Algorithmic Non-Discrimination Claims in India

1. Meaning and Legal Nature

Algorithmic non-discrimination claims arise when an algorithm, automated decision-making system, AI model, scoring mechanism, or data-driven classification system produces unlawfully discriminatory treatment against an individual or group.

Examples include algorithms used for:

• recruitment and employee evaluation;
• credit and loan scoring;
• insurance pricing;
• admission and examination systems;
• welfare-benefit allocation;
• policing and risk assessment;
• facial recognition;
• healthcare prioritisation;
• housing and tenancy;
• platform moderation;
• advertising and profiling;
• immigration or visa assessment; and
• government decision-making.

Indian law does not presently recognise a single, standalone statutory cause of action called an “algorithmic non-discrimination claim.” Instead, such claims are constructed through existing constitutional, statutory, contractual, employment, consumer, tort, privacy and administrative-law principles.

The central legal question is:

Can a person challenge an adverse decision when the discriminatory result was produced, wholly or partly, by an algorithm rather than directly by a human decision-maker?

The answer can be yes, where the algorithmic decision violates an applicable legal right or duty.

2. Core Legal Framework

Algorithmic discrimination in India may engage several legal regimes simultaneously.

The applicable regime depends upon who deployed the algorithm, what decision it made, what protected interest was affected, and what legal relationship existed between the parties.

3. What Constitutes Algorithmic Discrimination?

Algorithmic discrimination can occur in several ways.

A. Direct discrimination

The algorithm expressly uses a protected characteristic.

For example:

An automated recruitment system rejects all applicants identifying as women for a particular category of job.

This is the clearest form of discrimination.

B. Indirect discrimination

The algorithm does not expressly use a protected characteristic but uses another variable that produces substantially discriminatory effects.

For example:

An employment model gives substantial negative weight to career interruptions, disproportionately disadvantaging women returning from maternity or caregiving breaks.

The algorithm may never receive the field “gender,” but the outcome may still create legally significant discrimination.

C. Proxy discrimination

A seemingly neutral variable operates as a substitute for a protected characteristic.

Examples include:

• postcode functioning as a proxy for caste or socioeconomic status;
• school attended functioning as a proxy for social background;
• language patterns functioning as a proxy for ethnicity;
• employment gaps functioning as a proxy for gender;
• disability-related accommodation requirements functioning as a proxy for disability.

D. Historical-data discrimination

An AI model trained on historically discriminatory decisions may reproduce those patterns.

For example:

If historical hiring records systematically preferred male candidates, a model trained on those records may learn that male applicants are more “successful.”

The resulting discrimination can therefore arise without any explicit discriminatory instruction.

E. Intersectional discrimination

A system may disproportionately disadvantage people because of the interaction of multiple characteristics.

For example:

• women with disabilities;
• elderly women;
• persons from disadvantaged social groups with disabilities;
• transgender persons belonging to economically disadvantaged groups.

Indian constitutional jurisprudence increasingly recognises that equality cannot always be understood through isolated categories.

4. Article 14: The Central Constitutional Provision

Article 14 provides equality before law and equal protection of laws.

The Supreme Court has developed Article 14 beyond the simplistic idea that all persons must always receive identical treatment.

The doctrine prohibits arbitrariness, irrational classification and unjustified differential treatment.

This becomes particularly important for algorithms.

An algorithm may appear mathematically neutral while producing arbitrary outcomes.

For example:

An automated welfare algorithm excludes applicants because their income data contains a formatting error.

The problem is not necessarily that the algorithm intentionally discriminated. The problem may be that the decision-making mechanism was irrational, inaccurate or procedurally unfair.

5. E.P. Royappa v State of Tamil Nadu

E.P. Royappa v State of Tamil Nadu, (1974) 4 SCC 3

This is one of the foundational Indian equality decisions.

The Supreme Court moved Article 14 away from a narrow classification-based approach and emphasised the relationship between equality and arbitrariness.

Relevance to algorithms

An algorithmic decision can potentially violate Article 14 where:

• the input variables are irrational;
• the classification lacks a reasonable basis;
• similarly situated people are treated differently without justification;
• the system produces arbitrary outcomes;
• the decision-maker blindly follows an automated score.

Principle

Algorithmic neutrality in design does not necessarily produce constitutional equality in operation.

6. Maneka Gandhi v Union of India

Maneka Gandhi v Union of India, (1978) 1 SCC 248

The Supreme Court substantially expanded the relationship between Articles 14, 19 and 21.

State action affecting liberty must satisfy standards of fairness, reasonableness and non-arbitrariness.

Algorithmic significance

Where the State uses an automated system to make a decision affecting:

• liberty;
• movement;
• livelihood;
• benefits;
• licences;
• employment;
• access to public services,

the existence of an algorithm does not eliminate the requirement of fair procedure.

Affected persons may therefore challenge:

• unexplained automated decisions;
• inability to contest erroneous data;
• absence of meaningful review;
• unreasonable classification;
• automated decisions having serious consequences without procedural safeguards.

7. Ajay Hasia v Khalid Mujib Sehravardi

Ajay Hasia v Khalid Mujib Sehravardi, (1981) 1 SCC 722

The Supreme Court explained that Article 14 applies to State action and to entities that fall within the constitutional concept of “State.”

Algorithmic significance

This is important where an algorithm is operated by:

• a government department;
• statutory authority;
• public corporation;
• government-controlled entity; or
• another entity performing a sufficiently public governmental function.

The government cannot necessarily avoid constitutional scrutiny merely because it has outsourced an algorithm to a private technology company.

Example

Suppose a government welfare department purchases an AI system from a private vendor and the system automatically rejects applications.

The fact that:

“the algorithm was supplied by a private company”

does not necessarily remove constitutional responsibility from the public authority.

8. Anuj Garg v Hotel Association of India

Anuj Garg v Hotel Association of India, (2008) 3 SCC 1

This is especially important for discriminatory classifications.

The Supreme Court invalidated a statutory restriction that prevented women from working in certain establishments.

The Court emphasised that paternalistic assumptions cannot ordinarily justify discriminatory restrictions.

Algorithmic significance

Suppose an employment algorithm assumes that:

• women are less suitable for night work;
• mothers are less committed employees;
• persons with disabilities are less productive;
• particular social groups are inherently risky.

Such assumptions cannot become legally legitimate merely because they have been converted into mathematical variables.

Principle

Encoding a stereotype into software does not transform the stereotype into a constitutionally permissible classification.

9. C.B. Muthamma v Union of India

C.B. Muthamma v Union of India, (1979) 4 SCC 260

The Supreme Court addressed discriminatory service rules affecting women in public employment.

The case is significant for demonstrating that apparently formal employment rules may perpetuate gender discrimination.

Algorithmic relevance

AI employment systems can replicate precisely this problem through apparently neutral variables.

For example:

• marital status;
• maternity-related career breaks;
• age;
• travel availability;
• work-history interruptions.

If these variables disproportionately disadvantage women, the employer may face an equality challenge depending upon the applicable legal framework.

10. Air India v Nergesh Meerza

Air India v Nergesh Meerza, (1981) 4 SCC 335

The Supreme Court examined discriminatory employment conditions imposed upon air hostesses.

The Court scrutinised service conditions involving:

• retirement;
• pregnancy;
• marriage;
• employment continuation.

Algorithmic significance

The case illustrates that employment policies cannot escape equality scrutiny merely because they are incorporated into formal organisational systems.

An AI-based HR system applying discriminatory rules may therefore reproduce an unlawful employment policy at greater scale.

11. Jeeja Ghosh v Union of India

Jeeja Ghosh v Union of India, (2016) 7 SCC 761

The Supreme Court strongly emphasised dignity and the rights of persons with disabilities.

Algorithmic relevance

AI systems can create disability discrimination through:

• inaccessible interfaces;
• speech-recognition systems failing persons with speech impairments;
• facial-recognition systems performing poorly for particular disabilities;
• recruitment algorithms treating accommodation needs negatively;
• automated examination systems refusing reasonable accommodation;
• healthcare algorithms assigning lower priority to disabled persons.

A system can therefore discriminate even where disability is not expressly used as an input.

12. Vikash Kumar v UPSC

Vikash Kumar v Union Public Service Commission, (2021) 5 SCC 370

This is a particularly important modern equality decision concerning reasonable accommodation.

The Supreme Court emphasised that equality for persons with disabilities may require positive accommodation rather than merely identical treatment.

Algorithmic significance

A genuinely equal AI system cannot necessarily be designed around a “one-size-fits-all” model.

For example, an examination algorithm may need to account for:

• assistive technologies;
• additional time;
• alternative input methods;
• accessibility requirements.

Simply applying identical automated rules to everyone can itself produce substantive inequality.

13. Rajive Raturi v Union of India

Rajive Raturi v Union of India, (2024) 6 SCC 418

The Supreme Court's disability-rights jurisprudence further strengthens the concept of accessibility and substantive equality.

Algorithmic importance

Digital systems used by:

• governments;
• educational institutions;
• employers;
• banks;
• healthcare institutions;
• public-service providers

must be assessed for accessibility.

An algorithm that systematically excludes persons with disabilities can potentially produce a legally actionable form of discrimination.

14. National Federation of the Blind v UPSC

National Federation of the Blind v UPSC, (2013) 10 SCC 772

The Supreme Court recognised the importance of equal access to public employment and examination opportunities for persons with visual disabilities.

Algorithmic relevance

Automated examination and recruitment platforms must not create technological barriers that effectively exclude persons with disabilities.

This may involve:

• inaccessible interfaces;
• incompatible screen readers;
• automated rejection of assistive technology;
• inappropriate biometric authentication;
• inaccessible online examinations.

15. NALSA v Union of India

National Legal Services Authority v Union of India, (2014) 5 SCC 438

The Supreme Court recognised constitutional protection for transgender persons and emphasised equality, dignity and identity.

Algorithmic significance

Modern automated systems frequently classify individuals according to identity data.

Potential problems include:

• binary-only gender fields;
• automated identity verification failures;
• denial of services because official records do not match gender identity;
• discriminatory profiling;
• algorithmic advertising or employment decisions.

A system that cannot accommodate legally recognised identities may generate equality and dignity concerns.

16. Navtej Singh Johar v Union of India

Navtej Singh Johar v Union of India, (2018) 10 SCC 1

The Supreme Court emphasised constitutional morality, dignity, autonomy and equality.

Algorithmic significance

Algorithmic systems must not reproduce historically discriminatory assumptions concerning sexual orientation.

For example:

An advertising system might systematically exclude same-sex couples from housing advertisements.

Or:

An employment model might treat certain identity-related information as an adverse indicator.

Such systems may raise constitutional and statutory issues depending upon the context.

17. K.S. Puttaswamy v Union of India

K.S. Puttaswamy v Union of India, (2017) 10 SCC 1

The Supreme Court recognised privacy as a fundamental right under Article 21 and the broader constitutional framework.

This case is extremely important for algorithmic discrimination because modern AI systems depend heavily upon personal data and profiling.

Algorithmic profiling may involve:

• behavioural data;
• location;
• financial information;
• health information;
• browsing behaviour;
• biometric information;
• employment records;
• social relationships;
• inferred characteristics.

The privacy issue is not limited to whether data was collected.

It also concerns:

What the system infers from the data and how those inferences affect the individual.

18. K.S. Puttaswamy (Aadhaar) v Union of India

K.S. Puttaswamy (Aadhaar) v Union of India, (2019) 1 SCC 1

The Aadhaar litigation is particularly useful for analysing large-scale automated identification and authentication systems.

The Court considered questions involving:

• proportionality;
• legitimate State purpose;
• data architecture;
• authentication;
• exclusion;
• privacy;
• safeguards.

Algorithmic discrimination relevance

Automated identity systems may produce serious consequences when:

authentication failure → denial of benefit/service.

A technologically generated error can therefore become a constitutional injury if the affected person is denied an important entitlement without adequate safeguards or alternative mechanisms.

19. Shreya Singhal v Union of India

Shreya Singhal v Union of India, (2015) 5 SCC 1

The Supreme Court struck down Section 66A of the Information Technology Act.

Although the case was not about AI discrimination, it is highly relevant to digital regulation.

Algorithmic significance

The Court demonstrated that technological mechanisms remain subject to constitutional standards.

A computer system cannot convert an otherwise unconstitutional restriction into a lawful one.

This principle is important for:

• content moderation algorithms;
• automated censorship;
• platform recommendation systems;
• account suspension;
• automated speech classification.

20. Anuradha Bhasin v Union of India

Anuradha Bhasin v Union of India, (2020) 3 SCC 637

The Supreme Court emphasised proportionality and constitutional review of restrictions involving digital communications.

Algorithmic relevance

Where automated systems restrict access to digital services, courts can examine:

• legal authority;
• necessity;
• proportionality;
• duration;
• procedural safeguards;
• availability of review.

Thus:

“The system decided automatically” is not itself a legal justification.

21. A.K. Kraipak v Union of India

A.K. Kraipak v Union of India, (1969) 2 SCC 262

This case is foundational to the development of natural justice in administrative law.

Algorithmic significance

Suppose a government algorithm generates a “high-risk” classification and the authority automatically acts upon it.

Questions arise:

1. Was the decision-maker legally authorised?
2. Was relevant material considered?
3. Was irrelevant material considered?
4. Was there bias in the decision-making process?
5. Did the affected person have an opportunity to contest the adverse material?

Automation cannot automatically eliminate natural justice.

22. State of Orissa v Dr. Binapani Dei

State of Orissa v Dr. Binapani Dei, AIR 1967 SC 1269

The Supreme Court established an important principle that administrative decisions having civil consequences must satisfy procedural fairness.

Algorithmic application

An automated decision that causes:

• termination;
• loss of benefits;
• licence cancellation;
• adverse professional classification;
• exclusion from an examination;
• denial of an important service

may attract procedural fairness requirements.

The more serious the consequence, the stronger the case for meaningful human review and opportunity to contest the underlying data.

23. Mohinder Singh Gill v Chief Election Commissioner

Mohinder Singh Gill v Chief Election Commissioner, (1978) 1 SCC 405

This case is important for administrative decision-making and the requirement that public decisions be justified on legally relevant grounds.

Algorithmic relevance

A public authority should not necessarily be permitted to defend an adverse decision after litigation by introducing entirely new reasons that were not part of the original decision.

This creates an important issue for AI systems:

What was actually the reason for the algorithmic decision?

If the authority cannot identify the legally relevant basis of the decision, meaningful judicial review becomes difficult.

24. S.N. Mukherjee v Union of India

S.N. Mukherjee v Union of India, (1990) 4 SCC 594

The Supreme Court recognised the importance of recording reasons in administrative decisions.

Algorithmic importance

A black-box decision creates a potential problem where the affected person receives only:

“Your application was rejected because the system determined that you were ineligible.”

That may be inadequate where law requires reasons.

The legal requirement does not necessarily mean disclosure of source code.

It may instead require disclosure of sufficiently meaningful reasons explaining:

• relevant criteria;
• factual basis;
• applicable rule;
• significant adverse information;
• decision-making pathway.

25. Tata Cellular v Union of India

Tata Cellular v Union of India, (1994) 6 SCC 651

The Supreme Court established important principles governing judicial review of government decisions and contracts.

Algorithmic procurement relevance

Suppose a government selects an AI vendor through an algorithmic scoring system.

A disappointed bidder may challenge:

• arbitrary scoring;
• undisclosed evaluation criteria;
• conflict of interest;
• irrational weighting;
• unequal treatment;
• procedural irregularity.

The technology used in procurement does not remove Article 14 review.

26. Internet and Mobile Association of India v RBI

Internet and Mobile Association of India v Reserve Bank of India, (2020) 10 SCC 274

The Supreme Court examined the proportionality of regulatory restrictions affecting virtual-currency businesses.

Algorithmic relevance

It demonstrates the importance of proportionality where regulatory action significantly affects technological businesses.

The same reasoning may become relevant where an automated regulatory or risk-classification system imposes serious restrictions.

27. Constitutional Test for Algorithmic Non-Discrimination

A useful Indian-law framework can be expressed as follows:

Step 1 — Identify the classification

What distinction does the algorithm make?

Examples:

• gender;
• disability;
• caste;
• religion;
• age;
• location;
• language;
• identity;
• economic status.

Step 2 — Identify the legal right

The claimant must identify the legal interest affected.

For example:

• public employment;
• government benefit;
• education;
• liberty;
• privacy;
• dignity;
• access to public services.

Step 3 — Identify the source of discrimination

The discriminatory effect may originate from:

• training data;
• labels;
• variables;
• proxies;
• model architecture;
• thresholds;
• ranking mechanisms;
• deployment conditions;
• human use of algorithmic outputs.

Step 4 — Establish differential treatment

The claimant should show that similarly situated persons were treated differently or that a formally neutral system produces a legally significant discriminatory disadvantage.

Step 5 — Examine justification

The respondent may attempt to establish:

• legitimate objective;
• rational connection;
• necessity;
• proportionality;
• statutory authority;
• objective justification.

Step 6 — Examine procedural safeguards

The court may consider:

• notice;
• explanation;
• human review;
• opportunity to correct data;
• appeal;
• audit;
• independent review.

Step 7 — Establish injury

Potential injury includes:

• denial of employment;
• loss of promotion;
• denial of welfare benefits;
• denial of credit;
• exclusion from education;
• denial of healthcare;
• reputational damage;
• privacy injury;
• financial loss;
• dignity injury.

28. Direct vs Indirect Algorithmic Discrimination

A claimant therefore does not necessarily need to prove that a programmer consciously intended to discriminate.

The legal question may instead concern the operation and consequences of the system.

29. Burden of Proof and Evidence

Algorithmic discrimination presents an unusual evidentiary problem.

The claimant may not know:

• what variables were used;
• what model was deployed;
• what training data was used;
• what threshold caused rejection;
• whether a proxy variable was used;
• whether human intervention occurred.

Useful evidence may include:

Technical evidence

• model documentation;
• model cards;
• audit reports;
• training-data documentation;
• feature importance;
• validation reports;
• error rates;
• false-positive rates;
• false-negative rates;
• logs;
• version history.

Legal evidence

• employment policy;
• procurement contract;
• statutory rules;
• departmental guidelines;
• internal decision criteria;
• reasons for rejection.

Comparative evidence

The claimant may compare:

similarly situated persons who received different outcomes.

This can be particularly important in equality litigation.

30. Statistical Evidence

Statistical disparity can be highly relevant but is not automatically equivalent to unlawful discrimination.

For example:

Suppose:

• Group A: 80% approved;
• Group B: 45% approved.

That disparity may raise a serious question.

But a court would still need to examine:

• why the disparity exists;
• the relevant legal classification;
• whether the difference is justified;
• whether the system is accurate;
• whether protected characteristics are involved;
• whether the decision concerns a constitutional or statutory right;
• whether there is an applicable legitimate objective.

Therefore:

Statistical disparity may be evidence of discrimination, but it is not necessarily conclusive proof of legal liability.

31. Human Oversight

One of the most important safeguards is meaningful human review.

Merely saying:

“A human approved the AI decision”

may not be sufficient if the human simply rubber-stamps the algorithm.

Meaningful human review should potentially involve:

1. access to relevant information;
2. ability to question the AI output;
3. authority to override the algorithm;
4. independent assessment;
5. consideration of individual circumstances;
6. documented reasons.

This is particularly important where the decision has serious consequences.

32. Public-Sector vs Private-Sector Algorithmic Discrimination

Public Sector

The strongest constitutional protections may apply.

Potential grounds include:

• Articles 14, 15, 16 and 21;
• administrative law;
• natural justice;
• proportionality;
• statutory duties;
• disability rights;
• privacy.

Judicial review may be available.

Private Sector

The position is different.

A private company is not automatically subject to every constitutional equality obligation merely because it uses an algorithm.

Possible claims may instead arise under:

• employment law;
• contract;
• consumer protection;
• disability legislation;
• privacy/data protection;
• tort;
• sectoral regulation;
• specific statutory anti-discrimination provisions.

Therefore, one should not mechanically apply Article 14 to every private AI system.

33. Algorithmic Discrimination in Employment

Common examples include:

• AI recruitment;
• automated CV screening;
• performance scoring;
• promotion recommendations;
• attendance monitoring;
• productivity scoring;
• dismissal-risk predictions;
• salary recommendations.

Potential legal issues

An employer could face a claim where the system:

• systematically penalises women for career breaks;
• disadvantages disabled employees;
• uses discriminatory historical performance data;
• penalises employees for legitimate accommodation;
• treats protected characteristics as negative predictors.

Cases such as C.B. Muthamma, Air India v Nergesh Meerza, Anuj Garg, Vikash Kumar and Jeeja Ghosh provide important legal principles.

34. Algorithmic Discrimination in Banking and Credit

AI may determine:

• loan eligibility;
• credit limits;
• fraud risk;
• interest rates;
• customer risk;
• insurance risk.

Potential claims can involve:

• arbitrary classification;
• inaccurate personal data;
• proxy discrimination;
• lack of explanation;
• unfair contractual practices;
• consumer protection;
• privacy.

Where a regulated financial institution is involved, sectoral regulatory obligations may substantially strengthen the claimant's case.

35. Algorithmic Discrimination in Government Welfare

This is particularly significant.

An automated welfare system may reject an eligible person because:

• biometric authentication fails;
• databases do not match;
• income information is outdated;
• an algorithm classifies the person as ineligible;
• a household is incorrectly classified.

The constitutional concern becomes particularly serious when the affected benefit relates to basic livelihood or essential public services.

The Aadhaar judgment and Article 21 jurisprudence are particularly relevant.

36. Algorithmic Discrimination and Disability

Disability discrimination requires a substantive equality approach.

An algorithm may be discriminatory even if it applies the same rule to everybody.

For example:

An online examination system gives every student exactly the same amount of time.

Formal equality exists.

But a candidate legally entitled to reasonable accommodation may nevertheless be unlawfully disadvantaged.

Therefore:

Equal treatment is not always equivalent to equal opportunity.

The decisions in Jeeja Ghosh, Vikash Kumar, National Federation of the Blind and Rajive Raturi are especially relevant.

37. Algorithmic Discrimination and Privacy

Privacy and discrimination increasingly overlap.

An AI system can infer sensitive information even where the user never expressly supplied it.

For example, a model may infer:

• health condition;
• economic status;
• political preference;
• disability;
• religious affiliation;
• sexual orientation.

Such profiling can create both:

1. privacy harm, and
2. discrimination risk.

The constitutional privacy jurisprudence beginning with Puttaswamy is therefore central to algorithmic discrimination analysis.

38. Possible Remedies

Depending upon the context, remedies may include:

Constitutional remedies

Under Articles 32 and 226:

• writ of certiorari;
• mandamus;
• prohibition;
• declaration;
• quashing of discriminatory decision;
• directions for reconsideration.

Civil remedies

• damages;
• injunction;
• declaration;
• specific relief.

Employment remedies

• reinstatement;
• back wages where legally available;
• correction of employment record;
• reconsideration;
• compensation.

Disability remedies

• reasonable accommodation;
• accessibility;
• corrective action;
• compensation where authorised.

Data-related remedies

Depending upon the applicable legal framework:

• correction;
• access;
• grievance redress;
• restriction/cessation of unlawful processing;
• other statutory remedies.

39. Liability of Different Actors

Algorithmic discrimination may involve several actors.

Importantly, the existence of multiple actors does not automatically make each actor legally liable. The claimant must connect each defendant to a relevant legal duty and causation.

40. Defences

A respondent may argue:

1. the algorithm did not use a protected characteristic;
2. the disputed variable was objectively relevant;
3. the difference was statistically explainable;
4. the system was used only as an advisory tool;
5. a human independently reviewed the result;
6. the claimant was not similarly situated;
7. the classification had statutory authority;
8. the measure pursued a legitimate objective;
9. the adverse outcome resulted from inaccurate information supplied by the claimant;
10. there was no legally recognised injury.

The strength of these defences depends heavily on the applicable legal framework.

41. Important Distinction: Bias vs Legal Discrimination

Not every technically biased algorithm necessarily creates a legal claim.

For example:

An AI system performs slightly less accurately for one demographic group.

That is a technical bias problem.

It becomes a legal discrimination claim when the bias intersects with:

• a protected legal interest;
• a statutory prohibition;
• constitutional equality;
• disability rights;
• employment rights;
• consumer rights;
• privacy;
• contractual obligations; or
• another recognised legal duty.

Thus:

Technical bias ≠ automatically legal discrimination.

But technical bias can provide important evidence of discriminatory treatment.

42. Leading Case-Law Summary

43. Practical Legal Test

A useful test for an Indian algorithmic non-discrimination claim is:

Algorithmic Classification → Differential Impact/Treatment → Legally Protected Interest → Applicable Legal Duty → Unjustified or Disproportionate Differential Treatment → Causation → Legally Recognised Injury → Appropriate Remedy

For a public authority, this can be expanded to:

Legal Authority → Equality → Non-Arbitrariness → Legitimate Objective → Rational Connection → Proportionality → Procedural Fairness → Human Review → Reasons → Effective Remedy

44. Conclusion

Algorithmic non-discrimination is best understood not as a completely new branch of Indian law, but as a technological manifestation of existing equality, privacy, administrative-law, employment, disability and consumer-law principles.

The most important proposition is that:

A discriminatory decision does not become lawful merely because the decision was produced by software rather than a human being.

Similarly, an organisation cannot necessarily defend an unlawful decision simply by stating:

“The algorithm made the decision.”

Courts are likely to examine the legal authority for using the system, the data and variables used, the classification produced, the discriminatory effect, the justification offered, procedural safeguards, human oversight, reasons, causation and resulting injury.

The strongest Indian authorities for developing this field are E.P. Royappa, Maneka Gandhi, Anuj Garg, C.B. Muthamma, Air India v Nergesh Meerza, Jeeja Ghosh, Vikash Kumar, Rajive Raturi, NALSA, Puttaswamy, Puttaswamy (Aadhaar), A.K. Kraipak, Binapani Dei and S.N. Mukherjee.

In short: Indian courts already possess substantial doctrinal tools to scrutinise algorithmic discrimination even though there is not yet a single comprehensive Indian statute creating a standalone “algorithmic non-discrimination” cause of action.