Legal Narrative
Abstract
Artificial intelligence has transformed discrimination from an identifiable human act into a potentially invisible process of classification, prediction and automated targeting. Algorithms increasingly influence healthcare, employment, credit, insurance, policing, political communication, content recommendation, surveillance and national security. The conventional discussion of algorithmic bias focuses predominantly on discriminatory outputs such as biased recruitment. This article argues that such an approach is incomplete. Algorithmic systems may affect an individual before a formal decision is made by constructing profiles based on health, location, behaviour, political preferences, economics, psychological, political, medical or security related.
This also examines whether Article 14 of the Indian Constitution can reach such automated decision making. It develops the concept of an “Algorithmic Article 14” under which constitutional equality would extend beyond the final automated decision to the data, classification, prediction, targeting and feedback mechanisms through which the decision is produced. It analyses the black-box problem, proxy discrimination, digital redlining, algorithmic amplification, filter bubbles, hyper nudging & predictive chilling effects. The article proposes that Article 14 must be supplemented by transparency, algorithmic impact assessments, bias audits, human review, explainability and effective remedies.
Keywords: Algorithmic Bias, Black Box, Algorithmic Amplification, National Security, Chilling Effect, Article 14
Introduction
The emergence of Artificial intelligence and automated decision-making has fundamentally altered the manner in which individuals are classified, evaluated and treated by institutions. Decisions that were traditionally made through identifiable human judgment are increasingly influenced, assisted or replaced by algorithmic systems capable of processing vast quantities of personal and behavioral data. The significance of this transformation, however lies not merely in the speed or efficiency of automated decision making, but in its capacity to produce unequal consequences without apparently discriminatory rule or an identifiable human discriminator. Traditional anti- discrimination law generally begins with an observable act; an individual is denied employment, credit, healthcare and other opportunities on a particular ground. This phenomenon may be understood as pre decisional discrimination, where the discriminatory effect occurs at the stage of visibility, classification or targeting rather than at the final decision. Recent scholarship has increasingly moved beyond the simplistic understanding of the “black box” as merely a problem of technical secrecy and has examined the design choices, objectives, inputs, constraints and tolerated errors embedded within algorithmic systems. The question for equality law is consequently not only whether an algorithm is transparent but whether the architecture of the system itself permits unjustified differential treatment.
This distinction is crucial for Article 14 of the Constitution of India. Article 14 guarantees equality before law and equal protection of the laws, while Indian constitutional jurisprudence has developed the principle that arbitrary State action inconsistent with equality. The technological challenge is that discrimination may now occur without an openly discriminatory rule.
The 2019 Facebook housing-discrimination proceedings in the United States provide an important illustration. The U.S. Department of Housing and Urban Development alleged that Facebook's advertising mechanisms used data, machine learning and prediction techniques in ways that could recreate protected-class groupings and restrict access to housing advertisements. Meta subsequently entered into settlement and altered its systems. It demonstrates that algorithmic systems can reproduce inequality even where discrimination is embedded in the architecture of targeting rather than expressed as an explicit rule.
Algorithmic Discrimination Beyond the Final Decision
Algorithmic discrimination cannot be confined to the final decision produced by an automated system. The black box problem makes it difficult to determine how data, variables, objectives and optimization choices translate into an outcome. A 2026 analysis of predictive uncertainty argues that legal attention should be directed towards the design choices embedded within machine-learning systems: what data are selected, what objectives are optimized, what constraints are imposed and which errors are tolerated. Thus, the black box should not become a constitutional black hole into which accountability disappears.
The central proposition of this article is that Article 14 should evolve into what may be called an “Algorithmic Article 14.” The conventional model is State action proceeded by decision affects individuals’ then judicial review. But the model should be following the step as data collection, profiling, classification, prediction, targeting, human decision contributes to individual harm. Constitutional review should therefore be capable of examining the relevant stages where the State is responsible for the system.
This is particularly important because algorithmic discrimination can be:
1) Direct - the system explicitly uses a protected characteristic
2) Indirect – It implies neutral variables disproportionately disadvantage a group
Another kind includes proxy based, historical, predictive, geographical and feedback based.
Multidimensional Impact of Algorithmic Bias
The consequences extend beyond conventional employment discrimination. In healthcare, Obermeyer et al. demonstrated that a widely used health risk algorithm systematically underestimated the needs of black patients because healthcare expenditure was used as a proxy for health need.
One of the most important influential AI bias investigations came from MIT Media Lab’s Joy Buolamwini in 2018, She discovered that facial recognition by major tech firms misidentified darker skinned women at higher rate than lighter skinned men.
Economic and geographical profiling is another area of concern. Computer systems that make decisions about who gets a credit card, an insurance policy, a home mortgage, or a job can use location, employment, or behavioral information to deny services to minorities. The issue of digital redlining focuses on the ability of technology to replicate existing general social and geographic trends. The 2026 lawsuit against Workday’s AI-driven hiring software is also relevant to this problem because the case demonstrated that the gaps in employment information could be used as a proxy for disability.
Political decisions making creates another dimension. The Cambridge Analytica controversy demonstrated how personal information could be used for voters profiling and political microtargeting. Algorithmic systems can consequently create different information environments for citizens through filter bubbles, algorithmic amplication and hyper nudging. The problem becomes more acute when algorithms are used for surveillance. An individual may refrain from participating because of anticipated consequences of being identified known as predictive chilling effect and such circumstances bring Article 14 into direct conversation with Article 19 & 21 of the Constitution.
A related but comparatively under examined problem is geographical algorithmic discrimination, where predictive systems associate neighborhoods, districts or locations with crime, poverty, political activities and all can become proxies for those residing there. Here comes the concept of digital redlining is particularly relevant because systems can reproduce existing social and economic boundaries through apparently neutral classifications. Another is National Security that creates an equally difficult balance. Recent international reporting has also highlighted the growing exploitation of online and gaming environments for extremist engagement with younger audiences.
Article 14 & The Algorithmic Chain
Article 14 can be used as a constitutional yardstick to measure against State arbitrariness; in E.P. Royappa v. State of Tamil Nadu, the Supreme Court expanded the notion of equality to include the nexus between arbitrariness and equality, and Shayara Bano v. Union of India subsequently reaffirmed manifest arbitrariness as a constitutional basis in suitable contexts. These concepts form a possible doctrinal basis upon which to examine automated State decision-making. The constitutional analysis must therefore run through the whole chain, as to collect data then profiling followed by classification after that do the prediction then targeting after that reach to decision with human review & feedback. Wherever State action is involved, an algorithm should not be allowed to accomplish indirectly what the State could not accomplish constitutionally directly. Article 14's chief limitation remains its application to State action; purely private algorithms therefore require complementary statutory and regulatory mechanisms.
Algorithmic decision-making can also impinge upon Articles 19 and 21. In K.S. Puttaswamy v. Union of India, the Supreme Court recognized privacy as constitutionally protected and associated it with dignity, autonomy and liberty.⁶ Anuradha Bhasin v. Union of India further commented on the constitutional significance of a chilling effect on expression.⁷ Consequently, profiling and surveillance via algorithms can simultaneously involve issues of equality, informational privacy, freedom of expression and peaceful participation.
Towards An “Algorithmic Article 14” its framework and Recommendations
The article suggests a new 'Algorithmic Article 14' as an interpretative framework, but its remit should probe beyond the destination in the form of an output. It must question;
1) The ends that the system is seeking to achieve
2) The data and variables that inform it
3) Discriminatory proxies
4) Distributions
5) Predictive uncertainty
6) Explainability
7) Meaningful human review and
8) Access to an effective remedy
Here the important question raises that what if humans stop questioning the machine?
Human oversight should not be reduced to the mere presence of a human at the end of an automated process; meaningful oversight requires sufficient authority, information and competence to interrogate, explain and, where necessary, override the algorithmic outcome.
Conclusion
Algorithmic decision-making has turned discrimination from an identifiable human act into a process that may operate invisibly through data, profiling, prediction and targeting. The cases and examples discussed show that unequal effects may arise not only from the final decision but also from the algorithmic architecture preceding it, including biased datasets, proxies, geographical classifications, predictive models and feedback loops. Thus Article 14 must not be limited to asking what an automated system decided, but must also look at how and on what basis the individual was classified.
The proposed concept of “Algorithmic Article 14” aims to extend the constitutional scrutiny where State action is involved to the relevant algorithmic chain while recognising Article 14’s limitation to State action. Its aim is not to prohibit automation but to ensure that technological efficiency does not mask arbitrariness, discrimination or denial of effective remedies. Ultimately, constitutional accountability must remain attached to the exercise of public power, even if this power is increasingly mediated through automated systems.
References
1) See Ziad Obermeyer et al., Dissecting Racial Bias in an Algorithm Used to Manage the Health of Populations, 366 Science 447, 447–53 (2019) (demonstrating how an apparently neutral cost-based proxy generated racial disparities in identifying patients for additional care)
2) See U.S. Dep't of Hous. & Urban Dev., HUD Charges Facebook with Housing Discrimination Over Company's Targeted Advertising Practices (Mar. 28, 2019). HUD alleged that Facebook's advertising system used user data, machine learning, and prediction techniques in ways that could reproduce protected-class groupings and restrict access to housing-related advertisements.
3) See E.P. Royappa v. State of Tamil Nadu, (1974) 4 SCC 3 (recognizing the constitutional relationship between equality and arbitrariness in State action)
4) See Holli Sargeant, From Estimation to Discrimination: Algorithmic Bias, Predictive Uncertainty, and Anti-Discrimination Law, 89 Mod. L. Rev. 593, 593– 627 (2026). The recent literature specifically examines how predictive uncertainty and design choices affect the legal assessment of algorithmic discrimination. For a current interdisciplinary treatment of the mismatch between technical fairness and anti-discrimination law.
5) See Algorithmic Proxy Discrimination and Its Regulations, 54 Computer L. & Sec. Rev. 106021 (2024) (discussing how facially neutral variables may operate as proxies for legally protected characteristics and why such relationships can be difficult to detect in opaque systems).
6) Ziad Obermeyer et al., Dissecting Racial Bias in an Algorithm Used to Manage the Health of Populations, 366 Science 447, 447–53 (2019), https://doi.org/10.1126/science.aax2342. The study found that Black patients assigned the same risk score were, on average, sicker than White patients because healthcare expenditure was used as a proxy for healthcare need
7) Id
8) Fed. Trade Comm'n, In re Cambridge Analytica, LLC, No. 182 3107, Administrative Complaint (July 24, 2019). The FTC alleged that Cambridge Analytica obtained Facebook information and used it for voter profiling, microtargeting and related services to political campaign
9) U.N. Office of Counter-Terrorism, Artificial Intelligence and Preventing Violent Extremism, noting that AI, algorithmic systems, social-media and gaming platforms and other digital environments are increasingly exploited by terrorist actors and their supporters, including through harmful narratives and algorithmic amplification
10) See Bridging Fairness in Machine Learning: A Legal Critique of Technical Strategies, 61 Computer L. & Sec. Rev. 106339 (2026) (arguing that technical fairness mechanisms alone may fail to capture individualised harm, proportionality and procedural safeguards)
11) See Predictive Policing and Predictive Justice: Ethics, Data Protection, and the AI Act, 61 Computer L. & Sec. Rev. 106282 (2026) (highlighting transparency, validation and fundamental-rights concerns associated with predictive systems in policing and justice)
CASES
12) E.P. Royappa v. State of Tamil Nadu, (1974) 4 SCC 3; Shayara Bano v. Union of India, (2017) 9 SCC 1. Royappa linked arbitrariness in State action with the constitutional guarantee of equality, while the majority opinions in Shayara Bano recognized manifest arbitrariness as a ground of constitutional invalidity in the context considered there
13) K.S. Puttaswamy (Retd.) v. Union of India, (2017) 10 SCC 1. The nine-judge Constitution Bench recognized privacy as a constitutionally protected right and connected privacy with dignity, autonomy and liberty.
14) Anuradha Bhasin v. Union of India, (2020) 3 SCC 637. The Supreme Court discussed the chilling-effect doctrine in relation to restrictions affecting freedom of expression and emphasized the importance of evidence in establishing a legally cognizable chilling effect.
