Automated Decision-Making and Explainability Law: The Right to Reasoned Decisions in Indian Algorithmic Jurisprudence

I. I. The Black-Box Enigma in Consequential Governance and Commerce

The defining paradox of modern artificial intelligence lies in the tension between predictive accuracy and interpretability. Deep neural networks, multi-layered transformer architectures, and gradient-boosted decision ensembles derive their extraordinary predictive power from complex, non-linear mathematical transformations across billions of parameters. While these systems excel at optimizing complex statistical objectives, they operate as functional 'black boxes'—even the computer scientists who authored the underlying code cannot trace the precise causal chain that led an algorithm to generate a specific output for a specific individual.

In high-stakes environments, this opacity ceases to be an academic curiosity and becomes a direct challenge to the Rule of Law. When public authorities utilize automated decision-making (ADM) systems to determine eligibility for welfare pensions, flag tax returns for criminal audits, or deny bail recommendations, the absence of an intelligible explanation violates the most foundational precepts of administrative justice.

In Indian constitutional jurisprudence, administrative and judicial power is bounded by the non-negotiable obligation to provide reasoned decisions. This treatise examines how Indian courts are applying constitutional principles of natural justice to automated decision-making systems, exploring the emergence of an enforceable statutory and constitutional 'Right to Explanation'.

II. II. The Constitutional Doctrine of Reasoned Orders under Article 14

In Indian administrative law, the requirement to give reasons is an indispensable component of natural justice (*audi alteram partem*). In a celebrated line of Supreme Court authorities—including *Siemens Engineering & Manufacturing Co. of India v. Union of India* (1976 2 SCC 981), *S.N. Mukherjee v. Union of India* (1990 4 SCC 594), and *Kranti Associates Pvt. Ltd. v. Masood Ahmed Khan* (2010 9 SCC 496)—the apex court held that the recording of reasons is the heartbeat of every administrative and judicial decision.

Reasons introduce clarity, prevent arbitrary discretion, assure the affected individual that their case was conscientiously considered, and provide the indispensable factual basis for judicial review by superior courts under Article 226 or Article 32 of the Constitution. An unreasoned order is inherently arbitrary, and under the *E.P. Royappa* doctrine, arbitrariness is a direct violation of the equality guarantee under Article 14.

When an administrative authority substitutes its discretionary judgment with an automated algorithmic output—such as summarily cancelling a ration card or disqualifying a civil service aspirant based solely on a computer score—the failure of the algorithm to articulate specific, individualized reasons renders the administrative action unconstitutional on its face. The state cannot insulate itself from constitutional review behind the algorithmic complexity of a neural network.

III. III. Statutory Dimensions of Explainability: DPDP Act 2023 and Sectoral Regulations

Beyond constitutional writ jurisprudence, statutory mandates increasingly demand algorithmic transparency. Under Section 11 of the Digital Personal Data Protection Act 2023, Data Principals possess the statutory right to access information about their personal data, including a summary of personal data being processed, the processing activities undertaken, and the identities of other Data Fiduciaries with whom data has been shared.

While the DPDP Act does not codify a word-for-word copy of GDPR Article 22's 'Right not to be subject to automated decision-making', the Section 11 right to information, read alongside Section 8(1) obligations to process data for lawful purposes, establishes an operational necessity for explainability. Where automated profiling directly impacts personal rights, a complete refusal to disclose the processing logic violates the transparency duties owed by Data Fiduciaries.

Sectoral regulators have advanced even further. The Reserve Bank of India, through its Digital Lending Directions, mandates that credit decisioning algorithms must be transparent, auditable, and documented. Regulated lenders cannot deny credit to consumers based on uninterpretable algorithmic scores without providing a clear, understandable statement of the primary factors that contributed to the adverse credit determination.

IV. IV. Technical Mechanisms for Explainability: Local vs Global Interpretability

To satisfy legal and regulatory standards of explainability, machine learning engineers have developed two primary branches of algorithmic interpretability: Global Interpretability and Local Interpretability. Global interpretability seeks to explain the overall behavior of the entire model across the entire feature space, utilizing techniques such as feature importance rankings, partial dependence plots, and surrogate decision trees.

However, in legal disputes, an aggrieved individual is not concerned with the model's global statistical tendencies; they demand to know why *their specific application* was rejected. This requires Local Post-Hoc Interpretability. Methodologies such as Local Interpretable Model-agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP)—grounded in cooperative game theory—calculate the precise marginal contribution of each individual feature to a specific decision.

Furthermore, the most legally valuable form of explanation is Counterfactual Recourse ('Actionable Recourse'). A legally robust counterfactual explanation informs the individual: 'Your loan application was rejected because your debt-to-income ratio was 45% and credit utilization was 60%; if your debt-to-income ratio were reduced to 35% and utilization to 40%, the algorithm would have approved the application.' Providing actionable recourse transforms opaque automated scoring into fair, actionable administrative guidance.

V. V. Judicial Review of Algorithmic Discretion before High Courts

When public bodies or statutory corporations deploy opaque algorithms, affected citizens can invoke the extraordinary writ jurisdiction of High Courts under Article 226 of the Constitution. In judicial review, Indian courts apply the Wednesbury principle of unreasonableness and the doctrine of proportionality. If an automated decision is so irrational that no reasonable authority could ever have reached it, or if it imposes a disproportionate burden without legitimate justification, courts will issue a writ of certiorari quashing the decision.

Furthermore, superior courts will issue a writ of mandamus compelling the public authority to disclose the algorithmic parameters, training benchmarks, and validation records. High Courts have repeatedly held that the state cannot enter into proprietary secrecy agreements with commercial software vendors that deprive citizens of their constitutional right to inspect the standards governing their statutory entitlements.

Where an algorithmic system commits a demonstrable processing error—such as an automated optical mark recognition (OMR) grading system misallocating marks due to software bugs—courts will direct an independent manual re-evaluation, affirming that technology must serve justice rather than subvert it.

VI. VI. Chambers Counsel for Algorithmic Transparency and Administrative Justice

Advocate Ritik Khurana Chambers represents citizens, commercial enterprises, and public interest litigants challenging arbitrary automated decisions, administrative black-box determinations, and algorithmic exclusion before the High Court of Punjab & Haryana, the Delhi High Court, and appellate tribunals.

Our technology litigation team collaborates with forensic data scientists to deconstruct opaque algorithmic outputs, extract SHAP/LIME feature attributions, and formulate compelling constitutional petitions under Articles 14 and 21. For institutional clients deploying enterprise automated decisioning systems, our chambers designs legally defensible Explainability Charters, Adverse Action Notice architectures, and algorithmic appeal workflows.

In an increasingly automated republic, the right to a reasoned explanation is the ultimate constitutional guarantee that human agency remains sovereign over artificial intelligence.