Artificial intelligence in courts: A (legal) introduction to the Robot Judge

Stefan Philipsen & Erlis Themeli

The exercise of government power is increasingly automated. Modern technology makes it possible to reduce the direct human involvement in a great variety of government domains. Human involvement in domains like tax and social security is already limited to the most complex cases. One of the questions this development raises is whether artificial intelligence (AI) will also impact other branches of government, besides the executive. Last month the Estonian ministry of Justice ordered the design of a ‘Robot Judge’ to help the judiciary fight backlogs in the Estonian small claims procedure. And as futuristic as a ‘Robot Judge’ may sound, the Estonian efforts do not stand alone. Also in the Netherlands, the use of AI by the judiciary is on the political agenda. This makes sense since the use of AI by the judiciary holds many promises. Procedures are expected to be cheaper, faster, and less biased. However there are, as with the automation of executive government decisions, also concerns. In this blog we give a short introduction to the development of the Robot Judge.

The term Robot Judge offers an interesting frame in discussing the use of AI in courts. Some fear that computers may take over the world, while others are intrigued by the horizons new technologies and artificial intelligence may open. However, as the term Robot Judge is used more and more, it will probably take some time before AI is capable of replacing human judges in every court case. This holds in particular for the type of court cases that often first come to mind of lawyers. These are the high profile and highly complex cases that often deal with a great number of witnesses and court files, and regularly result in innovative legal interpretations. Despite AI being able to predict the outcome of court cases (for example decisions of the European Court on Human Rights), existing AI is far from capable of replacing humans in the most complex court cases. AI used in predicting the outcome of court cases is, at best, a first step towards such systems.

The impact of AI on courts
Judges do not immediately have to fear for their jobs. However, this does not mean that AI will not have a considerable impact on courts in the near future. The application of AI in and by courts can take at least three forms. First, AI can be used by justice administrators to better streamline their court’s caseload. In addition, the use of AI in court management promises better insight into the large amount of valuable data the judiciary possesses. Second, AI will be used more and more in courts to support judges in their judicial capacity, for example to identify, organise, and select relevant jurisprudence, to detect patterns in jurisprudence, to help highlight arguments presented by the parties, or create arguments to be used in the judgment. Such AI-assistance will touch more directly on the outcome of the case. In some countries like the US, AI is already used in bail cases to help judges assess the risk of recidivism. It is important to note that although these support systems formally leave the decision to humans, AI can have a big impact on the outcome of the case. It is understandable that when AI is used in courts to support judges, these judges may often not contradict the AI because it is not easy to contradict a system that reviews thousands of cases. This holds in particular when you have to deal with time constraints. AI in courts might therefore be like the use of AI as courts. The former must be distinguished from the latter because formal and substantive human involvement has a principled value. At the same time, the use of AI as courts does not fully exclude the possibility of some form of very general oversight or marginal review by a judge or member of the court staff. The use of AI as courts is the third way in which AI might impact the court system in the future.

Differentiate between case types
As stated, the use of AI as courts – AI completely replacing human judges – will remain futuristic for the most complex cases for the foreseeable future. However, such AI application is not completely impossible for all court cases. In private law, for example, there seem to be cases that might very well be suited for AI decision-making. These types of cases have at least two characteristics that make them less controversial to automate. We refer here to cases that are non-rivalrous and non-complex. In general, the bigger the rivalry between parties, the more attention and human involvement is required. Non-rivalrous cases are cases in which parties fully agree on the desired outcome of the case. Parties might even collaborate with each other and with the court in bringing about the desired outcome. In these situations the court is mainly granting a title. Take for example divorce cases in which both partners have an equal economic position and agree on the details of their separation, but the only way to effectuate the divorce is by a court judgment.

In addition to the level of rivalry, the more complex a case becomes the more difficult it is for an AI-system to resolve it. This refers first of all to the complexity that is directly related to the specifics of the case, like the number of witnesses, documentary evidence, and the number of separately represented parties. However, probably more important than these case specifics is the complexity of the legal rules applicable to the case. The more straight-forward a legal rule is, the less exceptions or vague terms it uses, the easier it is for AI systems to adjudicate the case. This also means that the desire to apply a ‘Robot Judge’ in the future might have consequences for the design of legal rules. Online Dispute Resolution (ODR) mechanisms that make use of AI can serve as an example. In designing such alternative ways of dispute resolution the possibility of implementing AI is often immediately taken into account.

An illustration of a non-complex rule might be the rule in civil procedure that prescribes that in the case of uncontested claims, the court must grant the claim. Such cases are not only simple they are non-rivalrous as well.

Concerns for the future
Whether it is the use of AI in the management of cases, to support judges, or in deciding cases, AI applications in courts comes with legal challenges. For example, challenges relating the protection of data. Debates on the future use of AI in courts will often also refer to human rights as an important tool for accessing the desirability of AI. This includes the right to a fair trial and closely related principles of good governance such as transparency and accountability. In evaluating the compatibility of AI with human rights, it helps to distinguish between the use of AI in and as courts. Especially with regard to the use of AI in courts to support judges, the possible human rights problems seem small. At least the conclusion will often be that the application of AI is not more problematic than the human performance. Take for example an AI system identifying legal arguments or mapping the relevant case law. Such systems will not perform worse (arguably) than a judge randomly entering search terms in a search engine. With regard to the application of AI in courts, human rights will often not set a direct boundary, but will mainly serve as a foundation for the formulation of ethical standards.

With regard to the use of AI as courts or the ‘Robot Judge’, the human rights concerns are more evident. This is mainly due to the fact that these human rights were written for situations in which a human would preside over court procedures. Human involvement is closely connected to the common perception of what a fair procedure is. And, for some cases or fields of law, like criminal law, it will be particularly important to hold on to this conception. A question that AI will explicitly put on the table is whether this also holds for the non-complex non-rivalrous civil cases distinguished above. Why should parties that agree on the desired outcome of the case not be allowed to choose for a Robot Judge, if this means that they will have a court decision in a fraction of the time it would usually take? Even if the written decision does not contain all the grounds it’s based on. Can the right to a fair trial be waived in such cases?

The future holds many unanswered questions with regard to the use of AI in courts. However, it is necessary to answer them soon since technological developments move fast – and often faster than the law. The law needs to guide this technological development.

Stefan Philipsen, Assistant Professor Constitutional Law, Utrecht University and member of the Montaigne Centre for Rule of Law and Administration of Justice, s.philipsen@uu.nl.

Erlis Themeli, Postdoc Erasmus School of Law in the ERC funded Building EU Civil Justice, Erasmus University Rotterdam, themeli@law.eur.nl.