Advancing
Research Through
Open Knowledge

The Requirement Of An Intelligible Differentia As A Constraint Upon Machine-Learned Classification By The State: Article 14 Of The Constitution Of India And The Statutory Residue Of The Criminal Tribes Act, 1871

The literature upon algorithmic discrimination rests upon two foundations, neither of which is Indian. The first is the doctrine of disparate impact, an outcome-oriented enquiry directed at what a classifier has done to a protected group. The second is Article 22 of the General Data Protection Regulation, which the Digital Personal Data Protection Act, 2023 conspicuously declines to reproduce. The Indian equality guarantee is differently constructed. Since State of West Bengal v. Anwar Ali Sarkar , a classification effected by the State must rest upon an intelligible differentia bearing a rational nexus to its object; and the decision in Shayara Bano v. Union of India has added that a classification defensible by no determining principle is manifestly arbitrary and void. The adjective has never been required to perform any work. This article contends that intelligibility is a condition upon the legibility of the classification itself rather than a description of its outcome, and that a classifier whose decision boundary cannot be stated therefore fails the first limb of the classification test by reason of opacity alone, independently of proof of disparate effect. Three conclusions follow. The requirement is addressed simultaneously to the court, to the administrative authority and to the person classified, and post hoc explanation satisfies none of them. A defence founded upon demonstrated accuracy would convert Article 14 from a reason-giving norm into a consequentialist one. Finally, the records upon which Indian policing systems are trained carry the sediment of the Criminal Tribes Act, 1871, so that what repeal removed from the statute book the training corpus has restored silently.