Martin Karner: is artificial intelligence a substitute or a supporter of a judge?
The pressure on the efficiency of the court system will not diminish in the future, but will rather increase. Therefore, we should continue to search for solutions that would help the courts to process disputes faster than before or reduce the number of disputes, writes Martin Karner.
The steam engine was invented in the 18th century and industrial revolutions started. The first wave involved a transition from handiwork to mechanized production. In the course of the second industrial revolution in the late 19th and early 20th centuries, electric energy was introduced and industries learned to produce in large quantities and for a low price.
The development of information technology and the Internet in the second half of the 20th century entailed the third industrial revolution, which is also named as the first information revolution. The internet spread around the world, and computer-based knowledge systems started to affect the daily work.
Currently, in the early 21st century, the fourth industrial revolution or the second information revolution, is occurring with the enormous growth of the computation power of computers. This again has created an opportunity for artificial intelligence to emerge. Artificial intelligence gives the possibility to utilize skilfully all the existing knowledge by combining it with computation power.
The digitization of the judiciary process started at the end of the third industrial revolution. However, for the first time, the fourth industrial revolution, aka second information revolution, provides an opportunity to make the judicial power very fast, efficient and of high quality.
In Estonia there are currently as many judges as there were 20 years ago; however, the number of civil matters to be adjudicated has doubled. The number of matters has increased mainly on account of simpler debt claims, and there are also other factors that contribute to more efficient administration of justice. However, one of the most important factors is definitely the development of technology, which makes it possible to automatize the handling of disputes to a greater or lesser degree.
The pressure on the efficiency of the court system will probably increase in the future, rather than decrease. It is therefore important to continue searching for solutions that would help courts to resolve disputes more quickly or to reduce the number of disputes.
Net Group proposes a hypothesis that artificial intelligence may be one of the tools that could be used to improve the efficiency of the administration of justice. The question is not if, but rather to which extent and when. The article presents some examples of how artificial intelligence has already been used or an attempt has been made to use it as a substitute or at least as a support for the administration of justice. We also provide ideas on how to introduce similar solutions in Estonia.
The main objective of Net Group is to initiate a discussion among people working in the Estonian court system. Only through discussions it is possible to detect, where the judge’s ”shoe actually pinches”, to be able to start looking for a solution. The solution does not necessarily have to be to replace a human judge with artificial intelligence but to use artificial intelligence as a helping tool at some stage of judge’s work.
What is artificial intelligence?
In most cases, artificial intelligence refers to the creation of complex algorithms to forecast the results of processes or to detect patterns. Artificial intelligence is definitely not the only technical tool that should be considered for increasing the efficiency of administration of justice.
In the Yearbook of Courts, the Chief Justice of the Supreme Court highlights a problem that digital signing of decisions is time-consuming, which is why justices of the (Supreme) Court still sign decisions on paper. It is not necessary to develop solutions based on artificial intelligence to solve such problems. Adapting the legal framework for digital signatures and developing a relevant technical solution will probably suffice.
How has artificial intelligence been used in the administration of justice?
Computers, databases, software and self-learning systems have frequently replaced human work. It is said that computers substitute in case of routine work, while creative work remains with people. This is true. However, what is routine work and what is creative work?
Many people who do routine work think they are creative, which, as they later learn, might not be the case. This is all the more surprising when decision-making processes that only a few ”gurus” think they are proficient in, are hacked with algorithms. Then a software robot makes decisions hundreds of times faster than the human.
The same applies to court proceedings. Many decision-making processes are probably routine, predictable, and can be carried out faster and more efficiently by software robots.
Examples include the verification of the receipt of a state fee if an action is filed and the verification of other data to be submitted, but also the appointment of a judge, booking of the time of the hearing, analysis and forwarding of simpler letters and preparation and sending of summonses. Providing evidence, weighing of arguments and making the decision – these stages require thinking more outside the defined framework: ”if-then-else”.
As the examples below show, even the drafting of the final decision can be left to the computer, as the computer is able to find patterns from previous precedents and big data, provide the core of the decision and write the initial text of the reasoned final decision.
If all routine activities and routine decision-making were given to software robots, it would be possible that the task of the future judge would ”only” be to serve justice. It literally means a fair assessment of circumstances and delivering the final judgement. The judge would have more time for this in the future, because the computer has presented all the arguments of the case at the speed of lightning, and even assessed their quality.
The judge can then assess the arguments, proceeding from their value, social criteria and context. A computer might not and could not consider fully all these factors.
Two extracts from the Yearbook of Courts 2018 by the Chief Justice of the Supreme Court Villu Kõve:
“Man vs. machine.” I believe that in the long run, court proceedings will become paper free, but now the user-friendliness of information systems still leaves much to be desired. Digital processing of large matters is a particular problem, because we require smarter solutions in order to cope with the whole volume of information. Solutions that give the judge an overview of the current state of the court matters before them would also be beneficial. For example, if a judge has a hundred matters simultaneously at hand, managing these all is very complicated. Information technology and artificial intelligence could be helpful, if these were smartly put to work for users, based on the internal logic of court proceedings.”
”I have a habit of looking through both the State Gazette and the Official Journal of the European Union every morning, and the amount of information I see is frightening. I think that even if someone wished to stay informed about everything that is published in the Official Journal of the European Union, it is simply not physically possible, since every day brings so much new information. We should find solutions for efficient management of this flood of information. Maybe this is where artificial intelligence solutions could be utilized? How to make sure that judges receive the information they require for their work, both about legislation and judicial practice? For example, issues related to civil proceedings are rarely reflected in the Official Journal of the European Union, but when it happens, this is very important information. In my opinion, we should pay more attention to the delivery and mediation of information, while also considering how to motivate judges to improve themselves.”
Dear readers, who work in the court system on a daily basis – have you wondered, how the digital tools of the 21st century could help you work more efficiently? We have discussed this with several colleagues and on the grounds of their perceptions we present a vision of where a software tool could help a court clerk to work substantially more efficiently. I claim that it is possible to automatize work processes, make the administration of proceedings simpler and logically structure decisions.
Even activities that usually take a few minutes (such as making a query to a register) add up to a major loss of time. It is important that the main worktime of a judge is used on resolving essential issues. Technology can be utilized to create relevant arguments and establish analogies, leaving therefore more time for the decision-maker to look deeper into the matter. We believe that through more efficient administration of proceedings we can make the work more pleasant and easier for judges.
We present our vision with the example of the processing of one simplified case. We have divided the functionalities that simplify work processes into three.
Administration of a proceeding
The described vision does not cover all the bottlenecks of the current judiciary process, but hopefully, we will still be able to demonstrate the capabilities of the modern technology. Artificial intelligence is not behind all the outlined solutions, but it could be one option for solving the problem. We hope that you will be inspired by our description and will want to join us in thinking about where the bottlenecks in the work of court clerks are and what are the efficient solutions.
We are not talking here about the automation of decision-making, but about the automatics of those work processes, which do not create substantive value. We use the following case as an example, calling it conditionally KIS 3.0. With the help of this case we describe new functionalities that can be built on an existing platform.
Description of the case. On August 1, 2021, Jüri Tamm files an action with the Harju County Court. After the application, the KIS 3.0 proceedings administration view shows automatically a new case, where it needs to be decided, whether to accept the case.
In brief, the application of Jüri Tamm said the following:
”I took my car VW Passat, number 001TTT, to maintenance at the company OÜ Mehaanik on October 1, 2018. On the following day the engine of the car stopped and after taking the car to another workshop for repairs, it turned out that there was no oil in the engine and the engine replacement costs 3200 euros. I hold OÜ Mehaanik responsible for causing the damage and claim from the company the 3200 euros + 1500 euros for time loss and stress involved. I also request that an expert confirm the opinion that Mehaanik OÜ has made a mistake.”
The application is written in free text on several pages.
A new case appears in the proceeding administration view of a court employee, who must decide, whether to accept the action. Participants in the proceeding, requests and payment of the state fee have been automatically identified in the statement of claim and entered as metadata into the information system, and also displayed separately in the proceeding administration view.
Relevant queries have been made about the parties to the proceeding (automatic standard queries, such as from the population register about active legal capacity, etc., and about essential objects in the case, such as the vehicle).
The results of the queries are also displayed in the detailed view of the proceeding. This means that when deciding on whether to accept the case, it is immediately visible that Jüri is the owner of the vehicle and the legal representative of OÜ Mehaanik is Priit Kask.
If necessary, it takes one click to make queries also to other registers (such as the land register), with results displayed immediately in the KIS 3.0 view. A court clerk can quickly and without leaving the KIS 3.0 view decide on acceptance and initiate a new workflow.
If the judge decides to accept the action, the system automatically draws up an order on acceptance, which is pre-filled with substantive information, provides the defendant a standard time limit for response and sends out the order.
Administration of the proceeding
After that it is possible to view and manage separately the following sections in the administration view: activity; parties to the proceeding; circumstances; requests; requirements; documents..
When the defendant receives the action, they submit a response, where they state that the action is unfounded and should be dismissed and that the limitation period of the action has expired.
The plaintiff Jüri, in turn, sends a large number of additional documents to confirm his statements of case, where the text includes the sentence: ”I request immediate seizure of the current account of OÜ Mehaanik, because to my knowledge the financial situation of OÜ Mehaanik is bad and money is being extracted from the company.”
This ensures that it is possible to assess immediately the urgency of solving the action and that nothing is overlooked. In the view of requests it is also possible to monitor the status of all requests.
Automatic preparation of an order for securing an action: if the judge decides to satisfy the request, the system automatically draws up an order for securing the action, pre-filled with substantive information, and sends the relevant information to the bank after signing of the order.
In all documents submitted to the courts, court employees can simply mark any necessary facts and add these along with a time stamp to the chronological sequence of events. In the present case, for example, the clerk can mark the following: 1) when Jüri had his car repaired at Mehaanik OÜ; 2) when the car engine stopped; 3) when he requested compensation of damages from Mehaanik OÜ, etc. In extensive cases, the chronological timeline of events makes it easier to get an overview of the case.
Intelligent search and finding similar parts of text
The plaintiff and the defendant have submitted many documents and explanations in the proceeding. KIS 3.0 provides several easy possibilities for intelligent searches. By intelligent search we mean a search that can find answers to a keyword or phrase even if these look different in the text (e.g. different inflections, misprints).
Furthermore, the search recognizes other words that have a similar meaning in the context (for example, a search for ”expert analysis” will also return parts of the text that talk about expert assessment). Intelligent search is able to find all those sections, which have a connection to the searched phrase.
For example, if marking a description of a circumstance in the statement of claim (a passage of text), an algorithm will search the entire case for similar parts of the text (for example, finds the sections, where the defendant has responded to the plaintiff about the circumstances).
Use of previous practice, knowledge management
After adding a claim from the statement of claim to the claims section of the proceeding, an algorithm automatically recognizes the type of claim, which in the current example is a claim for damages.
Prerequisites for this type of claim are added automatically to the claim. Since the claim in the present case is a claim for damages, a list of prerequisites appears next to the claim.
The court clerk can now check conveniently within one system the preconditions for satisfying or rejecting a claim. Such tool helps to create a logical structure for satisfying or rejecting claims in court judgements. It is possible to manage all prerequisites individually, add comments, view their status. In their response the defendant has previously indicated that the action has exceeded the time limit, therefore the court clerk can add a prerequisite of exceeding the time limit to the prerequisites of the claim.
KIS 3.0 gives the possibility to display the texts of sections of law in documents without having to search for these, for example, in the State Gazette. The algorithm is also able to display the practice of the Supreme Court (or a lower court) regarding the relevant section.
In order to find the correct judicial practice and the relevant part of the text, it is possible to search, using narrower, case-specific keywords. In case of the current example, we are interested in compensation for damage on the basis of the Law of Obligations Act (LOA), breach of contract and compensation for damage.
What is already being done elsewhere in the world?
The overview of the use of artificial intelligence in law discussed in this article is definitely not complete. The objective is to give an idea of what is possible and how the solutions of artificial intelligence can affect the judiciary process.
Utilisation of artificial intelligence in court systems is rather modest. The exception is China. In the European Union attempts have been made to allow artificial intelligence to decide some legal issues independently.
In France, the Rennes and Douai courts tested the Predictrice tool designed to automatically determine the amount of redundancy pay in the event of dismissal. The aim was to bring clarity and harmonize the payment of redundancy pays. The project was terminated, because the tool failed to take into account all the nuances and determine the payment fairly.
In the Netherlands, an artificial intelligence solution has been developed for an online court system, making decisions in some types of debt recovery proceedings. Studies show that under very limited conditions, artificial intelligence can be highly successful in making independent decisions, if certain preconditions have been met.
In the United States, the COMPAS system is used to assess recidivism, such as in case of deciding on bail and release from prison. This is an instructive example of how much care must be taken, when using artificial intelligence in the legal system. The artificial intelligence is trained on the grounds of previous decisions and because of this the algorithm can make biased decisions. In case of COMPAS it has been found that it considers the recidivism rate of black people to be higher than it actually is.
Australia is using the Split Up system. This artificial intelligence assists judges in divorce proceedings. The artificial intelligence helps to decide, which assets to divide and in which proportions.
In Finland, a project named ANOPPI has resulted in the development of tools for the automatic anonymisation and pseudonymisation of judgments as well as automated creation of content descriptions to make judgments more accessible to the public.
In Estonia personal data are also automatically anonymised in court judgements. This does not involve artificial intelligence, but is nevertheless a good example of the feasibility of automation.
In the use of artificial intelligence, the courts of the People’s Republic of China are well ahead of others. The Hebei Supreme Court in China has developed an intelligent court management system Intelligent Trial 1.0, which helps courts automatically digitize statements, classify documents, search for relevant laws, judgments and documents, automatically create documents (such as notices) and coordinate tasks in workflows.
The court system in Beijing uses a robot Xiaofa, which can answer to 40,000 litigation questions and is able to handle 30,000 legal problems. There are more than a hundred robots working in Chinese courts to help the courts in various issues.
In the city Hangzhou, a cyber court was founded in 2017. Citizens can turn to the cyber court through WeChat (a popular messaging app) and the court is run by artificial intelligence via video, involving all parties. The cyber court has jurisdiction over online trade disputes, copyright and webstore product liability issues.
A similar cyber court will soon be opened in other courts. Such cyber courts have conducted more than 3,000,000 legal proceedings, received around 120,000 litigations and pronounced judgements in almost 90,000 matters (as of the end of 2019).
Courts can exploit machine learning also in a completely different way. If artificial intelligence is capable of reading detailed information from court judgements, it is also able to analyse unintentional bias.
Each judgement or reasoning provided by a judge can be placed in a context (taking into account circumstances) and see, whether the judgements delivered by the judge are statistically significantly different from the judgements of other judges. Bias can also be viewed in respect of other parties: for example, if suspects with certain characteristics treated differently from others or if requests by attorneys of one sex are rejected more often than that of the other sex, etc.
Private sector utilizes artificial intelligence much more than courts. Artificial intelligence solutions of the private sector do not have a direct effect on improving the efficiency of the work of courts, but they nevertheless affect the work of courts.
One field, where artificial intelligence is used, is forecasting. Researchers have predicted the rulings of the European Court of Human Rights with 79% accuracy and the rulings of the United States Supreme Court with 70% accuracy.
A Canadian start-up Blue J claims that they are able to predict judgments in tax issues with 90% accuracy.  While the above-named examples are simply facts of academic interest, the private sector has made artificial intelligence with its predictions to work for them. Companies such as Ravel, Lex Machina and Bloomberg Law not only predict judgments, but also the specific behaviour of the parties, such as how a particular judge reacts to a specific type of request or activity of a party to the proceeding.
The company Ross Intelligence offers a legal analysis service, which processes legal documents. In addition to making predictions, Bloomberg Law also deals with intelligent search, automatic classification of documents, and intelligent workflow management. The product Points by Law by Bloomberg Law analyses the wording of judgements and based on the analysis gives to lawyers wording recommendations for legal documents (such as applications) in order to have a higher likelihood of achieving the desired result.
With certain types of disputes, going to court used to be common. Today, platforms can arise alongside the court, for resolving such disputes. The application Wevorce offers a service that organizes divorces and also helps to divide assets. In the United Kingdom, Keogh Solicitors and St John’s Buildings barristers’ chambers developed a fully automatic system for insurers and solicitors, which processes injuries resulting from traffic accidents.
The system initiates the proceeding digitally and decides on the type of the litigation.  Both solutions use artificial intelligence, which has helped them to make their solutions more efficient and competitive.
While the above-named solutions can lower the work load of the courts, the opposite is also possible. Artificial intelligence can help to automate some processes and make going to court significantly easier and less costly. 19-year-old Joshua Browder developed the DoNotPay app, which helps to dispute parking fines quickly and easily.
In a simple user interface, the artificial intelligence asks questions from those who wish to dispute the fine, and then gives them recommendations. In a little more than a year, 250,000 cases reached DoNotPay, resulting in 160,000 annulled fines (four million dollars in parking fines were not received).  By today, DoNotPay has evolved into an artificial intelligence lawyer, which enables you to sue other people in several areas with a single touch of a button.
Parking fines and other fees in England and Wales can be appealed online: https://www.trafficpenaltytribunal.gov.uk/want-to-appeal/.
The potential of artificial intelligence is demonstrated also by the LawGeex project, where artificial intelligence was made to compete against 20 professional lawyers. Their task was to work through five nondisclosure agreements in four hours and identify threats to the client.
The artificial intelligence solution had not seen these nondisclosure agreements, but had been trained to identify threats with the help of machine learning on tens of thousands of nondisclosure agreements. For lawyers it took an average of 92 minutes to review the five agreements and they were able to identify an average of 85 per cent of the threats (the best result was 94 per cent). Artificial intelligence reviewed the five agreements in 26 seconds and identified 94 percent of the threats. 
In the light of the world practice, it would be worth to reason further, could an artificial intelligence judge make decisions independently in Estonia, under certain limited conditions? The well-known technology magazine Wired has already published a story that Estonia is starting to create an artificial intelligence judge. The news had a lot of resonance in the world. Indicative of the future direction is the fast track proceeding for payment order, but the same could be applied to actions for support, judgments rendered in absentia, actions for quick loans and compensations paid by airlines.
We welcome response to the ideas presented here. We are ready to participate in the introduction of new solutions with our advice and capacities.
The article appeared initially in Yearbook of Courts 2019
 In detail at: https://www.netgroup.com/blog/lets-dive-into-data-science/. Throughout the article, the authors have also used: https://ec.europa.eu/info/sites/info/files/justice_scoreboard_2019_en.pdf, as well as information received at interviews with judges and advocates general.
 V. Kõve. Development of the legal and court system. Presentation at the General Assembly of judges on February 8, 2019 in Tartu. Yearbook of Courts 2018, pp. 19 and 21. On a computer network.
 F. Z. Gyuranecz, B. Krausz, D. Papp. The AI is now in session – The impact of digitalization on courts (2019). On a computer network.
 H.W.R. (Henriëtte) Nakad-Weststrate, A.W. (Ton) Jongbloed, H.J. (Jaap) van den Herik, Abdel-Badeeh M. Salem. Digitally Produced Judgements in Modern Court Proceedings. International Journal of Digital Society (IJDS), Vol 6, Issue 4 (December 2015). On a computer network.
 E. T. Israni. When an Algorithm Helps Send You to Prison. The New York Times (26.10.2017). Online: https://www.nytimes.com/2017/10/26/opinion/algorithm-compas-sentencing-bias.html.
 J. Wu. AI Goes to Court: The Growing Landscape of AI for Access to Justice. (6.08.2019). On a computer network.
 A. Hietanen. Finnish Project on the Anonymization of Court Judgments with Language Technology and Machine Learning Apps. On a computer network.
 J. Wu (reference 6).
 B. Harris. Could an AI ever replace a judge in court? (11.07.2018). Online: https://www.worldgovernmentsummit.org/observer/articles/could-an-ai-ever-replace-a-judge-in-court.
 In brave new world of China’s digital courts, judges are AI and verdicts come via chat app. The Japan Times (7.12.2019). Online: https://www.japantimes.co.jp/news/2019/12/07/asia-pacific/crime-legal-asia-pacific/ai-judges-verdicts-via-chat-app-brave-new-world-chinas-digital-courts/#.Xhbkq25uKUl.
 F. Z. Gyuranecz, B. Krausz, D. Papp (reference 3).
 In detail at: https://www.bluejlegal.com/.
 D. Hargrove. Technology predictions for 2020 – the impact of AI in the legal sector (25.12.2019). Online: https://www.itproportal.com/features/technology-predictions-for-2020-the-impact-of-ai-in-the-legal-sector/.
 S. Gibbs. Chatbot lawyer overturns 160,000 parking tickets in London and New York. The Guardian (28.06.2016). Online: https://www.theguardian.com/technology/2016/jun/28/chatbot-ai-lawyer-donotpay-parking-tickets-london-new-york.
 In detail at: https://www.lawgeex.com/resources/AIvsLawyer/.