The role of a judge is a complex one. It can incorporate activism, complex interactions with people, dispute settlement, case management, public and specific education activities, social commentary as well as adjudicatory functions that might be conducted with other judges or less commonly in some jurisdictions with lay people (juries).³ The extent to which judges are engaged in each activity varies across jurisdictions and between judges. Some judges may be more “responsive” than others, and others may show more emotion and compassion or be oriented towards therapeutic justice — interventions focussed on procedural justice that emphasise “voice” and respect.⁴ Given this variation, it is difficult to determine how developments in artificial intelligence (“AI”) may reshape the judicial role. However, the writer contends that developments may change the interactive nature of the role, varying the adjudicative function with the potential to remove judges from an adjudicative function altogether. Whilst developments in “Judge AI” or “Judicial AI” are in their infancy, there are indicators that it will become more relevant and there are already developments, although somewhat unpopular, to introduce Judge AI in relation to some categories of dispute.⁵

In terms of these developments and those in relation to AI, what will judging involve in the next 10, 20 or 30 years? More specifically, are there aspects of the judicial function that will ensure that judging remains a human activity at least in relation to some categories of dispute? Each of these questions can be informed to some extent by examining recent changes in the context of how lawyers, courts and others are currently using technology. What is abundantly clear is that the roles of those involved in justice and judging is rapidly changing and newer, more disruptive technologies have already reshaped some aspects of the justice system. Whilst the use of technology by lawyers may not immediately result in a transformation of the judicial role, it will no doubt change how some functions are exercised. For example, the shift to increasing use of AI in the form of predictive coding,⁶ predictive analytics⁷ and machine learning⁸ suggests that law firm use of AI is already changing how material is presented to judges and how client risk is assessed.

These developments have not taken place without some controversy. In the United States (“USA”), it was recently noted that predictive coding was already being used to determine whether recidivism was more likely in criminal matters and to assist in making decisions about sentencing.⁹ Importantly, many of these current developments may have an impact on judges by removing some task related functions but are unlikely to entirely reshape the judicial function or role. The writer suggests however, that recent developments in AI are likely to have a more profound impact on judges and judging into the future, and this requires us to consider the role of the judge within modern society as well as the significant issues linked to privacy, policy, intellectual property and societal and individual need that are raised by both AI and Judge AI more specifically. This article explores the nature of these developments in the context of the adjudicative role of judges and considers issues that arise when considering Judge AI and which include whether a framework exists that could enable developments to take place and if so, what particular issues arise that relate to legal authority, translating law into code, the use of discretion and understanding of syntax and semantics.

In this regard, “AI” in this article refers to a field of science, engineering and technology which focuses on the creation of intelligent machines¹⁰ and is an umbrella term which encompasses many branches of science and technology and will often involve the creation of complex algorithms to enable outcomes to be determined. AI can include machine learning, natural language processing, expert systems, vision, speech, planning and robotics.¹¹ Schatsky, Muraskin and Gurumurthy offer a practical definition of AI, stating that it is “the theory and development of computer systems able to perform tasks that normally require human intelligence”.¹² AI is an evolving concept and over time, technological advances mean that computer programs and systems become more capable of performing tasks and functions. As machines become more capable, routine tasks and functions once considered integral to AI are removed from the definition and no longer perceived to be a novelty, allowing the field to concentrate on the essential, complex functions of intelligence.¹³ Importantly, at present, most AI innovation is being led by corporate research and development processes and developments in this sector may have little regard to societal good¹⁴ or the deeper implications of AI innovation, particularly in the justice sector. Judge AI used in this context is more specifically focussed on judicial tasks — particularly adjudication. Furthermore, as noted above, the increasing use of AI to perform complex functions may include developments in affective processing that could emerge in the near future.

As noted, there are already some examples of AI informing human decision-making in the justice sector. In the USA and other jurisdictions,¹⁵ AI is already changing judicial decision-making and, in the legal sector, there are predictive analytics developments that enable predictions to be made regarding the outcome of litigation.¹⁶ The impacts of these technologies are currently emerging in some civil disputes and are forecast to have more significant future impacts¹⁷ and are particularly relevant in the criminal jurisdiction. Judge AI or, more specifically, the impact that AI may have on judging is already raising concerns amongst some senior judicial commentators. In a recent interview, Chief Justice John G Roberts Jr (USA) was asked “[c]an you foresee a day, when smart machines, driven with artificial intelligences, will assist with courtroom fact-finding or, more controversially even, judicial decision-making?” The Chief Justice responded “[i]t’s a day that’s here, and it’s putting a significant strain on how the judiciary goes about doing things”.¹⁸

As I have noted in previous work, there are three main ways in which technology is already reshaping the justice system.¹⁹ First, and at the most basic level, technology is assisting to inform, support and advise people involved in the justice system (supportive technology). Second, technology can replace functions and activities that were previously carried out by humans (replacement technologies). Finally, at a third level, technology can change the way that judges work and provide for very different forms of justice (disruptive technology), particularly where processes change significantly and predictive analytics may reshape the adjudicative role.²⁰ It is at these second and third levels that issues emerge in terms of the impact of technology on the role and function of a judge insofar as the adjudicative function is concerned.

At present, using the taxonomy above, most justice reform that is supported by technology has focussed on the first and second level of technological innovation that may or may not use very simplified forms of AI. For example, more recent technological developments supplement and support the operation of many court-based processes. As a result of this first level of supportive innovation, many people now locate justice services online and obtain information about justice processes, options and alternatives (including legal alternatives) through web-based information systems. People also increasingly locate and obtain legal support and services online, and the growth in online legal firms who may provide “unbundled” legal services has been significant over the past three years.²¹

Some web-based information (including digital video), video-conferencing (including internet-based group video calls),²² teleconferencing and email can supplement, support and replace many face-to-face in-court approaches and could be defined as a second level “replacement” technological approach. At this second level, justice is supported by technology and in some circumstances this can alter the environment in which court hearings take place.²³ For example, online court processes are increasingly used for some types of disputes and in relation to criminal justice matters (particularly bail applications).²⁴

Other technologies may merge into the “third level” and support negotiation as well as judicial processes by enabling people to access more sophisticated online “advice” that is supported by AI, or to consider options and alternatives or engage in different ways. Newer developments in legal expert systems that are focussed on predictive analytics support these shifts.²⁵ In contrast to traditional rational decision-making approaches, some of these more sophisticated technological programs are designed to encourage the development and refinement of a number of options (rather than producing one outcome).²⁶ These areas of technological innovation, at the “third level”, have the capacity to be more disruptive than previous innovations that supported a “graft and grow” approach and assumed that judging processes would not change in the context of their basic procedural stages.²⁷

In this context, there are some opportunities for AI processes to support judges and potentially supplant them. Initially, however, the impacts are likely to be confined to lower level decision-making. Such advances are not without controversy. For example, in New Zealand, Alistair Knott of the University of Otago’s AI and Law project has raised concerns about the use of a computer-based prediction model to handle claims and profile claimants under the country’s state accident compensation scheme (Accident Compensation Corporation (“ACC”)).²⁸ In Mexico, simpler administrative decision-making is already being supported by AI. For example, the Mexican Expertius system is currently advising judges and clerks “upon the determination of whether the plaintiff is or is not eligible for granting him/her a pension”.²⁹ There are, however, important issues about whether such processes will be supported in the context of judicial decision-making and as Harvey has noted, “what is at stake [in developing Judge AI] is continued confidence in and adherence to the rule of law”.³⁰

Although AI processes have emerged over the past 50 years,³¹ until the last decade they have been mainly directed at processes outside the justice sector. More recently, within the justice area they have been directed at technical as well as legal analysis. AI programs are likely to initially focus on tasks or part of the analytical function undertaken by judges and it is clear that the AI already utilised in document discovery has the potential to transform some judicial work. Current document discovery programs utilise predictive coding to read and analyse millions of pages of discovered documents and are able to select relevant material in a fraction of the time that human labour would require.³² Such programs could also be used to search through legal documentation in civil disputes, and commentators have predicted that the use of intelligent machines will increase in the legal sector — being used for the generation of legal documents and the prediction of legal outcomes (predictive analytics).³³ There are many advantages in that such AI programs are more time and cost efficient than humans and can work without stopping for sleep or breaks.³⁴ As noted above, outside of the legal profession, automated computer systems have also become prolific within government administration. These automated systems can process transactions, progress applications and make decisions without human input.³⁵

There are other pressures relating to technology that are causing a rethink of the judicial role and are linked to the creation of new court environments. In this regard, there has been a growing focus on online courts and what they may provide.³⁶ These pressures are partly a response to growing evidence of unmet legal aid, concerns about access to justice more generally and the growth in large scale online dispute resolution systems which are already being used to support some court and tribunal systems.³⁷ Such changes are arguably leading to the “democratisation of justice” and although they are not oriented towards Judicial AI, they may support the development of Judge AI by essentially building a framework which enables Judge AI to be used. At present, few of these proposals engage with Judge AI and are ordinarily focussed on increasing online activity. For example, proposals for the creation of an online court have gained traction in recent years. These developments essentially involve replacing a physical court and litigation process with an online alternative that encourages the resolution of a dispute but retains the stature and powers of a physical court of law.³⁸

The UK Civil Justice Council recommended the introduction of Her Majesty’s Online Court for civil disputes under the value of £25 000.³⁹ It was intended that the court would operate with a tiered system: the first tier would allow disputants to evaluate their problems though inputting information into an online system which would categorise their issues, provide information about their rights and entitlements, and suggest options available to resolve the dispute.⁴⁰ This tier encourages parties to resolve the dispute on their own on the basis of the information provided by the system. The second tier involves online facilitators reviewing information and documents provided by the disputants and assisting with the resolution of the matter by mediating, advising, or encouraging negotiations.⁴¹ This tier is reminiscent of court-connected alternative dispute resolution (“ADR”). The third and final tier was to involve online adjudication by the judges of the court based on electronic submissions, online pleadings and arguments, and telephone conference facilities.⁴² This determination could be binding and enforceable, with the same force as a decision made in a physical courtroom. Lord Justice Briggs suggested a similar model be introduced.⁴³ This proposal also called for the possibility of face-to-face hearings in the third tier as an absolute last resort,⁴⁴ and included details about how the online court would allow litigants to appeal a decision into the mainstream court system.⁴⁵ Lord Justice Briggs noted that the court would need its judges to adopt a less adversarial and more investigative approach.⁴⁶ In England and Wales, the plans to introduce Judge AI in relation to some categories of dispute were dropped in 2017 (less controversial but significant measures associated with the introduction of online dispute resolution referred to above are proceeding).⁴⁷

Chief Justice Warren of the Supreme Court of Victoria has suggested another model where technology is supportive: the distributed courtroom.⁴⁸ A physical courtroom remains central in this model, but the participants are replaced by life-size screens or holographic projections to enable judges, lawyers, jury members and parties to appear in court from any location of convenience. This model is facilitated through online video-conferencing technology, such as Skype, but still preserves the option of a physical space for the court, and the option of physically attending court. Should such courts be effectively implemented, the foundations for a move to an AI judge would be already in place. An AI judge at the centre of an online court program would allow litigants to provide the system with information remotely, and have a decision dispensed from within the program itself.

Similarly, developments in Online Dispute Resolution (“ODR”), a form of ADR where parties use the internet and technology to help resolve their dispute cheaply and efficiently, might also support and enable the development of AI by creating the machinery or platform within which it could eventually flourish. In ODR, disputants are not required to meet in person, as the ODR process can happen remotely through an internet connection. AI decision-making is already being used within the field of ODR. These systems are labelled expert systems, which are programmed by experts in the field and possess rule-based algorithms to assist the program to make decisions based on information received from the parties.⁴⁹ Legg explains that these processes “collect facts from users through interview-style questions and produce answers based on a decision-tree analysis”.⁵⁰

In the Netherlands, an advanced ADR program called Rechtwijzer incorporates ODR components that could be used to assist couples in the separation or divorce process. Rechtwijzer asks questions about the parties and their relationship, and provides options based on this input information.⁵¹ The program also “provides information, tools, links to other websites and personal advice” which encourages the parties to resolve their dispute between themselves.⁵² If resolution is not reached, the final step involves Rechtwijzer providing the parties with information and contact details of professional third parties such as mediators, legal representatives, and other dispute resolution processes.⁵³ Evaluations of Rechtwijzer found participants were satisfied with their experiences,⁵⁴ but a majority still felt the need to have a third party check over the agreement made through the system.⁵⁵ Whilst Rechwijzer will largely be replaced by a new system and online arrangements into the future, its creators have noted that the primary obstacle in terms of the success of such ODR arrangements relate to the incapacity of courts, lawyers and government to fully embrace these types of innovations.⁵⁶

These developments in ADR also suggest that the further introduction of AI systems into legal practice is likely. If these techniques can be used effectively within the field of ADR, then it follows that the introduction of AI programs into the court system is also feasible. Designers and implementers may draw on the experiences of these ADR programs to help perfect any AI judge programs, or alternatively AIs more specifically designed to assist judicial officers and learn from failures and successes in relation to such arrangements.

As noted above, newer technologies can assist people to resolve disputes at an earlier time or refine the issues that need to be presented to judges. For example, technology can assist people to develop options and use AI to develop alternatives, and can be used to run evaluative, advisory and determinative processes. In this regard, some disruptive technologies are linked to Artificial Legal Intelligence (“ALI”) which can be viewed as a system that has the capacity to render expert legal advice or decision-making.⁵⁷

The impact of AI on the justice system is significant as it has the capacity to be blended with existing adjudicatory or non-adjudicatory processes, and there have been questions raised about whether these processes will have an impact on the role of lawyers and judges as technology replaces some human decisionmaking and analysis processes.⁵⁸ It seems well accepted that the impact outside the justice sector is likely to be significant and there are numerous predictions that AI together with other advances will mean that many current employment arrangements will no longer exist in 20 years with many current tasks being replaced by AI supported processes.⁵⁹ However, there has so far been little discussion about more senior legal sector roles and whether these developments (and the creation of Judge AI) will mean that judicial work will change with some judges being completely replaced by newer technologies.

Clearly some aspects of judicial work will be conducted by technological processes into the future, particularly where AI systems can be built. In this regard, legal information and AI systems can already use sophisticated “branching” and data searching technology to create elaborate decision trees that can suggest outcomes to disputes.⁶⁰ In addition, more evolved AI supports systems which do not just emulate human intelligence but create additional and different intelligent systems — neural networks.⁶¹ Essentially, what takes place is that the system asks a number of questions or uses existing data about users and poses questions about the dispute to enable an accurate description of the dispute to be built. The computer then forms a conclusion by applying the law to the dispute description. It does this by applying rules for specific sets of facts. Finally, the computer can perform tasks based on the description given.⁶² This process may enable indicative decisions or even final decisions to be expressed. Such systems can be continuously updated and reflective in that machine learning enables systems to improve and be constantly revised with new data sets.

However, does this mean that judges will be replaced by technology? Arguably not, or at least not initially. This is partly because there are so many factors that impact on judicial decision-making. The Australian Law Reform Commission has noted that such factors include induction and intuition, as well as the capacity to assess the social impact of decisions.⁶³ However, if technologies can support decision-making (by, for example, enabling more accurate potential outcome identification by participants) they may play an increasing role in some forms of dispute (particularly in the family area)⁶⁴ and can support judicial processes and the making of decisions (by, for example, producing a draft or template decision that can then be considered by a human judge).

These types of technology have already been trialled⁶⁵ and have, so far, been the subject of limited extension because of connectivity, cultural, technological storage and access issues. Many of these issues are, however, declining in importance as humans become more technologically connected, and better able to store data. The increasing role that such processes will play may result in the possible diversion of more court-related disputes to ADR⁶⁶ and may also result in the replacement of more simple court decision-making processes by removing humans altogether (who may play an appellate or review function only). In this regard, the ability for AI decisions to be appealed or reviewed by human decision-makers is often cited as a necessary component of any automated decision-making system.⁶⁷

Such changes raise issues about the role of courts and judges in the future as well as raising challenging issues about how data is managed, categorised, and where and how executive and judicial functions are carried out and separated. In addition, as has been the case in the USA, there are issues about intellectual property (“IP”) and who may have control and input into outsourced Judge AI and how transparent algorithms are (see later discussion).

In addition, judges do much more than adjudicate or reach an outcome in relation to a dispute. They play a key role in case management and in the settlement of civil disputes. Judicial commentary informs how society can operate and many judges also play a role in an educative sense, both informing litigants and lawyers about approaches to be taken and also contributing to civic education at a broader level. Proponents of the view that judges can be replaced by AI are arguably missing the point in relation to what judges contribute to society which extends beyond adjudication and includes important and often unexamined issues relating to compliance and acceptance of the rule of law.

In terms of more simple adjudicatory functions it is clear that the task of performing many judicial functions requires human intelligence, and computer programs have yet to be developed to replace these functions or to interact with people with compassion, emotion or agile responsiveness. However, could advances in technology one day replace human judges in the courtroom with an AI programmed to preside over hearings and dispense more complex judgments and in what way might more affective technologies assist or support this work?

Harvey gives a simplified description of the process an AI judge would be required to take, using the example of algorithms already present in legal databases.⁶⁸ These databases employ natural language processing to assist with the sourcing of relevant material based on search terms. An AI judge would be required to go further than these databases, by reducing returned sources to a manageable and relevant sample and then deploying tools to compare these sources of law to a present case and engaging in analysis to make a determination of the outcome.⁶⁹ Harvey explains that this final step requires “the development of the necessary algorithms that could undertake the comparative and predictive analysis, together with a form of probability analysis to generate an outcome that would be useful and informative”.⁷⁰ However, human judge decision-making is largely retained in Harvey’s model.

Experiments have been conducted using AI computer programs to predict the outcomes of cases based on textual information (predictive analysis). Aletras and colleagues developed a program that textually analysed decisions relating to breaches of human rights in the European Court of Human Rights to discover patterns in judgments.⁷¹ The program learnt these patterns, and was able to predict the outcome of cases presented to it in textual form with 79% accuracy on average.⁷² This is an example of machine learning, where the computer system was able to “analyse past data to develop rules that are generalisable going forward”.⁷³ As noted, machine learning allows computer programs to learn complex tasks through experience, rather than through handcrafted computer functions.⁷⁴ Surden notes that machine learning may run into some limitations in the development of effective AIs that can predict legal outcomes. Machine learning techniques are only useful where analysed information is similar to new information presented to the AI.⁷⁵ Should an AI program be presented with a novel case where no similar precedent exists, it may not be well-suited in making a prediction or coming to an outcome.⁷⁶ These issues may also arise where the sample size of previous cases is not large enough for the computer program to discover patterns and create effective generalisations.⁷⁷

However, as AI researchers have had a number of clear successes outside of the legal field, these successes suggest that predictive analysis even where there are significant variations in terms of novelty can be “learned”. Recently, Google’s DeepMind researchers successfully trained an AI program, AlphaGo, to play the complex game of Go at a higher level than the European master of the game by training the neural networks of the program “directly from gameplay purely through general-purpose supervised and reinforcement learning methods”.⁷⁸ There are also many examples in the medical field with AI now increasingly being used for diagnostic purposes and in relation to some human functions.⁷⁹ While the law is more complex than any game, these successes suggest that Judge AI is able to learn how to apply the law by reading legislation and case law, and that applying these principles to factual circumstances is feasible. Given the developments in non-law areas and the rapid expansion of AI (and investment in this field), it seems likely that the development of more sophisticated Judge AI is probable within the next decade. More sophisticated Judge AI emerges as a more viable option as machine learning merges with more sophisticated predictive analytical processes.⁸⁰

Apart from the issues that are relevant in terms of the overall function of judges in our society, there are some specific factors that are especially relevant in the context of the development of AI and the adjudicative function of judges. Overall these factors suggest that AI can replace some adjudicative functions, however, the issues that emerge are whether this is appropriate and under what circumstances human judges should retain most adjudicative functions.

As noted above, whilst AI has the potential to replace current human judicial functions in terms of some aspects of adjudicative work, technological advances are more likely to support human judges in their judicial work. In this regard, a number of commentators have noted that the goal of the development of AI systems should be to complement current human work, allowing for greater efficiencies, rather than total replacement of humans.¹¹⁰ At times, these developments suggest that “co-bots” rather than robots will play a more important role in Judge AI.

AI programs that can produce a decision based on information input could be used to assist human judges, rather than replace them. These systems could produce a draft judgment based on the system’s determined outcome.¹¹¹ A human judge could then use this draft judgment to produce their own reasons, allowing for human oversight over the computer program, and enabling discretionary or social considerations to be made that may be beyond the capacity of the computer program.

A further technological advancement that may assist judges in their work is transhumanism. Although sounding like a concept out of the pages of science fiction, the philosophy and science of transhumanism is aimed at “fundamentally improving the human condition through applied reason, especially by developing and making widely available technologies to eliminate aging and to greatly enhance human intellectual, physical, and psychological capacities”.¹¹² Advances in this may allow the judges of the future to integrate computer circuits and programs into their bodies, or modify their physical or genetic makeup, to increase their intelligence and memory, increase their ability to manage and process information, and reduce the occurrence of fatigue.¹¹³

Perhaps the larger question is not “if” technologies will reshape the judicial function but “when” and to what extent. In this regard, disruptive technology is already reshaping the business of litigation.¹¹⁴ There are also significant changes in the way that courts are working. Technology is already being used by tribunal¹¹⁵ and court¹¹⁶ systems to provide support, intake and advisory processes that are intended to assist disputants to negotiate more effectively without having court staff or other practitioners involved. In the near future, many courts will continue to build and extend online platforms and systems that support filing, referral and other activities.¹¹⁷ These changes create a framework in which Judge AI can be fostered.

Policy approaches that increasingly result in the transfer of lower value matters or categories of disputes (such as insurance disputes) to tribunals and commissions suggest that judicial work is likely to continue to change over the next 20 years, and it is probable that AI is initially likely to play a more prominent role in tribunals and other decision-making contexts before being used in courts. However, these changes mean that the impact in respect of Judge AI is more likely to be significant, at least initially, in relation to smaller civil claims as AI support systems that are focussed more on predictive analytics spread throughout the administrative decision-making arena and are aided by online court platforms.¹¹⁸ If Judge AI (or perhaps “Adjudication AI”) is to be restricted initially to smaller claims, there are questions about where it might stop and how larger, more complex disputes will be dealt with and the extent to which judicial oversight will be maintained. In this regard, in non-legal domains, it seems that humans appear to be relatively comfortable with the replacement of some human functions by AI and robots. In some domains, there are clear advantages to using AI which can be more accurate, faster and cheaper than using humans. Will the same approach be adopted in respect of Judge AI? It seems likely that it will, particularly if decisions are made on a cost and time basis (rather than considering the broader benefits that human judges may bring to the justice system).

Any shift towards Judge AI clearly raises many issues which have so far been the subject of limited commentary.¹¹⁹ Some initial issues relate to the exercise of discretion and to what extent judges should — not could — be replaced by AI.¹²⁰ It is also important to understand that judges do far more than make decisions. They manage cases, provide a responsive and human framework, settle cases, manage court systems and processes as well as playing an important public and educative function. Drawing the boundaries of acceptable Judge AI requires consideration of ethical questions,¹²¹ as well as questions about who produces algorithms and Judge AI and the extent to which discretion and oversight will be maintained within the judiciary.

In addition, it may be unhelpful to maintain that Judge AI will only ever stand apart in some way from judges. As noted above, many technology futurists suggest that it is likely that humans will not necessarily be replaced by AI. Instead, human intelligence is likely to be supplemented by technological advances. This approach suggests that judges may remain human but be “supplemented”; that is, have their intelligence and analytical functions supported by AI. These enhancement approaches raise issues about judicial appointment, workload and retention and broader questions about how judges contribute to society, as well as the importance of responsive judging and a need to better understand and explore the impact that people experience when a human judge deals with their concerns.