate responsibility to another person, organisation or technology.
As technology is introduced, especially AI that could mask an abdication of human decision-making, guidelines, monitoring and auditing are essential to safeguard judicial processes. Our internal instructions are intended to provide those safeguards. Even when AI is used for something as apparently mundane as summarisation, we must monitor its output for accuracy.
Effectiveness
For the courts, effectiveness means reaching the right outcome as reliably as possible, with proper attention to the evidence and arguments. That objective has many dimensions.
Evidence adduced by the parties must be placed on the record and made readily available for judges to refer to, scrutinise and weigh. Witnesses should be able to testify effectively in the courtroom, with ready access to documents and other materials. Parties should receive accurate records that they can use seamlessly in argument and on appeal. These are only some of the elements needed to reach the right outcome. We seek to help parties, witnesses, lawyers and judges achieve that objective.
Transcription is an important starting point. Preparing and checking the record of proceedings takes time and resources. Removing that burden would benefit court staff, judges and parties. But the challenges are significant. Courtrooms contain different accents, multiple speakers, interruptions and sometimes more than one language. Their acoustics serve a different purpose and may not support accurate microphone capture. Sensitive microphones may also pick up conversations and asides not intended for others. None of these problems is insurmountable, but addressing them requires a higher level of accuracy and fidelity than an ordinary meeting, discussion or telemarketing call.
Technology can also improve effectiveness by helping judges manage growing demands. Judicial work at every level is becoming more complex as parties present greater volumes of evidence, including emails, WhatsApp chats, photographs and other documents. Submissions are longer, cover more issues and cite more authorities from Singapore and abroad. Ironically, wider use of AI may intensify some of these trends. Whatever the cause, mastering the evidence and submissions has become increasingly challenging. We have therefore been exploring how AI can assist in these areas.
The Courts have encouraged the use of government-provided AI tools, whose capacity to synthesise and process large volumes of material has improved steadily. We are grateful to our colleagues at GovTech and other agencies for their work. Questions remain about the proper sanitisation and protection of information. Restrictions and controls are necessary, but they create friction in using AI tools. We are seeking the right balance between preventing information leakage and enabling wider AI use. Other obstacles also inhibit adoption: our filing system does not yet support AI-assisted summarisation and querying across the board. We hope to address this as we develop our future case management system.
Beyond judicial work, substantial back-end administration is essential to the smooth running of the courts, just as it is in law firms and other organisations. AI can provide useful support for such routine work. We have been exploring government-provided products such as Copilot to lighten that administrative burden.
Access
Technology, especially AI, could greatly improve access to justice by giving court users accurate information about how to proceed, what is happening and what steps may be available next. The information is often available online in forms, rules, guidance and other resources. The difficulty lies in understanding and synthesising it, and applying it to the user’s circumstances.
Court users are also often under stress, whether as claimants or defendants. They may be dealing with unemployment, family breakdown or the possible loss of a home. We should not assume that someone approaching the courts has the time or composure to master our procedures. Providing useful information is part of our role. We can explain the process without taking sides; impartiality does not require us to leave people confused. Some helpful measures involve little or no technology: better processes, clearer language or simply a person available to answer questions. At other times, technology can provide concrete assistance.
Our work with Harvey AI in the Small Claims Tribunals illustrates how AI can assist productively. Translation and summarisation help parties understand the documents before them, including material submitted by the other side, so that they can follow the proceedings and participate effectively. This is a real benefit where lawyers are not permitted and would, in any event, probably be uneconomical.
We previously experimented with chatbots to assist members of the public, but found that they struggled with nuance and with materials that were related but not directly relevant. We set the idea aside, but may revisit it as AI improves. This reflects another aspect of our approach: we experiment, assess costs and benefits realistically, and try again when circumstances change.
Throughout all this, we will work to ensure that the human touch is present. We aim to be digital ready, and digital first to serve the needs of the majority of our users, but we will not be digital only. We do intend to make sure there is layered assistance for different needs.
Where the Machine Stops
This brings me to the title of this piece, inspired by a short story by E M Forster, “The Machine Stops”, that is worth reading. Here, however, I want to emphasise what we do not intend to use AI for.
Given AI’s ease of use and availability, there may be a temptation to use it not only to summarise, query or draft sections, but to formulate a decision and its reasons. It takes only a prompt. But that would amount to an abdication of the judicial role. We expect the human, not the machine, to judge. Attitudes may change: our children’s children may be entirely comfortable with machine judging. For now, however, it remains a clear boundary that we will reinforce.
Even when a judge retains discretion, but uses AI extensively, we have to be careful to ensure that the judge retains independence of thought, and does not let the tide of AI fluency sweep him or her to a particular conclusion: the judge must question and test what AI produces. We need to make sure that this capability is retained.
Despite these risks, we think there is nevertheless room to use AI for work surrounding the decision: drafting or summarising facts and arguments, refining language, and stress-testing reasoning through red-teaming. The AI can be asked to identify possible weaknesses, such as overlooked evidence, underdeveloped arguments or unanswered points raised by the parties. This may expose something we have missed. Not every suggestion will be sound; the judge must decide which warrant further attention.
Another area for exploration is the use of technology in high-volume settings to generate indicative assessments or possible resolutions for parties to consider. Where data sets are large and outcomes readily predictable, indications might be generated through rules-based systems or AI, which the parties can consider and weigh accepting, in the interests of time and convenience, but which can be fully argued before a human judge if need be. Such indications would not constitute judicial decisions. We know that this is a sensitive area, and we will explore its boundaries cautiously.
Governing the Machine
We have issued various guidelines on the responsible use of AI. Our general policy on AI use in court submissions is simple: if you use AI, you own the consequences.
Although there have been many horror stories about hallucinations in submissions by self-represented persons and lawyers, and even in judgments, their incidence should decline as AI training improves and more material becomes available. The Attorney-General’s Chambers (AGC), the Ministry of Law (MinLaw) and the Singapore Courts have made legislation and case law freely available to the public. This has expanded the material available for AI training, and AI models are now much more accurate when citing and referring to Singapore legal materials. With suitable prompts, particularly on paid models, we are approaching the point at which even general-purpose AI can readily produce reasonably accurate reports and summaries of Singapore law. That may create other challenges that warrant further exploration. Internally, we have guidelines for judges and administrators, with mechanisms in place to assess risks.
Looking ahead, we see two main challenges. The first is AI-fabricated evidence, whether videos, photographs or documents. An arms race is likely between AI systems used to fabricate evidence and those used to detect it. We must also consider other measures for identifying forged or fabricated material, and zooming in on provenance, authentication, security of metadata and traceability to ensure authenticity.
The second challenge is volume. Overseas, concerns have arisen that AI encourages more claims and more voluminous filings. If it helps people pursue worthwhile claims that would otherwise not reach court, that is welcome. The risk is that it may also generate unmeritorious claims at scale and overwhelm the courts. We have not yet seen that here, and our systems and processes may discourage it, but we must monitor whether such a trend emerges.
A broader concern, raised by the Chief Justice and my colleagues, is that AI may erode basic skills: sustained reading, summarisation, and the examination and development of ideas. We must consider how to keep judges and administrators sufficiently skilled. As I have noted in various fora, this is also a longer-term challenge for the profession.
The ability to assess work usually comes from having done it. Researching, drafting and wrestling with an argument develop the judgment needed to recognise an inadequate answer. If AI takes over much of that work, we must think carefully about how those skills will be acquired. Telling a junior lawyer to exercise judgment does not teach the skills underpinning that judgment, and it is doubtful, I think, that such skills can be developed without being exposed to actual situations.
Regional and International Exchange and Cooperation
An important part of what we do is to maintain links and exchange with other judiciaries around the world. Aside from bilateral meetings and discussions with our partners, we organise a regional technology forum for the courts, mostly in the Asia-Pacific, including many Australian jurisdictions, China, India, the Dubai International Financial Courts, South Korea and many of our ASEAN partners. This helps all of us keep abreast of developments, and issues. Importantly, it allows us to see that there are many possible responses to common challenges. We can weigh the experience of others, and consider our own approaches tailored to our specific contexts.
Conclusion
This brief article provides a picture of where the courts are, where we hope to be and what we aim to do. It is not offered as a model for other judiciaries, nor do we suggest that law firms should simply emulate it. Different institutions face different challenges.
There are significant challenges ahead for all of us, for the world in fact, in coming to grips with these advances in technology, especially AI. We do not have oracles to prophesy exactly and definitively where we will end up. All we have are ourselves, and each other. We will all need to be flexible and responsive to a changing environment. The effort will be hard, but it is something that MinLaw, AGC, the Bar, the Academy, the Universities and the Judiciary can work at together, to make it easier and for the benefit of us all.
