Click here to download the full paper (PDF)
Authored By: Sonika Choudhury, Advocate,
Click here for Copyright Policy.
ABSTRACT:
“The advent of artificial intelligence in the legal domain has ignited a profound jurisprudential debate: can algorithmic systems supplant the nuanced exercise of judicial discretion? This paper examines this question through the specific lens of the Indian judiciary – a constitutional democracy whose courts blend common law traditions with transformative social justice imperatives. It surveys global deployments of an AI in court systems, evaluates Indian experiments in legal technology, and critically analyses the constitutional, ethical and structural barriers to algorithmic adjudication. The paper argues that while AI tools offer significant promise in procedural administration, case management, and legal research, they cannot – and constitutionally must not- replace the deliberative exercise of judicial discretion. A framework of “argumentative AI” is proposed: embedding technology as a tool that enhances rather than supplants the human judge. The paper concludes with reform recommendations grounded in Indian Constitutional values”.
Keywords: Artificial Intelligence, Judicial Discretion, Indian Judiciary, Legal technology, Constitutional Law, Algorithmic Justice, Rule of Law, Due Process.
INTRODUCTION:
The courtroom has long been regarded as a sanctum of human reason — a space where trained legal minds weigh evidence, interpret statute, and apply precedent to the infinite variability of human circumstances. Yet the twenty-first century has brought an unprecedented challenge to this image: the rise of artificial intelligence and machine learning systems capable of processing vast legal datasets, identifying patterns, and generating decisions at speeds no human jurist can match. India’s judiciary finds itself at a particularly critical inflection point. With over five crore (50 million) cases pending across its courts as of 2024, chronic judicial vacancies, and mounting public frustration with delays, the temptation to deploy algorithmic solutions is considerable. [1] This paper poses a foundational question: can — or should — algorithms replace judicial discretion in India? The inquiry is not merely technical but deeply constitutional, philosophical, and sociological. Judicial discretion is not an imperfection to be optimized away; it is the mechanism through which law adapts to justice, through which the lived reality of a litigant encounters the abstract formulations of statute. The paper proceeds in eight parts. Part II examines the nature and constitutional status of judicial discretion. Part III surveys global AI deployments in courts. Part IV analyses India-specific initiatives and experiments. Part V addresses constitutional barriers. Part VI engages with ethics and bias. Part VII proposes a framework of augmentative AI. Part VIII offers reform recommendations, and Part IX concludes.
THE ARCHITECTURE OF JUDICIAL DISCRETION:
- WHAT IS JUDICIAL DISCRETION?:
Judicial discretion refers to the latitude vested in a judge to make choices among legally permissible alternatives based on reason, experience, and the particularised facts of a case. It is not arbitrary power; discretion is structured by statutory framework, constitutional mandate, precedent, and procedural rules. Yet within those constraints, the judge exercises genuine judgment — an evaluative, contextual, and often morally inflected act. The philosopher Ronald Dworkin distinguished between ‘weak’ discretion (where the applicable standard does not mechanically determine a unique answer) and ‘strong’ discretion (where the decision-maker is not bound by any standard). Judicial discretion in the Indian context is primarily weak in Dworkin’s sense: judges operate within a normative architecture but exercise judgment in applying it.[2]
- CONSTITUTIONAL FOUNDATIONS IN INDIA:
The Constitution of India does not use the term “judicial discretion” explicitly, but the concept pervades its structure. Article 142 grants the Supreme Court the extraordinary power to pass any order necessary for “complete justice” — a provision that epitomises discretion elevated to constitutional principle. [3] Similarly, the inherent powers of courts under the Code of Civil Procedure (Section 151) and the Code of Criminal Procedure (Section 482) are expressions of discretionary authority. The Supreme Court has repeatedly affirmed that discretion is essential to the judicial function. In State of Punjab vs. Gurmit Singh (1996), the Court observed that the absence of rigid rules in certain domains reflects a deliberate legislative choice to repose trust in the judiciary’s capacity for contextual judgment. [4]
Categories of Discretion:
Judicial discretion operates across multiple dimensions of adjudication:
- Interpretive discretion: Choosing among competing readings of ambiguous
statutory or constitutional text.
- Remedial discretion: Fashioning appropriate relief, particularly in public interest litigation and constitutional matters.
- Evidential discretion: Determining the weight and admissibility of evidence,
including credibility assessments of witnesses.
- Sentencing discretion: Calibrating punishment to the individual offender, victim, and social context.
- Bail and custody discretion: Assessing flight risk, threat to society, and personal
liberty in pre-trial detention decisions.
Each of these categories presents distinct challenges for algorithmic substitution.
- GLOBAL LANDSCAPE: AI IN COURTS:
Predictive Policing and Recidivism Tools: The United States Experience:
The most extensively studied AI deployment in the legal domain is the COMPAS
(Correctional Offender Management Profiling for Alternative Sanctions) system used in several US states for bail and parole decisions. A 2016 ProPublica investigation found COMPAS incorrectly flagged Black defendants as future criminals at nearly twice the rate of white defendants — a finding that triggered an ongoing debate about algorithmic fairness that has yet to be resolved. [5] The Wisconsin Supreme Court in State vs. Loomis (2016), held that while COMPAS could be used as one factor among others, due process precluded it from being the determinative factor, and defendants must have the opportunity to challenge the methodology.
European Approaches: Estonia, France and the Netherlands:
Estonia has pioneered “robot judges” for small claims disputes under €7,000, where an AI system makes binding decisions subject to human appeal. France, by contrast, enacted a law in 2019 explicitly prohibiting the publication of statistical data about ‘individual judges’ decisions — a direct response to concerns that predictive analytics could undermine judicial independence.[6]
The Dutch SyRI (System Risk Indication) case is instructive: In 2020, a Hague court struck down SyRI — a government algorithm used to detect welfare fraud — on grounds that it violated the right to private life under the European Convention on Human Rights. The court emphasised the opacity of the system and the impossibility of meaningful challenge.
China’s Smart Court System:
China’s deployment of AI in courts is the most expansive globally. Platforms like 206 and Faxin use AI to suggest verdicts, flag inconsistencies, and generate draft
judgments. While efficiency gains are documented, the system operates in a political and institutional context that limits meaningful evaluation of its impact on due process and judicial independence — values that animate the Indian constitutional framework in very different ways.
- AI IN THE INDIAN JUDICIAL CONTEXT:
The SUPACE and SUVAS Initiatives:
India’s Supreme Court launched SUPACE (Supreme Court Portal for Assistance in Courts Efficiency) in 2021 as an AI-powered tool to assist judges by processing case files, identifying relevant precedents, and surfacing key legal issues. Critically, SUPACE is designed as a decision-support tool: it presents information to judges but does not generate recommendations or predicted outcomes.
SUVAS (Supreme Court Vidhik Anuvaad Software) uses neural machine translation to translate Supreme Court judgments into regional Indian languages, dramatically improving access to justice for non-English speakers. This application represents an area where AI delivers unambiguous benefit without raising concerns about decisional substitution.
The National Judicial Data Grid:
The National Judicial Data Grid (NJDG) aggregates data from district and high courts across India, tracking pendency, disposal rates, and case-type distributions in near real- time. The grid has already been used to identify bottlenecks, prioritise old cases, and measure court performance — administrative functions where data-driven approaches are largely uncontroversial.
State-Level Experiments:
Several High Courts have experimented with AI-assisted case listing and scheduling. The Bombay, Delhi, and Telangana High Courts have piloted systems using historical data to predict hearing durations and optimise schedules. These experiments remain largely undocumented and unregulated — a governance lacuna this paper addresses in Part VIII.
The Bail Crisis: A Case Study-
India’s undertrial prisoner population is among the largest in the world, with undertrials constituting approximately 76% of the total prison population as of 2023. [7]
The argument for algorithmic bail assessment is superficially appealing: if decisions are inconsistent or arbitrary, standardization might deliver fairer outcomes. However, an algorithm trained on historical bail decisions will reproduce embedded biases —including those based on caste, poverty, and locality — with greater speed and scale, and with a veneer of objectivity that makes challenge more difficult.[8]
- CONSTITUTIONAL AND JURISPRUDENTIAL BARRIERS:
Articles 14 and 21: Equality and Due Process
Article 14 of the Constitution guarantees equality before the law and equal protection of the laws. Article 21 protects the right to life and personal liberty, which the Supreme Court has interpreted to include procedural due process through decisions from Maneka Gandhi v. Union of India (1978) onwards. 1819
If an algorithm incorporates variables correlated with group membership —
socioeconomic status, neighbourhood, education level — that are themselves products of historical discrimination, the algorithm perpetuates and entrenches inequality in violation of Article 14. If a litigant cannot access, interrogate, or challenge the algorithmic model that determined their fate, the due process guarantee of Article 21 is hollowed out.
The Doctrine of Non-Delegation and Judicial Power:
The Indian Constitution vests judicial power in the courts. Articles 124 and 214 establish the Supreme Court and High Courts respectively; Article 50 mandates separation of the judiciary from the executive. Delegating the core function of adjudication to an algorithmic system would constitute an impermissible delegation of judicial power.
The Right to a Reasoned Decision:
Indian courts have held that the duty to give reasons is an essential element of the rule of law. In Union of India v. Mohan Lal Capoor (1973), the Supreme Court held that reasons are the link between the mind of the decision-maker and the decision made.
“Reasons are the links between the mind of the decision-taker to the decision. They demonstrate that the mind was applied to the subject-matter for decision and show how discretion has been exercised.” — Union of India v. Mohan Lal Capoor (1973) 2 SCC 836[9]
A decision generated by a neural network — whose internal reasoning process is
opaque even to its designers — cannot satisfy this requirement.
The Right to Counsel and Adversarial Process:
The adversarial system — wherein parties present competing accounts of fact and law before an impartial decision-maker — is a constitutional feature of Indian courts, not a mere procedural convenience. AI-generated decisions would short-circuit this process. The advocate’s ability to adapt arguments to a judge’s questions and respond dynamically to emerging issues is irreplaceable by any current AI system.
- ETHICAL DIMENSIONS: BIAS, TRANSPARENCY AND ACCOUNTABILITY:
The Problem of Algorithmic Bias in the Indian Context-
Machine learning systems learn from historical data. In India, historical legal data is a record not only of legal precedent but of entrenched social inequities: discriminatory policing practices, caste-based biases in bail decisions, and economic disparities in access to legal representation. An algorithm trained on this data will not produce neutral outcomes; it will amplify existing patterns. [10]
Explainability and the Black Box Problem-
The most powerful AI systems — large language models, deep neural networks — are fundamentally opaque. This opacity is irreconcilable with the requirement of reasoned judicial decision-making.
Explainable AI (XAI) research seeks to address this through techniques such as LIME and SHAP. However, even these tools produce post-hoc rationalisations of algorithmic outputs rather than genuine insight into how the system ‘reasons’; The gap between the technical explanation and a legally adequate reason is substantial.
Accountability Gaps:
When a human judge makes an erroneous decision, accountability mechanisms exist: appeals, judicial review, professional discipline, and the possibility of public criticism.
When an algorithm errs, accountability becomes diffuse. India’s current legal framework — including the Digital Personal Data Protection Act 2023 — does not adequately address liability for algorithmic error in judicial contexts. [11]
Democratic Legitimacy:
Judges derive their legitimacy from the processes of appointment, oath, and the rules of judicial conduct that bind them. An algorithm derives its outputs from data and mathematical operations. It has no constitutional oath, no accountability to the public, and no capacity for moral reasoning. To vest decisional authority in an algorithm is to transfer a dimension of democratic sovereignty to a technical artefact — a transfer for which no constitutional warrant exists.
VII. The Case for Augmentative AI: A Middle Path:
Defining the Augmentative Model:
The binary framing — AI replaces judges versus AI is irrelevant — obscures the most important question: how can technology enhance the quality and efficiency of judicial decision-making without displacing the human judgment that justice requires? This paper proposes the framework of Augmentative AI: the deliberate and bounded use of AI tools to support, not supplant, judicial discretion.
Domains of Legitimate AI Deployment:
The following domains are appropriate for AI deployment within the augmentative framework:
- Case Management and Scheduling: AI can optimise court scheduling, reduce
hearing gaps, and predict case duration with significantly greater accuracy than
manual processes.
- Legal Research Assistance: AI tools can surface relevant precedents, statutory
provisions, and academic commentary — compressing research time without
replacing the lawyer’s evaluative function.
- Document Analysis: AI can identify key facts, dates, and legal issues in large
document sets — invaluable in commercial arbitration, tax tribunals, and insolvency
proceedings.
- Translation and Access: Building on SUVAS, AI translation can extend judicial
access across India’s 22 scheduled languages, closing a profound access-to-
justice gap.
- Administrative Tribunals: In highly rule-bound domains (tax assessment, pension disputes, routine regulatory decisions), AI-assisted decision-making under human supervision may be appropriate — provided robust appeal mechanisms exist.
The Human-in-the-Loop Imperative:
In every domain where AI is deployed in the judicial context, a human decision-maker must retain ultimate authority. This is not merely a safeguard against algorithmic error; it is a constitutional requirement. The doctrine of the ‘human-in-the-loop’ — embedded in emerging international frameworks including the EU AI Act (2024) — must be operationalised in Indian judicial technology policy. [12] The Augmentative AI framework positions AI as the equivalent of a highly competent law clerk: capable of prodigious research and organisation, but ultimately subordinate to the judge’s authority, judgment, and constitutional responsibility.
VIII. REFORM RECOMMENDATIONS:
This paper proposes the following reform agenda for the integration of AI in Indian courts:
Legislative Framework:
- Enact a Judicial AI Governance Act establishing standards for any AI tool
deployed in judicial or quasi-judicial proceedings, including mandatory explainability
requirements, bias audits, and liability provisions.
- Amend the Code of Criminal Procedure (or its successor legislation) to explicitly prohibit the use of AI-generated risk assessments as determinative factors in bail, sentencing, or parole decisions.
- Create a statutory right to contest any AI-assisted administrative decision that
adversely affects a citizen, with a corresponding duty on the deploying institution to disclose the algorithmic methodology.
Institutional Design:
- Establish an independent Judicial Technology Oversight Committee under the
Supreme Court, with representation from judges, lawyers, technologists, civil
society, and marginalised communities.
- Mandate public disclosure of any AI system used in courts, including its training data, methodology, accuracy rates, and error analysis disaggregated by
demographic groups.
- Require judicial impact assessments before any AI tool is deployed in a court
setting, modelled on the environmental impact assessment framework.
Judicial Training and Capacity:
- Integrate AI literacy into the curriculum of the National Judicial Academy and
State Judicial Academies, ensuring judges can critically evaluate AI-generated
outputs rather than deferring to them uncritically.
- Develop a mandatory certification programme for lawyers appearing before AI- assisted tribunals, covering the technical basis, limitations, and contestation
strategies for algorithmic evidence.
Access and Inclusion:
- Prioritise AI investments that expand access to justice — translation, legal aid
assistance, e-filing for remote litigants — over efficiency tools that primarily benefit institutional users.
- Conduct mandatory bias audits using disaggregated data including caste,
religion, gender, and economic status before any AI tool is deployed in criminal
justice settings.
- CONCLUSION:
The question posed by this paper — can algorithms replace judicial discretion in India — must be answered with an emphatic constitutional negative. Judicial discretion is not an inefficiency to be engineered away; it is the mechanism through which law becomes justice, through which abstract norms encounter the irreducible complexity of human lives. This is not a counsel of technological pessimism. The Indian judiciary faces genuine and urgent challenges — a crushing backlog, gross inequalities in access, linguistic barriers, and institutional strain — for which technology offers real solutions. The SUPACE initiative, the NJDG, and the SUVAS translation system are steps in a direction that can be both ambitious and constitutionally sound.
The critical distinction is between AI as a tool that empowers judges to exercise their discretion more effectively, and AI as a substitute that makes decisions in place of judges. The former is consistent with constitutional values; the latter is not. India’s constitutional framework — with its emphases on dignity, equality, due process, and the accountability of power — provides exactly the normative architecture needed to navigate this challenge. Law’s ultimate object is justice — a quality that requires not computation but wisdom. As Justice Krishna Iyer observed, no machine can substitute for the wisdom of a judge who has heard the living voice of the litigant. No algorithm has yet replicated wisdom. Until it does — if it ever does — judicial discretion remains not merely valuable but irreplaceable. [13]
Cite this article as:
Sonika Choudhury, “AI And Judicial Decision-Making: Can AI Replace Judicial Discretion In India?”, Vol.6 & Issue 4, Law Audience Journal (e-ISSN: 2581-6705), Pages 207 to 218 (1st June 2026), available at https://www.lawaudience.com/ai-and-judicial-decision-making-can-ai-replace-judicial-discretion-in-india/.
References:
Cases
Maneka Gandhi v Union of India (1978) 1 SCC 248
State of Punjab v Gurmit Singh (1996) 2 SCC 384
Union of India v Mohan Lal Capoor (1973) 2 SCC 836
AK Gopalan v State of Madras AIR 1950 SC 27
Puttaswamy v Union of India (2017) 10 SCC 1
Arnesh Kumar v State of Bihar (2014) 8 SCC 273
State v Loomis, 881 NW 2d 749 (Wis 2016)
Rechtbank Den Haag, SyRI Case, ECLI:NL:RBDHA:2020:1878 (5 February 2020)
Books
Dworkin R, Taking Rights Seriously (Harvard University Press 1977)
Hildebrandt M, Law for Computer Scientists and Other Folk (Oxford University Press 2020)
Pasquale F, The Black Box Society (Harvard University Press 2015)
Susskind R, Online Courts and the Future of Justice (Oxford University Press 2019)
Baxi U (ed), Law and Poverty: Critical Essays (NM Tripathi 1988)
[1] National Judicial Data Grid (NJDG), ‘Pendency of Cases in Indian Courts’ (2024) <https://njdg.ecourts.gov.in> accessed April 2026.
[2] Ronald Dworkin, Taking Rights Seriously (Harvard University Press 1977) 31–39. Dworkins distinction between ‘weak’ and ‘strong’; discretion has been foundational to debates about the limits of rule-based reasoning.
[3] Constitution of India, Art 142 “The Supreme Court in the exercise of its jurisdiction may pass such decree or make such order as is necessary for doing complete justice in any cause or matter pending before it…;
[4] State of Punjab v Gurmit Singh (1996) 2 SCC 384 [15] (SC, Verma CJ): The legislature has deliberately chosen not to circumscribe judicial power in this domain, reposing confidence in the judge’s capacity for contextual evaluation.”
[5] Julia Angwin and others, ‘Machine Bias’; Pro Publica (23 May 2016) https://www.propublica.org/article/machine-bias- risk-assessments-in-criminal-sentencing accessed April 2026.
[6] Code de’l organisation judiciaire (France), art L 111-13 (as amended by Loi n° 2019-222 du 23 mars 2019):
prohibiting the publication of statistical data identifying individual judges and their decisions.
[7] National Crime Records Bureau, Prison Statistics India 2023 (Ministry of Home Affairs 2024) Table 4.1: undertrials constituted 75.8% of the total prison population as of 31 December 2023.
[8] Priya Gupta, ‘Algorithmic Bail and the Constitution: Lessons from the United States for India” 39; (2023) 35 National Law School of India Review 88, 102–108.
[9] For an analysis of neural network opacity and its legal implications, see Finale Doshi-Velez and Been Kim,
‘Towards a Rigorous Science of Interpretable Machine Learning’; (2017) arXiv:1702.08608.
[10] Vrinda Grover, ‘Surveillance, Data and Discrimination in India’; (2022) 57(12) Economic and Political Weekly; Usha Ramanathan: A Unique Identity Bill 39; (2010) 45(35) Economic and Political Weekly 9.
[11] Digital Personal Data Protection Act 2023 (India). The Act creates rights of data principals against automated
processing but does not specifically address algorithmic decision-making in judicial or quasi-judicial proceedings.
[12] Regulation (EU) 2024/1689 of the European Parliament and of the Council on Artificial Intelligence (EU AI Act), Art 14 (human oversight); recital 48 (high-risk AI systems in administration of justice).
[13] Justice VR Krishna Iyer, ‘Social Justice: The Big Challenges’ in Upendra Baxi (ed), Law and Poverty: Critical Essays (NM Tripathi 1988);Justice is not a mechanical art. It is a human endeavour, and no machine can substitute for the wisdom of a judge who has heard the living voice of the litigant.
