º¬Ðß²ÝÉçÇø

A dialogue with experts in technology surrounding the growing interface between artificial intelligence (AI) and justice systems will be critical if judges are to develop a sound grasp of technological advancements, and the potential threats and challenges as well as opportunities that they will bring. The following is an edited version of the keynote speech that Singapore¡¯s Chief Justice Sundaresh Menon gave, titled ¡°Judicial Responsibility in the Age of Artificial Intelligence¡±, when the Supreme Courts of Singapore and India held their inaugural Conference on Technology in New Delhi recently to address the intersection of technology and judicial processes

A sound grasp of technological developments will become essential for our judges because technology is evolving at a very rapid pace, with far-reaching and profound implications. This has been particularly striking over the past year-and-a-half, as developments in generative artificial intelligence (AI) have taken the world by storm, and have radically reshaped our conversations about what the societies and systems of the future will look like. Mary Shelley¡¯s timeless novel, Frankenstein, was penned in the early 1800s, but its image of the creator and his ungovernable creation endures in the public consciousness today. Modern-day iterations of what has been termed the ¡°Frankenstein complex¡± have focused on the prospect of a ¡°technological singularity¡± ¨C the point at which AI will have surpassed the understanding and control of humans.

Some have suggested that we may one day also reach a point of ¡°legal singularity¡±, where AI and machines are as good as, if not better than, human lawyers and judges at providing solutions to legal problems, if not also in applying and even writing the law.

Although we are not there yet, and perhaps most of us might hope that we will not ever get there, it must be borne in mind that the capabilities of AI will only continue to grow. To borrow the words of Professor Richard Susskind, today¡¯s AI systems are ¡°the worst [they] will ever be¡±, and technology¡¯s great significance lies not in what it already is, but in what it is likely to become.

When we think of what technology might become, we have to recognise that the significance of those developments will not lie in the notion that technology will mimic what humans have historically done, or offer a cheaper, human-like alternative; rather, the significance of the ever-growing capability of technology will lie in its ability to offer solutions to the legal problems facing the world in ways that are completely different from the ways in which we lawyers and judges have been solving those problems for hundreds of years.

Just think of how Google Maps has provided a completely different solution for those who used to use a paper map to find their way; or how e-mail and smartphones have transformed the way in which we communicate and access information. These transformations had nothing to do with replicating their pre-digital counterparts. Another example is Professor Susskind¡¯s example of the autonomous vehicle that is driven not by a robot mimicking a human being, but by deploying data and computing power.

It is important to understand this because the impact of technology and technological advancements means that we stand today on the cusp of seismic shifts that will affect our justice systems in ways that we had previously not even imagined.

But as we explore the vast potential of AI and adapt to the changes taking place around us, we should first reflect on what must remain constant, on what we must remain anchored to. My central thesis today is that we must be guided above all by the goal of preserving and strengthening the rule of law, because that is the bedrock on which both our societies [India and Singapore] are founded, and its maintenance and protection are the ultimate responsibility of the judiciary.

This should guide how we discharge our two distinct but complementary roles ¨C namely, our traditional adjudicative role and our systemic role, which is emerging with rapid and growing significance ¨C if we are to ensure that the rule of law is not displaced by the rule of technology in this age of AI.

The adjudicative responsibility

Our adjudicative role as judges is our traditional role ¨C that is, to interpret and apply the law in each case in a fair and principled manner. The allure of AI and the possibility of ¡°AI judges¡± should not cause us to lose sight of the aspects of judging that remain, and should remain, a fundamentally human endeavour.

These are the aspects of our adjudicative responsibility that must endure. But at the same time, the potentially transformative impact of AI on our dispute resolution systems cannot and should not be ignored or reversed, and it will be crucial for human judges to be able effectively to manage the use of AI within the adjudicative process. Hence, there will be aspects of the adjudicative responsibility that will evolve.

The human endeavour of judging

I begin with the aspects of judging that must endure. Professor Michael Sandel has suggested that perhaps the most difficult philosophical question of the era is the role that human judgement, as opposed to ¡°smart machines¡±, will play ¡°in deciding some of the most important things in life¡±.

Although Professor Sandel was speaking of the cognitive exercise of human judgement generally, this question has particular resonance for us as judges, whose very profession revolves around the exercise of judgment on matters going to the heart of how our societies are organised. These include: decisions on a person¡¯s guilt or innocence; the appropriate punishment for offenders; and the provisions to be made for the children of a broken family. Given the weight and implications of these decisions, there are aspects of both the process and the outcomes of judging that, at least in certain fields, AI should not replace, if justice is to be both done and felt to be done.

Let me suggest three aspects of the process of judging that still require a human touch.

First, a judge should have empathy in engaging with the parties at various stages of the adjudicative process so that those parties feel that they have been heard, and that the judge has appreciated the impact that his or her decision may have on the people involved in each case.

A non-human adjudicator might be able to mimic empathy, but is unlikely to be able to exercise and convey genuine empathy and social intelligence in the way a human judge can.

Second, the process of judging must reflect the values of our justice system. Whereas these values will be instilled in the human adjudicators that are selected to serve as the face of justice in our societies, AI tools are not imbued with these values.

In particular, there is a compelling argument for criminal adjudication to be kept within the control of humans, who generally share our moral outlook and methods of reasoning ¨C especially as criminal law is, at its core, an expression of the values that our society is committed to promoting and defending.

Third, and more broadly, judging is an exercise that is intertwined with our shared humanity. As the former chief justice of the Federal Court of Australia, James Allsop has argued society¡¯s loyalty to our justice system is engendered not only by the accuracy and consistency of our output, but by the ¡°indefinable mixture¡± of the law and the judge dutifully undertaking the task of examining the dispute and applying the law with due diligence and care in a way that recognises the dignity and humanity of the people standing before justice.

This cannot be simulated by a machine, at least as far as I know. It is for this reason that some have argued that litigants may have a ¡°right to a human decision¡±.

Beyond the process of judging, I suggest that human judges and judgment may also be indispensable in producing at least some of the outcomes expected of our justice system.

First, judging often requires significant evaluative judgment ¨C it involves balancing competing considerations, which might be accorded different weights in different contexts, and which may even be incommensurable. Related to this is that judges must dispense individualised, custom-made, bespoke solutions. This requires more than just pattern detection or rule application based on existing data; instead, it calls for the exercise of discretion in applying legal principles to the facts of each new case. This may on occasion call for normative reasoning on whether a particular set of circumstances should be treated as exceptional, and whether it can be materially distinguished from the existing precedents.

Second, the development of legal principles may require judges to choose between several equally plausible interpretations of the law, or to decide whether established rules should be updated or departed from. These situations do not call for a backward-looking mechanical application of existing knowledge, but rather a conscious, reasoned and forward-looking decision on whether and how our jurisprudence should evolve, having regard to the broader societal context which the law serves.

This is especially important in common law systems, where cases are the ¡°atomistic building blocks¡± of the common law. These words of Oliver Wendell Holmes in 1881 still ring true today: the life of the law has not been logic but experience, as it ¡°embodies the story of a nation¡¯s development¡±, and it ¡°cannot be dealt with as if it contained only the axioms and corollaries of a book of mathematics¡±.

Judging in the age of AI

These fundamentally human aspects of judging suggest that the prospect of ¡°AI judges¡± replacing human judges is a distant, even a remote one. But in this age of AI, the role of the human judge will itself need to evolve. In discharging their adjudicative responsibilities, judiciaries will increasingly need to grapple with questions of how AI should, and should not, be used within the adjudicative process.

If used properly, AI can be a tremendously useful assistive tool that can improve the quality of decision-making and enable judges to surpass ordinary human limitations. AI may prove to be indispensable in dealing with the increasing technical, evidential and legal complexity that we see in many categories of disputes today.

AI tools can help judges manage and make sense of large volumes of documents and evidence, and facilitate research across vast repositories of legal precedents and authorities. AI tools could also help to identify the blind spots and subconscious biases of human judges by analysing trends in their decisions.

But these technological advancements also bring with them new pitfalls that judges must guard against to ensure that the use of AI in the adjudicative process does not undermine the rule of law. Let me highlight three key dangers.

(1) Proliferation of untruths

The first is perhaps the one that is best known, and that is what has been referred to as AI ¡°hallucinations¡± and the proliferation of untruths. Generative AI tools are potent sources of misinformation because of how quickly they can produce apparently credible output that is completely false; and unlike legal professionals, such tools are not bound by professional ethical obligations or values such as honesty and integrity.

Judges must therefore be especially vigilant to potential misuses, or careless uses, of AI tools by lawyers and parties. This underscores the need for judiciaries to take proactive steps to manage the use of AI in court proceedings, and for individual judges to be sufficiently familiar with the subject matter and the AI tools involved to be alive and alert to the possibility that something might be amiss.

While the danger of AI proliferating untruths is not unique to the judicial or legal professions, it is a particular concern for us because the rule of law is rooted in the pursuit of truth to achieve justice, and truth is the foundation of our adjudicative work. The spectre of unchecked falsehoods tainting judicial processes and decisions is therefore particularly damaging to public trust in the courts, and ultimately the integrity of our justice systems.

(2) Decline in transparency and accountability

The second key danger stems from an over-reliance on AI tools in judging. The need for judicial decisions to be reasoned and intelligible requires that judges themselves understand the AI tools that they may use, and that they are able to explain the role of those tools in the decision-making process. Otherwise, there is a real risk that aspects of judicial decision-making may become a ¡°black box¡±, lacking transparency and, perhaps more importantly, undermining accountability.

But the use of AI tools is often accompanied by an ¡°opacity problem¡±, which arises because their algorithms and underlying data sets are often undisclosed or unintelligible to laypersons, such that the output they produce cannot be meaningfully challenged. Crucially, when I speak of ¡°laypersons¡± here, this includes the great majority of judges. We may be well equipped to explain our legal reasoning, but our general lack of technical training and the opacity of the tools involved may impede our ability to properly interrogate these tools, and to understand, much less to precisely explain, why and how they have contributed to certain findings or conclusions.

As we navigate the age of AI, judges will also need to be armed with sufficient technical and domain knowledge to ensure that we understand the AI tools used and, more importantly, are conscious of their limitations.

The opacity of AI tools may not only affect whether justice is seen to be done; it may even cast doubt on whether justice is in fact done. AI tools are trained on data that may carry systemic racial, ethnic or other biases, which would undermine their reliability and often lead to unjust outcomes. Judges should therefore be vigilant in guarding against ¡°automation bias¡±, or the propensity to treat algorithmic output as authoritative without independent verification simply because it appears to be produced by an objective or scientific process.

(3) Abrogation of human aspects of judging

The third and broader danger posed by the use of AI is the risk that judges might, in using these tools, abrogate the fundamentally human aspects of judging that I spoke about earlier. As James Allsop has put it: ¡°The danger is not the machine becoming human; it is the human becoming the machine.¡±

AI tools may enable us to complete the tasks of locating, organising and synthesising material in a fraction of the time they would previously have taken, but that means we can and therefore should apply more thought and attention to the aspects of judging that require a human touch, even ¨C and perhaps especially ¨C when it might seem tempting to delegate these, too, to AI.

So, beyond cultivating technological expertise in relation to the use of AI tools, judges must also renew their commitment to their professional duties and their ethical responsibilities to exercise judgment, fully and fairly, in managing both the process and the outcomes of judging in each case.