A.  Introduction

[1]. Imagine being told that a computer has helped decide whether you get bail or how much compensation you receive. Many people would be uneasy. Yet around the world, courts already use artificial intelligence (AI) to summarise files, sort cases, and predict risks.¹AIJA, AI Decision‑Making and the Courts (AIJA 2023) 5–9.

[2]. AI will not replace judges. But it is changing how judges receive information, manage their workload and write their reasons.²R Purshouse and M Gousmett, ‘Relying on AI in Judicial Decision‑Making: Justice or Jeopardy?’ (PublicPolicy.i.e., 3 March 2025) 3–7. Used wisely, AI can reduce delay and improve consistency. Used badly, it can threaten fairness and public trust.

[3]. This essay explains, in simple terms, what courts actually do, how AI already supports real courts, what risks it creates, and how judges can use it safely. It is written for readers whose first language is not English. Sentences are short. The ideas are serious.

B.  What courts really do

[4]. Courts in many countries face heavy caseloads. Civil and criminal courts deal with more cases, more documents and more complex law each year.³AIJA (n 1) 10–13. In England and Wales, the County Court receives over 1.5 million civil claims a year.⁴A Uzelac, ‘Comparative Analysis of Judicial Statistics Reform’ (2022) 13 Jurnal Hukum dan Peradilan 495, 503–504, 507–509.

[5]. In the United States, federal district courts see more than 300,000 civil filings annually.⁵AIJA (n 1) 11–12. Most of these cases will never reach a full trial. Many end by default judgment, consent order, or settlement at an early stage.

[6]. Singapore and Malaysia show the same pattern. There are rising numbers of small claims and community disputes. At the same time, commercial and regulatory cases are more data‑heavy and technical.⁶Singapore Judiciary, Annual Report 2023 (2024) 16–18; Pejabat Ketua Pendaftar Mahkamah Persekutuan, Laporan Tahunan Badan Kehakiman Malaysia 2023 (Federal Court of Malaysia 2024) 22–29, 54–61.

[7]. So judges live with two pressures. First, very high volume in simple, routine matters. Second, very high complexity in a smaller number of big cases, with thousands of pages and long expert reports.⁷M Fabri and F Contini, ‘Artificial Intelligence, Judicial Decision‑Making and the Rule of Law’ in R Winkler and others (eds), Justice and Technology in Europe (SSM Italia 2025) 22–25. This is the world into which AI now enters.

C.   How AI already supports courts

[8]. AI can assist at almost every stage of a case. At filing, systems can classify new cases, flag urgent matters and send them to the right list or judge.⁸AIJA (n 1) 18–21. This can help courts control their backlogs.

[9]. During case‑management, AI can suggest timetables and directions. It can highlight cases suitable for mediation or simplified procedures. This can reduce delay and encourage settlement.

[10]. For evidence and facts, AI can search and cluster documents, review e‑mails and chats, and highlight contradictions or gaps. For law, it can find relevant cases and statutes, summarise judgments and suggest a structure for reasons.⁹Fabri and Contini (n 7) 24–30.

[11]. AI can also assist in sentencing and risk. Some tools predict the risk that a person will offend again. Others suggest guideline ranges or ‘bands” for awarding damages.¹⁰J Dressel and H Farid, ‘The Accuracy, Fairness, and Limits of Predicting Recidivism’ (2018) 4 Science Advances eaao5580. In administration, dashboards help leaders see where delay is growing and where more judges are needed.¹¹AIJA (n 1) 35–43.

D.   Real examples: Singapore, Estonia, Brazil, Argentina, United States

[12]. Singapore is often seen as a careful leader in digital justice. Its Small Claims Tribunals now use a generative AI tool that reads parties’ documents and produces short case summaries.¹²Singapore Judiciary, ‘Media Release: New Generative AI‑Powered Case Summarisation Tool to Help Small Claims Tribunals Users’ (9 September 2025). At first, tribunal magistrates used the tool. It is now being rolled out to self‑represented persons.¹³‘Judiciary to launch generative AI tool to summarise cases for Small Claims Tribunals’ Channel NewsAsia (Singapore, 9 September 2025).

[13]. The same tribunals also use an AI‑powered translation service that converts English information into Chinese, Malay and Tamil.¹⁴S Lum, ‘Small Claims Tribunals Roll Out AI‑Powered Translation Service for Users’ The Straits Times (Singapore, 15 April 2025). This reduces language barriers for ordinary people. In all these uses, AI does not decide the case. The human tribunal remains in charge.

[14]. Estonia and some Nordic countries have tested AI for small, simple disputes, such as traffic offences and low‑value claims. In Estonia, an AI system suggests an outcome based on past cases and fixed rules.¹⁵M De Stefano, ‘AI and Digital Justice in EU Labour Law: A Comparative Study on Algorithmic Management’ (2026) 18 European Labour Law Journal 45, 60–64. A human official reviews that suggestion and can change it. Parties keep a right of appeal.

[15]. Brazil and Argentina show AI at scale. In Brazil, systems such as VICTOR help the Supreme Court classify new appeals and spot those that raise constitutional questions.¹⁶I Ferrari and D Becker, ‘Artificial Intelligence and the Supreme Court of Brazil: Beauty or a Beast?’ (Supreme Courts and the World’s Commercial Courts, 22 June 2020). Other tools draft decisions in repetitive cases, freeing judges to focus on harder matters.¹⁷‘Brazil’ (Technology, Justice and the Rule of Law, University of Oxford, 2 March 2023).

[16]. In Argentina, the Prometea system manages deadlines, scans files, predicts likely outcomes and drafts orders. In some pilots, it reduced the time needed for certain tasks by up to 70 or even 80 per cent.¹⁸Office of the Public Prosecutor of Buenos Aires City, ‘Innovation and Artificial Intelligence’ (2019) 7–9. It has helped clear backlogs in overburdened offices.

[17]. The United States offers more cautious lessons. Risk‑assessment tools such as COMPAS have been used in bail and sentencing decisions.¹⁹Dressel and Farid (n 11). One study found that COMPAS was no more accurate than predictions by volunteers with no legal training and produced more false ‘high‑risk’ scores for Black defendants than for White defendants.²⁰ibid 4–6.

[18]. Research on public attitudes is also mixed. Some people think judges who rely on AI seem less legitimate. Others, especially members of minority communities that have faced discrimination, may trust AI‑supported decisions more than decisions based only on human judgment.²¹A Kramer and others, ‘Public Perceptions of Judges’ Use of AI Tools in Courtroom Decision‑Making’ (2025) Journal of Empirical Legal Studies (advance publication).

E.   Risks for fairness and trust

[19]. When courts adopt AI, they must protect certain basic values. These include open justice, fair procedure, equality of arms, judicial independence and impartiality, and clear, honest reasons.²²UNESCO, Guidelines for the Use of AI Systems in Courts and Tribunals (UNESCO 2025) 4–6.

[20]. Bias is a major danger. Past data often reflects racial, gender or class prejudice. If we train AI on that data, we risk freezing these patterns and making them worse.²³Dressel and Farid (n 11) 4–6; L Citron and FD Pasquale, ‘The Scored Society: Due Process for Automated Predictions’ (2014) 89 Washington Law Review 1, 23–30.

[21]. Opacity is another problem. Many models are ‘black boxes’. Judges and lawyers cannot see why the system produced a particular answer.²⁴Fabri and Contini (n 7) 31–34. Under pressure of time, judges may lean too heavily on AI outputs. Over the years, this may weaken their own research and writing skills.²⁵AIJA (n 1) 42–47.

[22]. There are also security and confidentiality risks. Court files contain sensitive personal and commercial information. If courts send this material to cloud‑based tools, there is a risk of leaks or misuse.²⁶UNESCO (n 19) 10–12. Finally, if people feel that “the computer” is judging them, public trust in courts may fall.²⁷Kramer and others (n 18).

[23]. International bodies now offer guidance. UNESCO’s Guidelines for the Use of AI Systems in Courts and Tribunals set out fifteen principles, including human rights, safety, information security, accuracy, reliability, explainability, auditability, and human oversight.²⁸UNESCO, ‘Guidelines for the Use of AI Systems in Courts and Tribunals’ (UNESCO 2025) 6–9; ‘UNESCO launches AI guidelines for courts and tribunals’ (Digital Watch Observatory, 3 December 2025). The Council of Europe’s ethical charter on AI in judicial systems states clearly that AI may help judges but may not replace them.²⁹Council of Europe, European Commission for the Efficiency of Justice, ‘European Ethical Charter on the Use of Artificial Intelligence in Judicial Systems’ (2018).

[24]. In England and Wales, judicial guidance also reminds judges that they remain personally responsible for everything they issue. It warns them not to put confidential material into public AI tools and to check all AI‑produced research carefully.³⁰UK Judiciary, ‘Artificial Intelligence (AI) Guidance for Judicial Office Holders’ (October 2025) paras 12–22.

F.   How judges should work with AI

[25]. Sometimes, I smile when I hear judges (particularly the clever ones) ask counsel, e.g., “How do your arguments [or facts] engage with the relevant section?” There are also other tell-tale signs. But this is a good thing.

[26]. Courts need a simple method for handling AI in cases. A three‑step approach is useful. First, decide what the AI output is. Sometimes it is evidence, such as a risk score or forensic match. Sometimes it is research help, like a suggested case. Sometimes it is an administrative tool, used for allocating cases.³¹Dressel and Farid (n 11) 6–8; UK Judiciary (n 24) paras 16–20.

[27]. If the output is evidence, it should be subject to normal rules of admissibility, disclosure and cross‑examination.³²Dressel and Farid (n 11) 6–8. If it is research help, judges should treat it like a memo from a law clerk: useful, but always to be checked. If it is an administrative tool, its rules should be public and open to review.³³UNESCO (n 19) 8–10; Council of Europe (n 23) principles 1–3.

[28]. Second, think about disclosure. If AI has materially influenced a decision – for example, by shaping the main issues, or suggesting a sentencing range – the parties should, in principle, be told.³⁴Purshouse and Gousmett (n 2) 14–18; UNESCO (n 19) 8–10. They should have a chance to question and challenge that use.

[29]. Third, keep records. Courts should log which tools were used, on which documents, when, and with what outputs. Contracts with vendors should allow courts to test systems and audit their performance.³⁵AIJA (n 1) 50–53. Judicial training should cover basic statistics and concepts like false positives and bias, so judges can interrogate the tools they use.³⁶AIJA (n 1) 53–55.

G.   What reform should look like

[30]. For AI to support, not weaken, judicial decision‑making, many actors must work together.³⁷Stimson Center, ‘AI in Global Majority Judicial Systems’ (Policy Brief, 2026) 20–24. Courts and judicial councils should set up technology committees, run pilot projects with clear goals, and design systems together with judges and staff.³⁸AIJA (n 1) 53–60.

[31]. Legislatures should define ‘high‑risk’ AI uses, such as sentencing or detention, and impose strict safeguards. They should set strict rules on court data and privacy, require vendors to disclose key information, and clarify who is liable when AI‑assisted processes cause harm.³⁹UNESCO (n 19) 10–14; Council of Europe (n 23).

[32]. Professional bodies and court users also have a role. They should set competence standards for lawyers who use AI. They should issue guidance on how to cite AI‑generated analysis and create procedures to challenge AI‑affected decisions, including access to technical experts.⁴⁰UK Judiciary (n 24) paras 12–22; Citron and Pasquale (n 20) 28–30.

[33]. The stories from Singapore, Estonia, Brazil and Argentina show that, in well‑chosen areas, AI can cut delay and help users without undermining judicial authority.⁴¹Singapore Judiciary (n 12); De Stefano (n 14) 60–64; Stimson Center (n 29) 13–16; Ferrari and Becker (n 15); Innovation and Artificial Intelligence (n 16). The United States’ struggles with opaque and biased risk tools show the opposite: if AI is unchecked, it can damage both fairness and trust.⁴²Dressel and Farid (n 11); Kramer and others (n 18).

[34]. AI is a tool. It can straighten what is crooked, or make crooked lines harder to see.

[35].  The choice lies with judges, lawmakers, lawyers and the public.

[36].  If they insist on transparency, fairness and human judgment at the centre, AI will strengthen, not weaken, judicial decision‑making.

∞§∞

We thank Getty Images of Unsplash for the image.

The author thanks Miss KN Geetha, Miss Lydia Jaynthi, Miss TP Vaani and Miss JN Lheela.

@Copyright reserved.

All content on this site, including but not limited to text, compilation, graphics, documents, and layouts, is the intellectual property of GK Ganesan Kasinathan and is protected by local and international copyright laws. Any use shall be invalid unless written permission is obtained by writing to gk@gkganesan.com