The use of artificial intelligence and machine learning in decisions that directly affect people's lives has grown substantially in India — both in government programmes and private sector services. From welfare disbursements to job screening, the outputs of these systems have real consequences for real people. Yet there is a dimension of this shift that has not received nearly enough attention in public or legal debate: that these algorithms are not neutral. They carry biases, sometimes subtle, sometimes glaring, and by doing so may be quietly undermining one of India's most fundamental constitutional guarantees — the right to equality under Article 14. This paper examines that intersection. The central argument is straightforward: machine learning systems are trained on data produced by human societies, and if those societies have histories of caste discrimination, gender inequality, or economic exclusion, the algorithm will replicate those patterns. A credit-scoring model may penalise applicants from particular postcodes. A hiring filter may systematically screen out certain names. A welfare verification system may fail disproportionately for manual labourers whose fingerprints have worn down. In every case, the discrimination is real — it is just dressed in the language of mathematics. Drawing on both computer science and constitutional law, this paper shows how algorithmic opacity, proxy variables, and skewed training data produce outcomes that cannot survive scrutiny under Article 14. Supreme Court precedents such as E.P. Royappa v. State of Tamil Nadu, Maneka Gandhi v. Union of India, and Shayara Bano v. Union of India are examined to show that the doctrine of manifest arbitrariness applies squarely to automated decision-making by the state. The paper also addresses the very real difficulties involved in actually regulating AI — from the impenetrability of deep learning models to the knowledge gap between judges and engineers, from jurisdictional complications with foreign-hosted systems to corporate trade secrecy laws that shield algorithms from scrutiny. Against this backdrop, the paper puts forward concrete suggestions: mandatory pre-deployment audits, a statutory right to explanation for affected individuals, constitutional compliance certification for high-risk algorithms, and a dedicated regulatory authority with genuine enforcement powers. The core claim running through all of this is simple: computational opacity is not merely a technical inconvenience. When it enables discrimination against citizens, it becomes a constitutional problem that demands legislative action.
Rauf et al. (Thu,) studied this question.