ABSTRACT

1 INTRODUCTION

The extent and intensity of technology use in courts vary significantly across regions and countries. Undoubtedly, one of the main factors behind such uneven an distribution of technology is legal regulation. The United States, China, and the European Union are competing to become the global leader in finding the right balance between effective regulation and the widespread application of digital technologies. In general, there are three different competing regulatory approaches: the American market-driven model, the Chinese state-driven model, and the European rights-driven regulatory model.¹

Artificial Intelligence (AI) is deemed the key enabling technology,² and with the European Union's ambition to position itself as a global leader in developing cutting-edge, trustworthy AI,³ it raises important questions: What types of AI tools are suitable for adoption in European countries? Specifically, in the context of courts, to what extent can AI be integrated into judicial processes? Critically, can AI be applied to the core of judges' work, decision-making? Could it fully take over the adjudication role?

In this context, two distinct models of applying AI in courts can be observed. The first model uses AI as a decision-support tool to enhance judicial efficiency, accuracy, and consistency while retaining final authority in human judges. The second model envisages AI as an autonomous decision-maker, or so-called robot judge, capable of conducting proceedings and issuing binding decisions with minimal or no human involvement. These contrasting approaches, supportive versus substitutive, form the analytical framework of this article.

Accordingly, the article explores the role of AI in judicial decision-making, focusing on identifying the distinctive European approach to this issue. To do that, the article is structured as follows. Section 2 examines the core aspects of the European approach to AI in adjudication, emphasising the balance between AI support and human oversight, adherence to ethical principles, implementation of safeguards, and analysis of key policy documents. Section 3 analyses the potential for AI to assist judges in the decision-making process while preserving human oversight, whereas Section 4 delves into the prospects of AI replacing judges in decision-making altogether. Finally, Section 5 concludes by summarising the article’s main findings.

2 EUROPEAN APPROACH TO AI IN ADJUDICATION

The European approach to AI in adjudication has developed through the combined efforts of both the Council of Europe and the institutions of the European Union, united by a commitment to ensure that technology serves rather than supplants human judgment. In 2018, the Council of Europe’s European Commission for the Efficiency of Justice (CEPEJ) recognised AI as one of the technologies with the greatest potential to enhance the efficiency and quality of justice, while warning against its uncritical use for predicting or automating judicial decisions.⁴ That same year, the CEPEJ adopted the European Ethical Charter on the

Use of AI in Judicial Systems, which affirmed that the deployment of AI in courts should improve efficiency and consistency but must always respect fundamental rights, judicial independence, and the right to a fair trial.⁵ Parallel discussions were unfolding within the European Union: following a 2017 request by the European Council to develop a coordinated European approach to AI,⁶ the European Commission’s 2018 Communication AI for Europe identified justice as a key application area and introduced a dual approach, promoting innovation while ensuring adherence to the Union’s values and ethical principles of accountability and transparency.⁷ Subsequent EU policy documents, including the 2019–2023 European e-Justice Strategy and Action Plan,⁸ further distinguished AI as a promising but high-risk technology, whose development must be accompanied by safeguards in data protection, ethics, and fundamental-rights compliance. Collectively, these initiatives reflect a distinctly European approach to AI in adjudication, one that encourages technological innovation while maintaining the primacy of human oversight, judicial independence, and fundamental rights.

Building on these initiatives, European institutions articulated a more detailed ethical and governance framework for the use of AI in the justice system. Between 2019 and 2021, a series of soft-law instruments were adopted by the European Commission, the European Parliament, the Council of the European Union, and the Council of Europe, each reinforcing the principle that AI technologies should support judicial activity without undermining human oversight or judicial independence.

In 2019, the High-Level Expert Group on AI, established by the European Commission, presented the Ethics Guidelines for Trustworthy AI, marking a key step in shaping the European vision of human-centric AI. Among the seven requirements for trustworthy AI identified in the guidelines, particular importance is placed on human agency, oversight, and respect for human dignity. In the context of adjudication, these principles mean that AI systems should act as tools serving and protecting human autonomy rather than replacing it, and that final decisions must remain under meaningful human control through mechanisms such as human-in-the-loop or human-in-command oversight.⁹ The same year, the European Commission issued the Communication Building Trust in

Human-Centric AI, reaffirming that AI systems should support individuals in making better, more informed choices in accordance with their goals. They should function as enablers of a flourishing and equitable society by strengthening human agency and fundamental rights, and should not decrease, limit, or misguide human autonomy. The Communication further stressed that human oversight helps ensure that an AI system does not undermine human autonomy or cause other adverse effects.¹⁰ In the White Paper on AI (2020), the Commission noted that the appropriate type and degree of human oversight may vary from one case to another, depending on the intended use of the systems and the effects that the use could have for affected citizens and legal entities.¹¹ This reflects the flexible yet indispensable role of human supervision, which is particularly relevant in the context of adjudication. Building on this understanding, the Commission’s 2020 Communication Digitalisation of Justice in the European Union: A Toolbox of Opportunities applied these principles directly to the justice sector. The document emphasised that final decision-making must remain a human-driven process. Only a judge can guarantee genuine respect for fundamental rights, balance conflicting interests and reflect the constant changes in society in the analysis of a case.¹²

The European Parliament also echoed this human-centric approach in its Resolution on a Framework of Ethical Aspects of AI, Robotics and Related Technologies, adopted in October 2020. The Resolution emphasised that technologies capable of making automated decisions and thereby altering the determinations of public authorities should be approached with great caution, particularly in the field of justice. It further underlined that member states should use such technologies only if there is detailed evidence of their reliability and if a meaningful human review is possible.¹³ Through this, the Parliament reaffirmed that technological innovation must not come at the expense of fundamental rights or human autonomy. The Council of the European Union adopted a similar position in its Conclusions of 8 October 2020, titled “Access to Justice – Seizing the Opportunities of

Digitalisation”. The Council stressed that the use of AI tools must not interfere with judges’ decision-making power or judicial independence. A court decision must always be made by a human being and cannot be delegated to an AI tool.¹⁴

Recent developments at the Council of Europe level further reinforce this human-centred approach. In September 2024, the Council of Europe adopted the Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law,¹⁵ the first binding international instrument addressing the lifecycle of artificial intelligence systems used by public authorities. The Convention establishes horizontal obligations requiring that activities involving AI remain fully consistent with human rights, democracy, and the rule of law, and explicitly calls on member states to ensure that AI systems are not used in a manner that undermines judicial independence, the separation of powers or access to justice. It further requires the adoption of measures ensuring transparency, oversight, accountability and responsibility for adverse impacts on fundamental rights, as well as context and risk-based safeguards, documentation of AI systems capable of significantly affecting human rights, and the availability of effective remedies, including the ability to contest decisions substantially informed by AI use. While not regulating adjudication as such, these requirements situate the use of artificial intelligence in the justice domain within a governance framework that presupposes human authority, responsibility and procedural fairness in the exercise of judicial power, thereby limiting AI to a supportive role rather than an autonomous one.

These soft-law instruments paved the way for the European Union’s next step: the adoption of the AI Act,¹⁶ the world’s first binding legal framework on AI, which anchors Europe’s human-centric approach and emphasises that AI should support rather than replace human judgement.

By classifying as high-risk those AI systems intended to be used by a judicial authority or on their behalf to assist a judicial authority in researching and interpreting facts and the law and in applying the law to a concrete set of facts,¹⁷ the EU recognises both the potential and the sensitivity of such tools. Given that high-risk AI systems must comply with the additional safeguards set out in Section 2 of the AI Act, including, inter alia, the requirement of effective human oversight under Article 14. The AI Act envisages only

supportive, not autonomous, uses of AI in adjudication. Human oversight must be designed into every high-risk system to ensure that natural persons can properly understand the system’s capacities and limitations, monitor its functioning, and intervene where necessary, including by overriding or reversing the system’s output. The AI Act further warns against automation bias and explicitly requires that no decision be made solely on the basis of AI-generated output without separate human verification. Taken together, these safeguards affirm that AI may support judicial reasoning, but adjudication itself remains an inherently human prerogative.

This position is explicitly reinforced in Recital 61 of the AI Act, which classifies as high-risk those AI systems intended to be used by or on behalf of a judicial authority to assist in researching and interpreting facts and the law and in applying the law to a concrete set of facts. The Recital states that such systems should be treated as high-risk given their potentially significant impact on democracy, the rule of law, individual freedoms, and the right to an effective remedy and to a fair trial. It makes clear that while AI tools can support judges’ decision-making power or judicial independence, they should not replace it: the final decision-making must remain a human-driven activity.

Taken together, these instruments articulate a coherent European vision of AI in adjudication, one that welcomes technological progress while safeguarding the primacy of human judgement. The European framework views AI as a means to enhance the efficiency, consistency, and accessibility of justice, yet insists that the authority to decide must remain with human judges.

This approach is consistent with the European e-Justice Strategy 2024–2028,¹⁸ adopted in January 2025, which situates artificial intelligence within the broader digital transformation of justice, and identifies AI as a technology capable of supporting the work of courts and justice professionals, while repeatedly emphasising the need to respect judicial independence, the rule of law, and fair trial guarantees. Notably, the Strategy refers to AI only in relation to clearly supportive functions, such as data analysis, transcription, anonymisation of judicial decisions and legal research, without envisaging its use as an autonomous decision-maker. This sets the stage for the following analysis, which examines how AI systems can, in practice, support judicial decision-making without compromising independence or accountability.

3 AI AS JUDICIAL ASSISTANTS: SUPPORTING HUMAN DECISION-MAKING

In the European Commission’s study on the use of innovative technologies in the justice field, it is indicated that, among the good practices currently in place in member states, there are already those that concern areas such as, inter alia, anonymisation of documents (for example, court decisions); speech-to-text and transcription; introduction of chatbots for strengthening the access to justice and public services, and Robot Process Automation for increasing efficiency and minimizing errors in repetitive tasks.¹⁹

In addition, the European Council notes that AI systems in the justice sector may, in the future, be capable of performing increasingly complex tasks, such as analysing, structuring, and preparing information on the subject matter of cases, automatically transcribing records of oral hearings, offering machine translation, supporting the analysis and evaluation of legal documents and court/tribunal judgements, estimating the chances of success of a lawsuit, automatically anonymising case law, and providing information via legal chatbots.²⁰ This raises the question: Could AI systems one day replace judges in their core function, i.e., decision-making? Under the current European legal framework, the answer is no. The EU law explicitly classifies AI systems intended to support judges as high-risk, thereby ensuring that AI may only operate in an assistive capacity. In Sourdin’s words, assistant ‘co-bots’ rather than replacement robot judges could play a more important role in the future.²¹ Furthermore, Zeleznikow noted that while robots are unlikely to replace judges, automated tools are being developed to support legal decision- making.²² Even considering that AI has the potential to surpass judges in adjudication, AI systems complementing judicial work, including in the decision-making process, are more feasible in the near future.

AI systems can support judges by making predictions about how a case should be decided, as well as by generating a draft judgment based on those predictions. Accordingly, a human judge retains the discretion to make on the final decision and bears responsibility for that judgment. For example, Harvey, a generative AI platform developed for legal professionals and used primarily in legal practice rather than in courts, offers an illustrative account of how AI systems could, in principle, support legal reasoning without

displacing human decision-making. There are databases that use natural language processing to assist with sourcing relevant material based on search terms. The system would need to go beyond these databases by reducing the returned sources to a manageable, relevant sample, then deploying tools to compare these sources of law with a present case and engaging in analysis to decide the outcome. 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.”²³ However, Harvey’s model retains the principle of human judge decision-making.²⁴

By combining the ability of predictive systems to identify patterns that influence projections, and the ability to generate a decision on top of the predictions, in accordance with a specific case and based on the information input, AI assistants can serve judges in adjudication. A human judge could then use this draft (since many judges, especially in appellate courts, already use drafts prepared by their legal assistants) to draft their own reasons for judgment. This use of AI would allow human oversight of the computer program and enable a human judge to take into account discretionary or social considerations that may be beyond computer program’s capacity or authority.²⁵ Much like legal clerks, AI systems can assist in preparing drafts, leaving the judge to exercise judgement, incorporate social and moral considerations, and ensure the final decision aligns with legal and constitutional principles.

The rationale for adopting such supportive systems in Europe is clear: it reflects the EU’s consistent policy that AI should assist, not replace, human judges. The European approach provides clear normative boundaries for such supportive uses. As established in the preceding section, the principle of human oversight, embedded in multiple EU instruments and codified in the Artificial Intelligence Act, requires that judicial decision-making remain under human control. By classifying as high-risk those AI systems intended to assist judges in interpreting facts and law, the EU recognises their potential value while drawing a firm line against their autonomous use in adjudication.

Critically, the principle of human oversight ensures that judges retain responsibility for the final decision, preventing accountability gaps and reinforcing public trust in the judiciary. This approach also aligns with the constitutional principle of judicial prerogative. For instance, under Article 109 of the Constitution of the Republic of Lithuania, justice is administered exclusively by judges,²⁶ a mandate that underscores why human responsibility cannot be outsourced to machines.

Recent European judicial practice illustrates both the possibilities and the risks associated with judges’ use of generative AI as a supportive tool. In a judgment of 7 June 2024, the Dutch Rechtbank Gelderland openly acknowledged its use of ChatGPT to obtain information relevant for the assessment of damages, including estimates concerning the lifespan of solar panels and electricity prices.²⁷ While the court treated ChatGPT as a source of background information rather than a decision-maker, subsequent scholarly commentary has critically highlighted the dangers of relying on generative AI without sufficient transparency, verifiability, and methodological clarity, particularly given the risk of hallucinations and the absence of identifiable sources.²⁸ This example demonstrates that even limited, supportive uses of AI by judges raise serious concerns about accountability, contestability, and the right to a fair trial, underscoring the importance of the safeguards discussed in Section 2, including effective human oversight and transparency requirements.

To sum up, the use of AI systems to assist judges in preparing judgements is in line with the current European approach to preserve the human oversight in the adjudication process, which would contribute to avoiding accountability gaps and preserving public trust, as well as being consistent with the courts’ fundamental role in administering justice, preserving judicial authority, and maintaining public trust, as exemplified by the constitutional principles and the importance of human responsibility in decision-making.

It should be noted that, given the state-of-the-art level of technology, examples from outside the European Union show that the implementation of AI (understood in a broad sense to include algorithmic and data-driven systems supporting judicial decision-making) already raises serious concerns about access to justice. In the United States, the COMPAS algorithm, used to assess the likelihood of recidivism, has been criticised for racial bias and a lack of transparency,²⁹ undermining the fairness and contestability of judicial decisions. In China, the Smart Courts system goes further by issuing “abnormal judgment warnings” when rulings deviate from prior cases,³⁰ thereby pressuring judges to conform to algorithmic patterns rather than exercising independent judicial reasoning. These experiences demonstrate that even supportive, non-autonomous systems, designed to aid rather than replace judges, may inadvertently compromise fundamental guarantees of access to justice if their operation is opaque, biased, or insufficiently overseen by humans.

4 AI AS AUTONOMOUS ADJUDICATORS: THE PROSPECT OF REPLACING JUDGES

What concerns the use of AI to replace judges in adjudication, according to Volokh, is that if an AI program someday passes a Turing test, and its developers can then teach it to converse, and even present an extended persuasive argument. Moreover, if the software can create persuasive opinions, capable of regularly winning opinion-writing competitions against human judges, we should, in principle, accept it as a judge.³¹ Agreeing with Mizaras, and in light of the European Court of Human Rights interpretation of the Convention as a ‘living instrument’ that evolves alongside societal developments, we cannot entirely rule out the possibility of an autonomous E-Judge in the future. While still a work in progress, such a judge could potentially be tasked with resolving smaller, repetitive cases of a non-complex nature, i.e., cases characterised by clear outcomes and minimal evidentiary or interpretive challenges. Realising this vision, however, would require society to approach AI with an open mind, while upholding rigorous legal and ethical standards.³²

Even though the current state-of-the-art AI, and the discussed European approach in terms of using this technology in courts, suggest that we should still be reserved in talking about autonomous application of AI in adjudication, in R. Susskind’s words, lawyers, judges, and policy- makers should be both humbled and open-minded about as-yetuninvented technologies.³³ A few considerations to bear in mind, supporting the idea that in the not-so-near future, the question may not be whether AI will be capable of replacing judges, but whether societies outside the European model will be willing to accept such a transformation, and what implications this might have for global standards of justice, will be discussed below.

Firstly, when considering the limits of the application of AI in courts, what is often underestimated is the so-called AI fallacy, i.e., the view that systems cannot replicate human lawyers and their judgement because they cannot, inter alia, exercise judgement or be empathetic. However, the problem with this point of view is thinking that the only way to get machines to outperform humans is to mimic human reasoning and the way humans work. However, the present-day AI systems operate do not by copying human beings, they take on the work in ways that are best suited to their unique capabilities, not ours.³⁴ The second wave of artificial intelligence has provided an answer to sceptical arguments, after maintaining for a long time that AI could never replace a human, because it does not have and will never have the human qualities. However, the new approach to reach the aimed

result in the most feasible way, rather than to mimic human-beings, opened new possibilities in terms of the ways AI could be used. It is not to mention that there will be a third, as well as a fourth, wave of artificial intelligence, promising even more sophisticated results, thus suggesting that AI could excel in judicial decision-making. As Susskind famously remarked, “patients do not want neurosurgeons; they want health.”³⁵

Secondly, as systems are improving and taking-over more and more tasks, which had previously been exclusively attributed to human abilities, also, as they, inter alia, are making more accurate predictions³⁶ and more proficient at playing chess,³⁷ it is likely that technological intrusion we are comfortable with delegating tasks will change to a great extent. In this era of increasingly capable machines, then, it is not outrageous to expect at some stage, whether twenty or one hundred years from now, that systems will outperform judges at their own game, by delivering reasoned judgements with explanations that will look and feel like the finest of human judgements but sourced through AI rather than the judicial ‘wetware’.³⁸ As technology progresses, society’s comfort with delegating tasks to AI will grow, and it is plausible that AI could deliver judgements even surpassing those of human judges.

From a European regulatory perspective, however, the prospect of autonomous AI adjudication remains incompatible with the existing legal framework. As discussed in Section 2, the AI Act classifies AI systems used to assist judicial authorities as high-risk and subjects them to strict human oversight obligations, while Recital 61 explicitly presupposes that adjudication remains a human-driven activity. Together with the Council of Europe’s insistence on judicial independence, accountability and effective remedies, this framework leaves little room for the lawful deployment of fully autonomous adjudicators in Europe under current conditions. Any move towards replacing judges with AI would require not only technological advances but also a fundamental reconfiguration of Europe’s constitutional and human rights-based understanding of adjudication.

To conclude, this evolution, i.e., AI systems analysing facts, applying the law more effectively than humans, and even delivering reasoned judgements that would rival or surpass those written by flesh-and-blood judges, would challenge the foundations of the European model of adjudication, which rests on human oversight as an indispensable precondition for the proper upholding of the principle of access to justice. Decisionmakers and stakeholders would need to reconsider the current boundaries of AI

applications in justice systems and potentially modify the existing legal framework to meet the demands of emerging realities while safeguarding enduring values and fundamental principles. While these claims remain speculative, they highlight that, in the long term, society may face not only the question of AI’s capabilities but also the question of whether we are comfortable delegating adjudication to a non-human system.

5 CONCLUSIONS

The analysis demonstrates that the European approach to AI in adjudication is firmly grounded in a human-centric, rights-based framework. The Council of Europe and the European Union both view AI as a tool to enhance the efficiency, consistency, and accessibility of justice, not as a substitute for judicial reasoning or human judgment. The AI Act and accompanying soft-law instruments establish a clear normative boundary: AI may assist judges in interpreting facts and law, yet the final decision must remain human-driven and subject to meaningful oversight.

For now, to ensure procedural fairness, judicial accountability, and public confidence, the role of AI in adjudication must remain limited to a supportive function, with judges retaining full authority and responsibility for their decisions. Properly implemented, AI systems can improve efficiency and consistency without undermining judicial independence or the constitutional mandate to administer justice.

At the same time, continued technological development may prompt a careful reassessment of how these safeguards operate in practice. The current European framework, developed through sustained collaboration between EU institutions and the Council of Europe, remains appropriate for present capabilities. Yet future advancements could necessitate a nuanced recalibration of the existing approach to ensure that its commitment to human oversight remains both principled and effective. The central issue will not be AI’s technical sophistication, but how far societies are prepared to rely on non-human systems without compromising the foundations of justice. REFERENCES

1. Angwin J and others, ‘Machine Bias’ (ProPublica, 23 May 2016) <https://www.propublica.org/ article/machine-bias-risk-assessments-in-criminal-sentencing> accessed 30 September 2025

2. Barry BM, How Judges Judge: Empirical Insights into Judicial Decision-Making

(Routledge 2023)

3. Bradford A, Digital Empires: The Global Battle to Regulate Technology (OUP 2023)

4. Delponte L, European Artificial Intelligence (AI) Leadership, the Path for an Integrated Vision (European Parliament 2018)

FOOTNOTES

REFERENCES

AUTHORS INFORMATION

RIGHTS AND PERMISSIONS

EDITORS

ABOUT THIS ARTICLE

DETAILS FOR PUBLICATION

AI DISCLOSURE STATEMENT

АНОТАЦІЯ