9 September 2025
1. INTRODUCTION
Thank you, Professor Foster, for your kind introduction.
I echo the acknowledgment of the Traditional Owners of the land on which we gather this evening, the Wurundjeri People of the Kulin nation. I pay my respects to elders past and present and all First Nations persons here this evening.
I am pleased to be joined here tonight by some members of Sir Ninian Stephen’s family, Sarah Stephen, the Honourable Chris Maxwell and their daughter, Alex. I also acknowledge Mr Peter Jopling AM KC, Chair of the Menzies Foundation, and Board Member, Mr Harvey Kalman. I am delighted to be here to deliver this address as part of the Sir Ninian Stephen Law Program. Thank you all for attending.
Sir Ninian’s distinguished career of public service traversed law and diplomacy on a global scale. Beginning his career at the Victorian Bar, he was made Queen’s Counsel in 1966, was appointed to the Supreme Court of Victoria in 1970 and to the High Court in 1972. Sir Ninian’s final year on the High Court coincided with the first year of the then Justice Brennan, who observed that he ‘witnessed the working of an incisive and cultivated mind responding with seeming ease to an unrelenting caseload.’1 Sir Ninian went on to serve as the 20th Governor General of Australia from 1982 to 1989, and in the years that followed, he was Australia's first Ambassador for the Environment and a highly esteemed international judge and mediator.2
The focus of this year’s program, named in his honour, is the law and emerging technologies. In 1982, the year that Sir Ninian retired from the High Court, Time magazine departed from its tradition of naming a ‘Man of the Year’ and instead named a ‘Machine of the Year’ – the personal computer. The following year, 1983, is often cited as the advent of the modern internet. The depth and pace of advances in technology since that time is breathtaking.
This evening, I want to explore Artificial Intelligence in the judicial process.
Artificial Intelligence (‘AI’) is a broad concept that encompasses many ways in which machine-based systems collate, synthesise, catalogue and generate information.3 AI systems vary in their levels of autonomy and adaptiveness after deployment.4 While AI was first used in the 1950s,5 the mainstream use of generative AI has soared over the last few years with the advent of publicly available Large Language Models (‘LLMs’) like ChatGPT. There are now generative AI products tailored to the legal market.
AI is here. There are a number of ways to demonstrate its emergence and scale. Investment is one metric. In 2015, Alphabet, Amazon, Microsoft and Meta had a combined capital expenditure of around $25 billion USD. In 2022, the year in which ChatGPT was launched, the figure was a little over $140 billion, and by 2030 it is predicted to exceed $500 billion.6 The growth has been in data centres housing the new generation of chips to power generative AI.
Another indicator is energy use. The International Energy Agency says that electricity usage for data centres is forecast to climb from 415 terawatt hours in 2024 to around 945 terawatt hours by 2030, that latter figure is roughly equal to the electricity consumption of Japan and about 3 per cent of global electricity demand.7 As computational power grows, electricity consumption increases exponentially. The so-called ‘hyperscalers’ who are engaged in the computational arms race are constructing data centres based on the availability of energy sources and water. The four that I have mentioned have struck deals with nuclear power generators to power data centres; Microsoft is reported to be working with Constellation Energy to restart the Three Mile Island nuclear plant in Pennsylvania to provide power for its data centres.8 Presently, water is used to cool the chips and in 2023, data sites in America consumed about 55 billion litres of water.9
It might be that Sir Ninian Stephen would have been more concerned about the implications of AI from his perspective as Australia's first Ambassador for the Environment than as a judge.
It is possible that the investment will fail: a recent report from MIT made the startling claim that despite a $30 – $40 billion enterprise investment in GenAI, 95 per cent of organisations were getting zero return.10 But the point to make is that the sheer weight of investment is likely to yield significant changes in the computational power and the pressure to adopt that power into different sectors, including the law, will be enormous if not irresistible.
The potential role of AI in a judicial context is broad. At one end of the spectrum, AI can be used as a tool to augment the work of a judge and support the decision-making process, and at the other extreme end of the spectrum, AI could replace decision-making by human judges altogether.11 As the Australian Institute of Judicial Affairs notes, ‘[d]epending on the purpose of the system, and the safeguards thought necessary to be built into it, human oversight can range from human (or technology)-in-the-loop to full autonomy’.12
In courts and tribunals, we must grapple with how we can harness AI in ways that maximise its potential benefits and minimise the risks. In my view, we should focus our attention on problems that can be ameliorated or solved with the use of AI. The fact that AI can do something does not necessarily mean that it can do it to the requisite standard or replicate the judicial process. And even if it can do so now or at some point in the future, we have a responsibility to interrogate what tasks and functions it is appropriate to delegate to AI and to what extent we should do so.
I am mindful that a lack of understanding and fear about new technology also poses risks in terms of missing out on the potential benefits. Victorian Courts and Tribunals have a culture of embracing digital innovation and have adopted new technology to great effect. For example, our eFiling systems, and our transition to online hearings in response to the COVID-19 pandemic. I am confident that AI will be no different. At present there is a trial being conducted in the Coroners Court of Victoria, and initial scoping work will shortly commence in the Supreme Court of Victoria to trial the use of AI.
I am cautiously optimistic about the potential benefits of AI, and the ability of Victorian Courts and Tribunals to explore and harness the technology in thoughtful and appropriate ways.
2. THE ROLE OF A JUDGE
Chief Justice Gageler recently referred to the rise of AI as posing an ‘existential’ threat for the judiciary.13 By that I take his Honour to mean that it ought cause us to reflect on the existence of the judicial system: its attributes, characteristics, purposes and how it might interact with or coexist with AI. It suggests that the emergence of AI is not just another tool or technology that can be used, but one which might reshape how courts operate and the nature of judicial determination.
I want to make some brief observations about the judicial method in order to frame the discussion. It is only once we have in mind the way the judicial systems works, identifying its strengths and weaknesses, that we can understand how AI might usefully be employed – where it might add value and where it might undermine the system’s very existence.
The most important thing that judges do is make orders. Orders embody the judicial determination of a legal controversy and govern the rights and duties of the parties affected. The effect of orders is to change an uncertain and contentious claim into a settled outcome. They provide authority for the executive to act, including in the most profound ways, such as incarceration or the eviction from property. Orders are the end point of the judicial process. If we want orders to be made more quickly and without prevarication, then AI will be a boon. But the ability to secure compliance with orders, by force of the executive, if necessary, remains the critical touchstone of the rule of law. In turn, compliance depends on there being a sufficient well of respect for the judicial process. One cannot divorce the validity of the order and the acceptance that it garners from the process by which it is made.
The question I want to explore is whether there is any role for AI, especially generative AI, in the process that leads to the making of those orders. To begin, I want to touch on some aspects of the judicial process: fact-finding, the exercise of discretion and the giving of reasons. These aspects cannot be divorced from and are bound by some fundamental conceptions.
First, we operate in a common law system. That system contains a body of rules and legal principle, but also a method. Whether or not described as ‘a strict and complete legalism’,14 it has a discipline that focuses on the case at hand and the strict application of principle, but which is capable of evolving to capture new perspectives or reflect changing times. It is not an inquisitorial system, in which a court sets the boundary of inquiry, but rather involves the adjudication of practical disputes tendered by the parties for determination.
Second, the power and authority of a court is conditioned by an obligation to accord procedural fairness, often described by reference to the two pillars of an impartial mind free of any taint of bias and an obligation to provide a reasonable opportunity for an affected party to be heard before a decision is made.
Third, the power is exercised in public, in open courts. Open, transparent justice is a critical aspect of our judicial system. In Russell v Russell, Justice Stephen (as Sir Ninian then was) noted that ‘a tribunal which as of course conducts its hearings in closed court is not of the same character as one which habitually conducts its proceedings in open court.’15
Sir Frank Kitto observed that:
‘The process of reasoning which has decided the case must itself be exposed to the light of day, so that all concerned may understand what principles and practice of law and logic are guiding the courts, and so that full publicity may be achieved which provides, on the one hand, a powerful protection against any tendency to judicial autocracy and against any erroneous suspicion of judicial wrongdoing and, on the other hand, an effective stimulant to judicial high performance.’16
3. THE USE CASE FOR AI
When one looks at the judicial system in Australia, there is much to be proud of. It involves the fair and independent determination of disputes by application of settled legal principles by an expert and well-trained judiciary. If it were prefect or could not be improved, then there would be no case for AI intervention. But nobody thinks that it is a perfect system.
When considering the potential role of AI in our judicial system, we should focus our attention on the problems that AI might address. The so-called ‘use case’.
The most common criticisms of the Australian judicial system are that it is slow, expensive and not accessible to everyone in the community. These are significant and enduring criticisms.
Delay is a significant burden on litigants and the community. It is a symptom of the volume of cases initiated, their complexity and the level of judicial resources available to hear and determine them. Judging is time consuming.
Costs are often both high in absolute terms and may seem disproportionate to the value of the claim.
There are also concerns about decisions that are made. Concerns are expressed about consistency and whether decisions are explained in a way that can be comprehended by those in the community.
Within the judiciary, some judicial officers feel overburdened. There is the burden of the number and complexity of cases that they have to deal with. In the criminal context, the sheer volume and speed of legislative change may be difficult to keep up with. In the civil context, technology has vastly increased the number of documents that may be produced and court books running into thousands of pages are not uncommon. Both the legal and factual material that judges have to master in a given case are often significant. Judgment writing takes an enormous amount of judicial time.
We are also increasing our understanding of the risk of secondary and vicarious trauma that can be associated with viewing or reading about distressing events.17 This is a potential problem not only for judicial officers but also for their staff.
Is there a role for AI to address these issues: to speed up judicial decision-making, to reduce costs and increase access to justice? Our legal system is one of individualised justice in which each case is considered on its merits. This takes time. There is no doubt that AI could be employed to significantly decrease delays and to increase efficiency in courts and tribunals, but in what circumstances is this appropriate, and at what cost?
4. THE ROLE OF AI
In considering the potential use of AI in our courts and tribunals, it is understandable that there will be discussion about the accuracy and capability of AI.18 Australian courts have already experienced receiving documents generated by AI that are inaccurate, misleading and which contain fictitious references. It is likely that this capability issue will be addressed both by a change in practice, including a recognition of the responsibility that exists on every party that relies on the contents of a document in court to ensure its accuracy, and an evolution in AI accuracy. I see these challenges as not existential but practical. It is unlikely that they provide a sufficient reason to resist entirely AI on the march.
The issues about the use of generative AI in the judicial process are more fundamental. They go to the extent to which the adoption of AI in judicial decision-making would threaten transparency and judicial independence if underlying algorithms are confidential and impenetrable, limiting understanding and critique of the decision-making process.19 While AI has the capacity to increase access to justice, it also raises issues of equity which must not be overlooked.20
Can I focus then on fact-finding, the exercise of discretion and the giving of reasons.
Fact-finding
Litigation is always backward looking. What happened as a matter of fact is the most important issue in the overwhelming majority of cases. Once the facts are known, the application of legal principle will often follow without too much controversy. And, it is the uncertainty of where the facts will land that encourages settlement of cases.
In those cases where the facts are to be found, the process involves the parties selecting the evidence they wish to adduce, the calling of witnesses in open court and the tendering of documents. In some cases, it will be the oral evidence of witnesses that will prove the most important and in others it will be the documents. It is now rare that a witness’ oral account is not tested and challenged against some documentary record.
There seems little doubt that AI will significantly affect the availability of evidence and the process by which evidence is located, collated, synthesised and adduced by the parties. The digital footprint has grown, and with the development of AI and vastly increased computing power, the amount of data will grow exponentially. Vast digital databases recording the details of our lives are not the sole province of the dystopian authoritarian regime. This will inevitably lead to an increase in the amount of material that is available for the purposes of any litigation. Emails, correspondence, file notes, purchase orders, contracts will increasingly have been created, modified or edited by a LLM. The clinical notes of doctors may be generated by AI from a recording of the consultation and may proffer differential diagnoses or suggest the need for pathology or radiology testing.
Expert evidence, which covers the full gamut of fields of expert knowledge, will be informed by AI. Where the expert is called upon to conduct analysis, modelling or estimation, it is likely that the various disciplines will have incorporated AI to a greater or lesser extent. Where the claim is for damages, the analysis often requires consideration of the counterfactual: what would have happened had the wrongdoing not occurred? This hypothetical exercise is often the critical element in the calculation of loss. Again, it is inevitable that the modelling required to fashion and consider the myriad potential alternatives will be aided by AI.
In this context, there will be the risk that AI will be used illicitly to manufacture evidence, including documents and images. In the context of fact-finding, AI presents serious challenges in relation to the authenticity of evidence and the potential inaccuracy and unreliability of AI-generated information which litigants and their legal representative may present to the court. These issues were recently explored by Chief Justice Bell of New South Wales.21 Again, it is likely these can be addressed by vigilance and with technological solutions. The question of authenticity is obviously important for courts but not uniquely so.
In preparation for trial, AI may assist the parties in the efficient identification of relevant documents and in ordering and summarising their contents in a way that will assist in case preparation. At trial, it is possible that AI will be used by counsel to summarise a witness’ evidence as it is given, identifying points of conflict with other parts of the evidence, to aid a challenge or warn the cross-examiner about lurking dangers and what topics to avoid.
What then of the judge? Would it be acceptable in a trial for the judge to use AI to assist by preparing a summary of the transcript of the evidence given that day? What about preparing an annotated chronology that the judge might find useful in following the evidence? Can summaries and categorisation, which AI does very well, reduce the need to view or read distressing material?
And then when the judge retires to her chambers having reserved judgment, is it acceptable for the judge to use AI to help summarise the evidence, highlight areas of conflict or even resolve factual disputes?
Judicial discretion – Sentencing
What about in the exercise of judicial discretion, meaning the power the law gives a judge to choose among several lawful alternatives,22 which must be exercised reasonably?23 Judges will be influenced by their personal experience and perception of the judicial function, and may be guided by intuition.24 However, they must also act from the perspective of the ‘reasonable judge’ who is governed by society’s values and sense of justice.25 An example of the exercise of judicial discretion is the sentencing process.
Imagine that judges have access to a reliable, secure AI platform, which has high quality sentencing data – including judgments, sentencing principles, sentencing statistics – into which can be uploaded all the evidence in relation to a case. The judge asks AI to recommend a sentence, or a sentencing range. The judge interrogates the output, spars with the AI. Why did you recommend that sentence? What factors did you take into account in reaching that recommendation? Present the arguments in favour of the lowest and highest sentences in the range. The outcome would favour consistency.
Reasons for judgment
What (if any) role should AI play in judgment writing? It is undeniable that AI is orders of magnitude faster than a human at summarising and synthesising information and could accelerate this process.
There is no one uniform approach to judgment writing, and ways of working vary between judges. However, in my experience, it is conventional for a judge to write the judgment, aided by an associate who is available to conduct legal research when required and proof-read the draft. This approach contrasts with the US where it is typical for junior members of appellate judge’s chambers, typically law clerks, to draft judgments.26
A judge could direct AI to produce a complete first draft of a judgment for the judge to review. It could identify and apply the relevant sentencing principles to the facts of the case, just as the AI medical notes prompt a differential diagnosis.
Less controversially, AI could be tasked with summarising the facts, the parties’ submissions and/or the applicable law.27 In the meantime, the judge could focus on drafting the sections of the judgment which contain the reasoning and conclusion.
Might this enable the judge to focus on what really matters in the deciding of a case, freed from the burden of setting out matters of context or uncontroversial history?
The problems and the opportunities
In each of the areas of fact-finding, articulating the relevant legal principle, identifying the range of potentially open discretionary decisions and in judgment writing, there is potential for AI to be deployed to reduce the time taken for each step. Is it consistent with the judicial process?
There are a number of issues that need to be confronted.
The first is that it is wrong to treat the different phases or tasks that I have identified as discrete aspects that can be managed or treated in isolation of one other. The steps taken by a judge from summarising the evidence to writing reasons for judgment are connected and form part of a single task. Justice Kitto has observed in relation to the exercise of judicial power: ‘…the process to be followed must generally be an inquiry concerning the law as it is and the facts as they are, followed by an application of the law as determined to the facts as determined.’28
Making findings of fact from a summary that the judge did not prepare may not be the same thing as making findings based on the evidence. The process of synthesis or summation can affect the emphasis that different parts of the evidence enjoy or diminish the weight of evidence that is not mentioned or developed in the summary. On the other hand, resources such as chronologies, summaries and lists of issues are often used to help the judge frame his or her consideration of the evidence. One of the most important steps in understanding the evidence and being in a position to make findings is ordering the material in a logical and comprehensible form. Further, LLMs can index and footnote a summary, enabling the reader to have direct access to the relevant underlying source material.
The second point goes to the nature of the judicial task. As already adverted to, it is not inquisitorial. It is sometimes said that it is the parties’ case. That means that it is for the parties to identify the issues that they choose to litigate.
Chief Justice Gageler has observed that a judge ‘is tasked not with the independent pursuit of some ultimate truth but with arbitration of a contest between parties who assert different versions of the truth.’29 Nor is it for the judge to wade through the material and try to come to some just answer free from the case presented. The use of AI to review the material, identify or resolve points of conflict or identify the legal issues that arise might be quick, the result might be correct, but does it represent a significant departure from the proper role of the judge?
The third related point is that judicial fact-finding is not a mathematic exercise based on probability theory. It is well established that the process of weighing up evidence and reaching a conclusion to the requisite standard cannot be reduced to a mathematical formula.30 The notion that fact-finding, which is not mathematical in nature, would be entrusted to AI based on algorithms, risks altering the nature of the process.
That is, fact-finding is often highly qualitative. Appeal courts recognise the advantage that a trial judge enjoys in having seen and heard the witness and watched the evidence unfold.31 A review of a documentary record is not the same thing as a trial. AI may well reach a point of sophistication where it could assess the veracity of a witness’ oral statement more accurately than a human. However, even assuming that it could – and this capability could be assessed objectively – would we ever feel comfortable assigning it that task when the stakes are so high?
Fact-finding is a delicate exercise undertaken by judge or jury. In my experience, it is done with extensive care and reflection. In deciding issues at trial – for example, whether a witness is telling the truth – juries are invited to draw on their life experience and common sense.32 In my view, fact-finding should remain the responsibility a human tribunal of fact, regardless of the capability of AI and the efficiencies that could be gained by outsourcing this task.
Fourth, the process of judicial decision-making must be transparent. The evidence and arguments which the parties seek to advance are exposed in open court. Any questions the judge has for the parties are also ventilated in open court. This process allows the parties to deal with concerns that the judge may have, and which arise during the hearing. To what extent should the parties be told about a judge’s use of AI? Should this extend to the identification of the product, the prompts that the judge uses or by providing a copy of any product? It is established that a judge does not have to expose his or her thinking process to the parties. Nor does a judge tell the parties what tasks he or she has given the associate. In my experience, it is common for judges to discuss the issues with their associate as part of the process of formulating the judge’s own view. This dialogue remains between judge and associate. Should we treat AI use as analogous to an associate or as analogous to the judge seeking external assistance to help decide the case?
Fifth, although AI could improve consistency, consistency of outcome is often in tension with individualised justice. Courts are cautioned against overreliance on current sentencing practices and sentencing statistics, which have the capacity to distort the instinctive synthesis.33 Sentencing statistics have been described as ‘a very crude guide as to the appropriate sentence in a case.’34 Courts have also cautioned against categorising offences as ‘low’, ‘mid-range’ or ‘high’ because it risks assessing current sentencing practices for previous offending of that type without considering the circumstances of the specific case.35
When AI is used to produce a ranking, score, recommendation or decision in the sentencing context,36 this moves beyond supporting or informing judicial decision-making, into the realm of exercising discretion. Reliance on such tools could threaten judicial independence.37
Sixth, judgment writing is a means of accountability. Reasons for judgment further judicial accountability.38 Using AI to draft a complete judgment would be an abdication of duty and one that does not account for the cognitive process of human decision-making. For many judges, putting fingers to the keyboard and drafting those earlier sections of a judgment is part of the process through which the case is digested, and reasoning is formulated. In this sense, the decision-making process is iterative, not linear, and the process of deciding a case and articulating the reasons for it are often inextricably linked.
A judgment informs the parties of the reasons for the court’s decision, which facilitates trust and confidence in the legal system. Sir Harry Gibbs, former Chief Justice of the High Court, observed that reasons are useful to a critical if not sceptical audience who ‘should be convinced that justice has been done, or at least that an honest, careful and conscientious effort has been made to do justice, in any particular case, and the delivery of reasons is part of the process which has that end in view.’39
They also equip an appeal court with the information required to determine whether the trial judge made a jurisdictional error. As precedent, they also play a vital role in the development of the law.
Judgments are the tip of the iceberg in the judicial system and the operation of the law more generally. They occur in a fraction of cases brought in the courts. They operate beyond the individual resolution of rights in that case to inform legal advice and legal adherence. AI tools are trained on existing material, and we are seeing significant demand for judgments as source material to train AI. We value the body of evolving court precedent as an enabler of broader decision-making. I suggest there is a use case for retaining the production of human wisdom, both for its own sake but also as the basis from which AI external users can draw.
5. CONCLUSION
There are strong reasons to be cautious about the use of generative AI in the judicial process. Judges are humans. Judicial decision-making is subject to human frailty. Many of the problems of delay and expense are a product of the fact that it is a very intensive process. Could AI increase the quality and/or consistency of decisions made, and is this a problem that needs to be solved?
In my view, there is inherent value in human judges making decisions.40 Fundamentally, courts are human institutions.41 Justice Stephen (as Sir Ninian then was) observed that ‘Courts necessarily reflect community values and beliefs’42 and they do that through the judges that are appointed. Sir Gerard Brennan, former Justice of the High Court, observed that in presiding over a criminal trial a judge needs to ensure that no one in the court room feels humiliated or has their dignity diminished.43 Values such as mercy cannot be quantified.
Article 24(1) of the Victorian Charter of Human Rights and Responsibilities provides that:
‘A person charged with a criminal offence or a party to a civil proceeding has the right to have the charge or proceeding decided by a competent, independent and impartial court or tribunal after a fair and public hearing’44
If AI makes a decision – or even assists a human to make a decision – can it be said that the charge or proceeding was decided by ‘a competent, independent and impartial court or tribunal’?
Sir Geoffrey Vos has suggested, in the context of article 6 of the European Convention on Human Rights,45 that machines probably cannot or should not be properly regarded as ‘an independent and impartial tribunal’ within the meaning of article 6 because they are a function of their training and cannot have human emotions.46
Based on current capabilities, it is an open question whether AI can be classified as a ‘competent, independent and impartial court or tribunal’ in the context of the Victorian Charter of Human Rights. A key underlying reason is that the very nature of AI is mathematical, stochastic, predictive. It reverts to the mean/average. While it seems to be generating or creating new information, it is a derivation of the data it has been fed or trained on. The ability to consider the unique nature of an individual case is a fundamental aspect of a judge’s role which AI cannot currently replicate.
The nature of judicial decision-making can be distinguished from many other areas of human enquiry that can be enhanced by AI. In areas such as meteorology forecasting or radiology, AI has the edge on the human brain when it comes to analysing big data and identifying patterns and trends. Judicial decision-making differs from many other areas in this respect. Judges make finely tuned, highly considered and qualitative assessments of individual cases. In a judicial context, I consider that the human brain is capable of absorbing and analysing sufficient information to make a sound, reasoned decision. While there is scope for AI to support judges and increase efficiency in certain regards, I am not convinced that substituting machines for humans would improve the quality of decisions made.
In my view, there are aspects of the judicial function which should not be entrusted to AI – fact-finding, the exercise of judicial discretion, and producing reasons for judgment. While the capability of AI and the public’s perception of it will necessarily evolve over time, my current view is that these integral elements of the judicial function should remain the province of human minds, replete with values, creativity, life experience and imperfections. To a real extent, the problems associated with the judicial process – time and cost – reflect the human involvement in the process.
But the use of generative AI is not necessarily inconsistent with human decision-making. In my view, AI is a prime candidate for tasks that involve identifying and collating relevant data, the acceleration of which would free up time for other work. Although these are small examples, when scaled across a large court, the cumulative effect of these efficiencies would not be insignificant.
There is potential for AI to assist in the realm of case management, including triage and allocation of cases.47 Inside courtrooms, both physical and virtual, AI may facilitate faster and cheaper transcription services.48
AI could reduce the instances of court staff being unnecessarily exposed to distressing material. This is a potential benefit which is currently being explored in the AI pilot program in the Coroners Court of Victoria.49
There is scope for AI to be utilised to facilitate and increase access to justice. For example, to provide swift, clear information to court users, including Self-Represented Litigants, in relation to process.50 It could also assist with communicating the law and outcome of cases with the public by producing succinct, clear summaries of cases as soon as they are handed down.
Finally, I foresee the potential for AI to be deployed to assist with low-value, high-volume claims in certain areas at some point in the future, which could reduce delays and costs. I note that the Singaporean judiciary have been collaborating with Harvey AI to explore the use of AI in the Small Claims Tribunal, including to assist Self-Represented Litigants.51
The challenge is to discern what is appropriate augmentation and what is tantamount to unacceptable substitution. There are a number of potential approaches: prohibit the use of AI altogether, disclose its use or trust judges to use it in a manner consistent with the judicial method. Last month, Justice Aidan Xu, Judge of the General Division of the High Court, Supreme Court of Singapore, observed that:
‘The aim is not to replace the submitted materials, and to depart from the adversarial process, but to better enable the judges to recall and digest information, pursue lines of questioning, and examine the coherence and strength of what has been put before them.’52
It will be important that the use of AI by the courts does not endanger trust. That the community, armed with the knowledge of what generative AI can do, will not simply assume that judges who use AI will use it to decide cases and produce reasons. This requires the courts to be open about the use of AI and will be aided by the publication of guidelines, preferably on a national basis.
We should focus our attention on issues in courts and tribunals that AI can ameliorate and solve. Explore, interrogate and collaborate – across courts and jurisdictions – and with the technology itself. Certainly, we must be mindful of the risks and challenges AI poses, but not to the extent that we miss out on harnessing its positive potential in the right contexts.
To that end, courts must understand the AI environment. We must be open to trial new technology to ensure that its use is consistent with the values that underpin the system. This will involve both investment and collaboration across courts and tribunals, within Victoria and across Australia, and partnerships with universities and research institutions.
AI is an evolving tool. Human minds created it, and I am optimistic that human minds can also harness it and take full advantage of its potential in appropriate ways in Victorian Courts and Tribunals. However, we must be vigilant and ensure that AI is only employed in ways which serve the best interests of the public and do not undermine trust and confidence in the judicial system.
[1]Father Frank Brennan SJ (ed), Gerard Brennan’s Articles and Speeches (Connor Court Publishing, 2025) vol 1, 18.
[2]See, eg, Justice Michael Kirby, ‘Sir Ninian Stephen – Internationalist’ (Speech, the University of Melbourne Faculty of Law Dinner in Honour of Sir Ninian Stephen on his 80th Birthday, 24 July 2003).
[3]Supreme Court of Victoria, Guidelines for Litigants: Responsible use of Artificial Intelligence in Litigation.
[4]OECD, ‘Explanatory Memorandum on the Updated OECD Definition of an AI System’ (OECD Artificial Intelligence Papers No 8, March 2024) 4.
[5]David Spiegelhalter, The Art of Statistics: Learning from Data (Penguin Press, 2020) 185.
[6]Tabby Kinder, ‘‘Absolutely immense’: the companies on the hook for the $3tn AI building boom’, Financial Times (online, 14 August 2025).
[7]Tabby Kinder, ‘‘Absolutely immense’: the companies on the hook for the $3tn AI building boom’, Financial Times (online, 14 August 2025).
[7]International Energy Agency, Energy.and.AI.(Report, April 2025) 14, 49, 63.
[8]Constellation Energy, ‘Constellation to Launch Crane Clean Energy Center, Restoring Jobs and Carbon-Free Power to The Grid’ (Media Release, 20 September 2024); Natalie Sherman, ‘Microsoft chooses infamous nuclear site for AI power’, BBC.News (online, 21 September 2024).
[9] Hyperscale and colocation sites. FT reporters, ‘Inside the relentless race for AI capacity’, Financial.Times (online, 31 July 2025), citing Arman Shehabi et al, 8680.United.States.Data.Center.Energy.Usage.Report.(Report, Lawrence Berkeley National Laboratory Publications, 20 December 2024) 55.
[10] Aditya Challapally et al, The.Gen.AI.Divide¿.State.of.AI.in.Business.868❶ (MIT NANDA, July 2025).
[11] Professor Sourdin uses the terminology ‘supportive Judge AI’ and ‘Judge AI’ – Tania Sourdin, Judges?.Technology.and.Artificial.Intelligence (Edward Elgar Publishing, 2021).
[12] The Australian Institute of Judicial Administration Inc (‘AIJA’), AI.Decision‗Making.and.the.Courts¿.a.guide.for.Judges?.Tribunal.Members.and.Court.Administrators (Guide, June 2022) 32.
[13] Michael Pelly, ‘The Stephen Gageler interview in full’, Capital.Brief (online, 8 July 2025).
[14] Swearing.in.of.Sir.Owen.Dixon.as.Chief.Justice (1952) 85 CLR xi, xiv.
[15] (1976) 134 CLR 495, 532.
[16] Sir Frank Kitto, ‘Why Write Judgments?’ (1992) 66 ALJ 787, 790.
[17] See, eg, C Schrever, C Hulbert and T Sourdin, ‘The psychological impact of judicial work: Australia’s first empirical research measuring judicial stress and wellbeing’ (2019) 28 JJA 141.
[18] See, eg, George Hammond, ‘AI sceptic Emily Bender: ‘The emperor has no clothes’, Financial.Times (online, 20 June 2025).
[19] COMPAS, a privately-owned risk assessment tool that is used in the context of sentencing in certain US states, has been critiqued on this basis.
[20] See, eg, Victorian Law Reform Commission (‘VLRC’), Artificial.Intelligence.in.Victoria"s.Courts.and.Tribunals (Consultation Paper, October 2024) 27, 79-80, citing Australian Human Rights Commission, Human.Rights.and.Technology (Final Report, 2021).
[21] Chief Justice Bell of New South Wales, ‘Change at the Bar and the Great Challenge of Gen AI’ (Address to the Australian Bar Association, 29 August 2025).
[22] Aharon Barak, Judicial.Discretion (Yale University Press, 1989) 7.
[23] Ibid 115-120.
[24] Ibid 121-4, 133-5.
[25] Ibid 124-133.
[26] Richard A Posner, Reflections.on.Judging (Harvard University Press, 2013) 238-248.
[27] Lord Justice Birss, a judge of the Court of Appeal of England and Wales, who specialises in intellectual property law, has acknowledged using ChatGPT to summarise an area of law for a judgment. Hilbaq Farah, ‘Court of appeal judge praises ‘jolly useful’ ChatGPT after asking it for legal summary’, The.Guardian (online, 15 September 2023). In the United Kingdom earlier this year, guidance for judicial officers on the use of AI, first published in December 2023, was updated – see HM Courts and Tribunals Judiciary, Artificial.Intelligence.(AI)¿.Guidance.for.Judicial.Officer.Holders, 14 April 2025.
[28] R.v.Trade.Practices.Tribunal·.Ex.parte.Tasmanian.Breweries.Pty.Ltd (1970) 123 CLR 361, 374; [1970] HCA 8.
[29] Justice Stephen Gageler, ‘Alternative Facts in the Courts’ (2019) 93 ALJ 585, 589.
[30] See, eg, R.v.GK (2001) 53 NSWLR 317, 323 [26] (Mason P), cited in Al.Mahmoud.v.DPP.(Cth).[2021] VSCA 240, [25] (Priest and Niall JJA).
[31] See, eg, R.v.Baden‗Clay (2016) 258 CLR 308, 329 [65] (French CJ, Kiefel, Bell, Keane and Gordon JJ); [2016] HCA 35; Pell.v.The.Queen (2020) 268 CLR 123, 144–5 [37]–[39] (Kiefel CJ, Bell, Gageler, Keane, Nettle, Gordon and Edelman JJ); [2020] HCA 12.
[32] Cavanaugh.(a.pseudonym).v.The.Queen [2021] VSCA 347, [273] (Walker JA).
[33] Ashdown.v.The.Queen (2011) 37 VR 341, 398 [167] (Ashley JA); [2011] VSCA 408.
[34] Russell.v.The.Queen [2011] VSCA 147, [61] (Kaye AJA).
[35] DPP.(Vic).v.Weybury [2018] VSCA 120, [33] (Maxwell P and Hargrave JJA), [54] (Priest JA); Lee.v.The.Queen [2018] VSCA 343, [31]–[32] (the court).
[36] Eg, COMPAS in the US.
[37] Monika Zalnieriute, ‘Technology and the Courts: Artificial Intelligence and Judicial Impartiality’, Submission to the Australian Law Reform Commission, Review.of.Judicial.Impartiality (4 June 2021).
[38] Soulemezis.v.Dudley.(Holdings).Pty.Ltd (1987) 10 NSWLR 247, 279 (McHugh JA).
[39] Sir Harry Gibbs, ‘Judgment Writing’ (1993) 67 ALJ 494, 494.
[40] I make these observations mindful that the technology is not static and there may indeed come a time when AI can replicate human emotion and creativity.
[41] Chief Justice James Allsop, ‘Technology and the future of the courts’ (2019) 38 University.of.Queensland.Law.Journal 1, 2.
[42] Onus.v Alcoa.of.Australia.Ltd (1981) 149 CLR 27, 42.
[43]Sir Gerard Brennan, ‘The Role of the Judge’ (Speech, National Judicial Orientation Programme, 13 October 1996) 7.
[44] Charter.of.Human.Rights.and.Responsibilities.Act.866❷ (Vic).
[45] European Convention on Human Rights. Article 6, Right to a fair trial, provides that: ‘[i]n the determination of his civil rights and obligations or of any criminal charge against him, everyone is entitled to a fair and public hearing within a reasonable time by an independent and impartial tribunal established by law.’
[46] Sir Geoffrey Vos, ‘Keynote speech’, (Human rights, Algorithmic Justice, and Global AI Policy Conference, Global Policy Institute, Durham University, 15 May 2025) [38]-[41], [48].
[47] See, eg, Supreme Court of Victoria, Submission.No.8❷.to.VLRC?.Inquiry.into.Artificial.Intelligence.in.Victoria"s.Courts.and.Tribunals (2025) 2-3; AIJA, AI.Decision‗Making.and.the.Courts¿.a.guide.for.Judges?.Tribunal.Members.and.Court.Administrators (Guide, June 2022) 27; Tania Sourdin, Judges?.Technology.and.Artificial.Intelligence (Edward Elgar Publishing, 2021), 120-5.
[48] See, eg, Supreme Court of Victoria, Submission.No.8❷.to.VLRC?.Inquiry.into.Artificial.Intelligence.in.Victoria"s.Courts.and.Tribunals (2025) [11]; AIJA, AI.Decision‗Making.and.the.Courts¿.a.guide.for.Judges?.Tribunal.Members.and.Court.Administrators (Guide, June 2022), 28.
[49] Coroners Court of Victoria, Submission.No.0.to.VLRC?.Inquiry.into.Artificial.Intelligence.in.Victoria"s.Courts.and.Tribunals (2025).
[50] See, eg, Supreme Court of Victoria, Submission.No.8❷.to.VLRC?.Inquiry.into.Artificial.Intelligence.in.Victoria"s.Courts.and.Tribunals (2025) [11]; AIJA, AI.Decision‗Making.and.the.Courts¿.a.guide.for.Judges?.Tribunal.Members.and.Court.Administrators (Guide, June 2022) 38. I note Singapore’s experimentation with AI chatbots to assist litigants and decision to shelve this for the time being – Justice Aidan Xu, ‘The Use (and Abuse) of AI in Court’ (Speech, IT Law Series 2025: Legal and Regulatory Issues with Artificial Intelligence, 6 August 2025) [29].
[51] Justice Aidan Xu, ‘The Use (and Abuse) of AI in Court’ (Speech, IT Law Series 2025: Legal and Regulatory Issues with Artificial Intelligence, 6 August 2025) [27]-[28]; Chief Justice Sundaresh Menon, ‘Judicial Excellence in a Challenging World: The Centrality of Trust’ (Speech, International Association for Court Administration Conference, 12 November 2024) [31]; VLRC, Artificial.Intelligence.in.Victoria"s.Courts.and.Tribunals (Consultation Paper, October 2024) 4.
[52] Justice Aidan Xu, ‘The Use (and Abuse) of AI in Court’ (Speech, IT Law Series 2025: Legal and Regulatory Issues with Artificial Intelligence, 6 August 2025) [17].