Constitution Never Approved Algorithmic Governance: Why India Needs Right To Human Decision-Making

In September 2017, an eleven-year-old girl named Santoshi Kumari died of starvation in Jharkhand's Simdega district after her family's ration card was cancelled because it could not be linked to Aadhaar. No official signed off on that cancellation in the way administrative law has traditionally understood a “decision.” A biometric authentication system flagged a mismatch, a database updated a status, and a family stopped receiving grain. Nobody applied their mind to Santoshi's case. That, in miniature, is the constitutional problem posed by algorithmic governance in India: decisions that affect life and livelihood are increasingly made, or effectively made, by systems that were never contemplated by the text of the Constitution and that resist the doctrinal tools built to discipline human decision-makers.

The Constitution of India was drafted for a state that acts through officers who apply their minds, record reasons, and can be questioned. Article 14 guarantee against arbitrariness, Article 21 promise of a fair procedure, and the natural justice principles that Indian courts have read into both, all presuppose a decision-maker capable of reasoning, of being persuaded, and of explaining itself. Nothing in the Constituent Assembly Debates, and nothing in seven decades of constitutional jurisprudence, anticipated a state that governs through scoring systems, authentication mismatches, and opaque risk models. Algorithmic governance was never approved by the constitutional design; it has simply arrived, and the doctrine is being asked to stretch to cover it.

Where the machines already govern

This is not a speculative or futuristic concern. Automated and semi-automated systems already mediate some of the most consequential encounters citizens have with the Indian state. Welfare delivery under the Public Distribution System and social security pensions has, for over a decade, depended on Aadhaar-based biometric authentication; when fingerprints fail to match, at a point-of-sale device or a server, the consequence is exclusion, not error correction. The government's own submissions before the Supreme Court have acknowledged authentication failure rates that translate, at national scale, into millions of people at risk of being wrongly denied entitlements they are legally owed. The Income Tax Department's faceless assessment scheme routes disputes through automated allocation and templated processing with limited scope for a taxpayer to make a case to an identifiable human being. Predictive policing tools and crime-mapping software, adopted by several state police forces, generate risk scores that shape patrolling and surveillance priorities without any statutory framework governing their design or accountability. And credit, insurance and even loan-recovery decisions taken by regulated financial entities increasingly rest on proprietary scoring models that borrowers cannot see and cannot meaningfully contest.

In each instance, the structure of the “decision” has changed. It is not that a human officer made a biased choice, a fact pattern administrative law knows how to handle. The decision was distributed across a database, a threshold, and a rule nobody in the room can fully articulate. The result is governance that is efficient and scalable, but largely unaccountable to the doctrines meant to constrain the state.

The doctrinal strain

Indian constitutional law's principal safeguard against arbitrary state action is the requirement that a decision reflect an application of mind and be capable of being defended with reasons. This lineage runs from the Supreme Court's insistence in Maneka Gandhi v. Union of India that state action affecting life and personal liberty must be fair, just and reasonable, through decades of natural justice jurisprudence, including A.K. Kraipak v. Union of India, which extended fair-hearing principles from quasi-judicial to administrative functions precisely because the line between the two had become difficult to draw. The right to know why a decision has gone against you, so that it can be tested and, where necessary, challenged, is not a procedural nicety; it is what makes judicial review possible at all.

Justice D.Y. Chandrachud's opinion in Justice K.S. Puttaswamy v. Union of India carried this logic into the informational age, describing privacy as encompassing the individual's right to be told why the state has acted as it has, and locating that right within the dignity guaranteed by Article 21. That reasoning was written with data collection in mind, but its implications for automated decision-making are direct. If dignity requires that a person be told why the state has intruded upon them, it is difficult to see how that requirement is satisfied by a system that cannot itself produce reasons a human being can understand, because no human being wrote the rule that produced the outcome.

The strain is structural, not a matter of poor implementation. Reasoned decision-making assumes a reasoner. Audi alteram partem assumes a hearing that can change an outcome. Article 14 proscription of arbitrariness assumes an actor whose discretion can be scrutinised for whim or malice. Machine-learning systems, and even simpler rule-based automation of the kind used in welfare authentication, do not reason in a sense a court can interrogate; they classify. A model can be biased or built on flawed data, but it cannot, as a human bureaucrat can, be asked to justify itself, and its operators can often genuinely say they do not know precisely why a given output emerged. Constitutional doctrine built for accountable human agency confronts a process that is, by design, opaque even to its own creators.

Why existing statutes do not close the gap

India's two most relevant statutes fall short of addressing this in different ways. The Right to Information Act, 2005 was designed to open up files, orders and reasons; it was not designed to compel disclosure of source code, training data, or scoring logic, and public authorities have routinely resisted such disclosure by invoking the commercial confidence exemption under Section 8(1)(d), treating proprietary algorithms as trade secrets even where they determine eligibility for statutory entitlements. Transparency built around the human-era assumption that a decision leaves behind a legible file does not transfer easily to systems whose “file” is a weighted model.

The Digital Personal Data Protection Act, 2023, meanwhile, is India's first general data protection statute, but it was drafted around consent and processing of personal data rather than around the consequences of automated decisions built on that data. Unlike the European Union's General Data Protection Regulation, which gives data subjects a qualified right under Article 22 not to be subject to a decision based solely on automated processing that produces legal or similarly significant effects, the DPDP Act contains no comparable entitlement. A citizen whose welfare eligibility, tax assessment, or credit access is determined by an automated system has, under Indian law, considerably weaker footing than a counterpart in the European Union, even though the Puttaswamy court grounded India's privacy jurisprudence in language at least as protective of dignity as its European counterparts.

Towards a right to human decision-making

What India needs, whether through judicial interpretation of Articles 14 and 21 or through legislative codification, is a right to human decision-making wherever automated or algorithmic systems materially affect life, liberty, livelihood or dignity. Such a right would rest on modest, workable elements rather than a wholesale rejection of technology in governance. First, meaningful human review before any adverse action, so that a person, not merely a re-run of the same model, examines a case before an entitlement is withdrawn or a penalty imposed. Second, a duty to disclose, in plain and accessible terms, the categories of factors an automated system relied upon, without necessarily requiring release of proprietary source code, on the pattern courts have already accepted for other forms of qualified disclosure. Third, a genuine right to contest an automated outcome before it takes effect, rather than after harm has occurred, reversing the current default under which exclusion happens first and appeal, if it exists at all, happens later. Fourth, periodic, independent algorithmic audits of high-stakes public systems, with findings placed in the public domain, modelled loosely on the risk-tiered obligations found in the European Union's AI Act.

None of this requires India to slow the digitisation of governance, which has brought genuine efficiency gains. It requires acknowledging that the Constitution's safeguards were built around a particular model of the state, one that reasons, explains and can be persuaded, and that this model has not been formally revised even as the practice of governance has moved on without it. Courts have shown, in Puttaswamy and in the natural justice line running through Maneka Gandhi and Kraipak, that they are capable of reading dignity and fairness into new contexts the framers did not foresee. Algorithmic governance is the next context in which that capacity will be tested. Until the right to a human decision-maker is recognised as part of the guarantee against arbitrariness, the Constitution's promise that no person shall be deprived of life or livelihood except by a fair, reasoned and answerable process will remain, for an increasing number of Indians, a promise addressed to officials who are no longer the ones deciding.

Author is an Assistant Professor at CHRIST (Deemed to be University), Delhi NCR Campus. Views are personal.