Artificial Intelligence in Administrative Decision-Making

Artificial Intelligence (AI) is increasingly being used to assist governments in processing information, identifying patterns, assessing applications, detecting irregularities and delivering public services. Its use in administration can improve speed and efficiency, but it also creates important questions concerning legality, fairness, transparency, natural justice, equality, privacy, accountability and judicial review.

Administrative decision-making traditionally involves human officials exercising powers granted by law. AI introduces a new possibility: a decision may be supported, recommended or substantially influenced by an algorithmic system.

The central Administrative Law question is therefore not simply whether AI can make administration faster, but whether AI-assisted administrative action remains consistent with the constitutional and legal limits on public power.

Meaning of Artificial Intelligence in Administrative Decision-Making

Artificial Intelligence in administrative decision-making refers to the use of AI systems to process information and assist or influence decisions taken by public authorities.

AI may be used for:

• Processing large quantities of government data;
• Detecting suspected fraud;
• Identifying patterns in regulatory compliance;
• Assessing eligibility for public schemes;
• Risk assessment;
• Tax administration;
• Predictive analysis;
• Resource allocation;
• Regulatory enforcement;
• Prioritising cases; and
• Providing recommendations to administrative officials.

AI can therefore operate at different levels.

Decision Support

The AI system provides information or recommendations, while the final decision is made by an official.

Automated Decision-Making

The system itself determines an outcome according to programmed rules or an algorithm.

AI-Assisted Decision-Making

An official makes the final decision but relies substantially on AI-generated analysis or recommendations.

The legal consequences may differ depending upon the degree to which AI influences the final administrative decision.

Growth of AI in Public Administration

Governments increasingly possess enormous quantities of digital information.

AI can process information much more rapidly than conventional administrative methods.

Potential applications include:

• Tax compliance;
• Welfare administration;
• Public procurement;
• Immigration administration;
• Fraud detection;
• Environmental regulation;
• Traffic management;
• Public health administration;
• Revenue collection; and
• Regulatory monitoring.

The increasing use of AI therefore makes it necessary to apply traditional Administrative Law principles to technologically assisted government action.

AI and the Principle of Legality

The first requirement of administrative action is legal authority.

An administrative authority cannot acquire a new substantive power merely because an AI system is capable of exercising it.

For example, if legislation gives an authority discretion to determine eligibility according to specified statutory criteria, an AI system cannot lawfully introduce additional substantive conditions without legal authority.

AI must therefore operate within:

• The Constitution;
• Parent legislation;
• Delegated legislation;
• Applicable administrative rules; and
• Procedural requirements.

Technology can assist the exercise of a legal power, but it cannot itself create that power.

AI and the Rule of Law

The Rule of Law requires governmental power to be exercised according to law.

In AI-assisted administration, this requires:

• A lawful basis for using AI;
• Clearly defined administrative powers;
• Consistent application of legal standards;
• Accountability for decisions;
• Procedural safeguards; and
• Access to remedies.

An administrative authority cannot avoid legal responsibility by arguing that an adverse outcome was produced by a computer or algorithm.

AI and Article 14

Article 14 guarantees equality before the law and equal protection of the laws.

AI systems may create Article 14 concerns if they produce:

• Arbitrary classifications;
• Unequal treatment;
• Discriminatory outcomes;
• Inconsistent results for similarly situated persons; or
• Decisions based on irrelevant factors.

An algorithm can therefore become an instrument of arbitrary administration if its design or underlying data produce unjustified distinctions.

E.P. Royappa v. State of Tamil Nadu (1974)

The Supreme Court recognised the connection between Article 14 and protection against arbitrary State action.

This principle is particularly relevant where an AI system influences the exercise of public power.

AI and Article 21

Article 21 protects life and personal liberty.

AI-based administrative systems may affect Article 21 where they process information concerning:

• Identity;
• Health;
• Location;
• Financial circumstances;
• Employment;
• Education;
• Communications; or
• Other personal information.

Where AI substantially affects an individual’s liberty, dignity or privacy, constitutional safeguards become particularly important.

AI and the Right to Privacy

AI systems often depend upon large datasets.

Government AI applications may involve the collection, combination or analysis of personal information.

Justice K.S. Puttaswamy (Retd.) v. Union of India (2017)

The Supreme Court recognised privacy as a constitutionally protected right.

The judgment is important for AI-based administration because large-scale data processing may affect:

• Informational privacy;
• Personal autonomy;
• Dignity;
• Control over personal information; and
• Freedom from unjustified State intrusion.

The use of AI therefore requires attention to the legal basis and safeguards governing government data processing.

AI and Data Protection

The effectiveness of an AI system depends substantially on the data supplied to it.

Poorly governed data can create administrative problems.

Important issues include:

• Lawful collection;
• Purpose of data processing;
• Data accuracy;
• Data security;
• Access controls;
• Retention;
• Disclosure; and
• Correction of inaccurate information.

The Digital Personal Data Protection Act, 2023 is part of India’s contemporary statutory framework concerning digital personal data, subject to its scope and applicable provisions.

Importance of Data Accuracy

An AI system can only be as reliable as the information on which it operates.

Incorrect government records can produce incorrect AI recommendations or decisions.

For example, an inaccurate database entry may cause an AI system to:

• Incorrectly classify a person;
• Deny eligibility;
• Flag a person for investigation;
• Assign an inappropriate risk category; or
• Produce an erroneous recommendation.

An effective administrative framework should therefore provide appropriate mechanisms for correcting inaccurate information.

AI and Natural Justice

Natural justice is particularly significant when AI contributes to an adverse administrative decision.

Its application depends upon the nature of the decision, governing law and consequences for the affected person.

The principal concerns include:

• Notice;
• Opportunity to respond;
• Disclosure of relevant material;
• Impartiality; and
• Reasons.

AI should not make these safeguards meaningless.

Right to Know the Basis of a Decision

Suppose a citizen’s application for a government benefit is rejected because an AI system classifies the person as ineligible.

The person may reasonably need to know:

• What legal criteria were applied;
• What information was considered;
• Whether the information was accurate;
• Why the person was classified as ineligible; and
• How the decision can be challenged.

Simply stating that “the algorithm rejected the application” may not provide a meaningful explanation where the law requires reasons.

AI and the Duty to Give Reasons

Reasoned decision-making is an important component of administrative accountability.

S.N. Mukherjee v. Union of India (1990)

The Supreme Court recognised the importance of recording reasons in administrative and quasi-judicial decisions, subject to recognised exceptions.

In AI-assisted administration, reasons become particularly important because individuals and reviewing courts may otherwise find it difficult to understand how an outcome was reached.

The use of a technologically complex system cannot by itself eliminate a legal duty to provide reasons.

Explainability and Administrative Law

Explainability refers to the ability to understand how an AI system reached or contributed to a particular outcome.

It becomes particularly important when an AI system affects:

• Benefits;
• Licences;
• Penalties;
• Regulatory action;
• Public employment;
• Tax liabilities; or
• Other significant legal interests.

The required degree of explanation may depend upon the nature and consequences of the administrative decision.

Algorithmic Bias

AI systems can reproduce or amplify bias present in their:

• Training data;
• Historical records;
• Variables;
• Classification methods;
• Design assumptions; or
• Human decision-making processes.

For example, if historical administrative data reflects unequal treatment, an AI system trained on that data may reproduce the same pattern.

This can create concerns under:

• Article 14;
• Anti-discrimination principles;
• Natural justice; and
• Administrative reasonableness.

Sources of Algorithmic Bias

Algorithmic bias may arise from:

Biased Training Data

Historical data may reflect past discriminatory practices.

Incomplete Data

Certain groups may be inadequately represented.

Proxy Variables

Apparently neutral information may indirectly correspond to legally sensitive characteristics.

Design Bias

The way a system is programmed may favour particular outcomes.

Feedback Loops

An algorithm’s earlier decisions may influence the data used to make later decisions, reinforcing existing patterns.

AI and Administrative Discretion

Administrative discretion is a central subject of Administrative Law.

AI can reduce discretion by standardising administrative decisions.

This may improve consistency, but it can also create legal problems.

Suppose legislation requires an official to consider individual circumstances before deciding whether to grant a benefit.

If the official simply follows an AI recommendation without considering the person’s individual circumstances, the statutory discretion may effectively have been surrendered to the machine.

This can raise concerns similar to the doctrine against fettering of administrative discretion.

Human Oversight

Human oversight is therefore particularly important in high-impact administrative decisions.

Officials should be capable of:

• Reviewing AI recommendations;
• Identifying obvious errors;
• Considering individual circumstances;
• Examining disputed data;
• Rejecting an inappropriate AI recommendation; and
• Taking responsibility for the final decision.

Human involvement should be meaningful rather than merely formal.

AI and Delegated Legislation

AI systems used by government may operate under rules and regulations made through delegated legislation.

Such rules must remain within the authority granted by the parent statute.

The use of AI cannot justify:

• Excessive delegation;
• Creation of new substantive obligations without authority;
• Unlawful restrictions on rights; or
• Circumvention of statutory safeguards.

The ordinary principles governing delegated legislation therefore remain applicable.

AI and Procedural Fairness

AI may change the traditional structure of administrative procedure.

A conventional administrative decision might involve:

Application → Officer → Evidence → Hearing → Decision

An AI-assisted system might involve:

Application → Data Processing → Algorithmic Assessment → Official Review → Decision

The second model can be lawful only if the legally required procedural safeguards remain effective.

AI and Disclosure of Evidence

Where an adverse decision is based upon information that must legally be disclosed to the affected person, the fact that the information was processed by AI does not automatically remove the disclosure requirement.

The individual may need access to relevant information necessary to meaningfully challenge the decision.

This becomes particularly important where incorrect data has contributed to an adverse outcome.

AI and Right to Cross-Examination

Cross-examination is not automatically required in every administrative proceeding.

Where credibility or disputed evidence makes cross-examination necessary for a fair decision, the use of AI cannot eliminate that legal requirement.

The appropriate procedural safeguards depend upon:

• The nature of the proceeding;
• Governing rules;
• The evidence involved; and
• The consequences of the decision.

AI and Public Procurement

AI systems may also be used in government procurement.

They may assist with:

• Evaluating tender information;
• Identifying irregularities;
• Comparing bids;
• Detecting potential collusion;
• Monitoring contracts; and
• Assessing supplier risk.

However, procurement decisions remain subject to principles of:

• Equality;
• Fairness;
• Transparency;
• Non-arbitrariness; and
• Judicial review.

Tata Cellular v. Union of India (1994)

The Supreme Court emphasised that judicial review of government contracts primarily concerns the legality and fairness of the decision-making process rather than replacing administrative or commercial judgment with that of the court.

This principle remains relevant where AI is used to assist procurement decisions.

AI and Welfare Administration

AI may assist governments in identifying beneficiaries and detecting irregularities in welfare programmes.

Potential benefits include:

• Faster processing;
• Fraud detection;
• Better targeting;
• Data analysis; and
• Administrative efficiency.

However, erroneous classifications can result in wrongful denial of benefits.

Therefore, welfare systems using AI should have mechanisms for:

• Verification;
• Correction;
• Human review;
• Appeals or grievances; and
• Protection against arbitrary exclusion.

AI and Digital Identity

AI can be integrated with digital identity and authentication systems.

The Aadhaar framework provides an important example of the constitutional issues associated with technology-based identification and government service delivery.

Justice K.S. Puttaswamy (Retd.) v. Union of India (2018)

The Supreme Court examined the Aadhaar framework in relation to identity, welfare delivery, privacy and proportionality.

The case illustrates how technological administrative systems can be assessed through constitutional principles.

AI and Transparency

Transparency becomes more difficult when AI systems are technically complex.

Administrative authorities may need to provide appropriate information concerning:

• The purpose of the AI system;
• The legal basis for its use;
• The type of information considered;
• The role of the system in the decision;
• Human oversight;
• Available review mechanisms; and
• Relevant limitations.

Transparency does not necessarily mean that every technical detail or source code must always be disclosed.

The important question is whether the affected person and reviewing institutions have sufficient information to understand and scrutinise the legally significant decision.

AI and Accountability

A central principle of Administrative Law is that public power must remain accountable.

AI can create an accountability gap if responsibility is unclear.

Responsibility may potentially be divided among:

• The government department;
• Officials using the system;
• Developers;
• Contractors;
• Data providers; and
• System administrators.

From an Administrative Law perspective, however, the existence of multiple technological actors should not obscure the responsibility of the public authority exercising the statutory power.

AI and Judicial Review

Judicial review remains applicable to AI-assisted administrative decisions.

Courts may examine:

• Whether the authority had jurisdiction;
• Whether the correct legal standard was applied;
• Whether relevant considerations were taken into account;
• Whether irrelevant considerations influenced the decision;
• Whether the process was procedurally fair;
• Whether the decision was arbitrary;
• Whether constitutional rights were affected; and
• Whether the decision-making process was proportionate where proportionality applies.

The court does not necessarily need to become an expert in the underlying technology to determine whether the administrative authority acted within law.

AI and Wednesbury Reasonableness

The traditional principle of reasonableness remains relevant to AI-assisted administration.

A decision may be vulnerable to judicial review where the administrative authority relies on an AI-generated outcome that is so unreasonable or disconnected from the statutory purpose that it cannot reasonably be justified.

The technological origin of a decision does not protect it from review.

AI and Proportionality

Proportionality becomes especially important where AI systems interfere with Fundamental Rights.

For example, an AI-based surveillance or risk-assessment system may affect privacy.

The authority may need to demonstrate:

• A legitimate objective;
• A rational connection between the measure and objective;
• Necessity or availability of less restrictive alternatives where applicable; and
• An appropriate balance between public objectives and individual rights.

AI and Judicial Review of Reasons

A court reviewing an AI-assisted decision may need to distinguish between:

The existence of an algorithmic output and the existence of a legally adequate administrative reason.

An administrative authority cannot necessarily defend a decision simply by producing a computer-generated score.

The legal question remains whether the authority exercised its statutory power according to law.

AI and Accountability Through Record Keeping

Electronic records can help establish:

• What information was supplied to the system;
• Which version of a model or rule was used;
• What recommendation was produced;
• Whether an official reviewed the recommendation;
• What final decision was taken; and
• When each step occurred.

Maintaining appropriate records is therefore important for:

• Internal accountability;
• Appeals;
• Audits;
• RTI processes where applicable; and
• Judicial review.

AI and Right to Information

The Right to Information Act, 2005 can become relevant where AI systems are used by public authorities.

Potentially relevant information may include:

• Policies governing AI use;
• Administrative guidelines;
• Records concerning decisions;
• Data-related policies;
• Contracts with technology providers; and
• Other information held or controlled by the public authority,

subject to the statutory exemptions and other applicable provisions.

The RTI framework therefore provides an additional transparency mechanism for technology-assisted administration.

AI and Digital Exclusion

AI-based administration may unintentionally disadvantage people who lack:

• Digital literacy;
• Reliable internet access;
• Access to technology;
• Ability to understand digital procedures; or
• Ability to challenge automated decisions.

The administrative objective of efficiency must therefore be balanced with meaningful accessibility.

AI and Persons with Disabilities

AI and digital administration should also account for accessibility requirements.

If an AI-driven public service is technically available but cannot practically be used by persons with disabilities, the system may create barriers to equal access.

Accessibility is therefore an important consideration in designing AI-enabled public administration.

AI in Regulatory Administration

Regulatory authorities may use AI for:

• Monitoring compliance;
• Identifying suspicious transactions;
• Detecting environmental violations;
• Analysing financial information;
• Predicting regulatory risks; and
• Prioritising inspections.

Such systems can help regulators handle large volumes of information.

However, regulatory action must remain based upon legally relevant criteria and cannot be justified solely by the existence of an algorithmic risk score.

AI and Government Liability

Errors caused by AI may produce administrative harm.

For example:

• An innocent person may be wrongly flagged;
• A benefit may be wrongly denied;
• A licence may be incorrectly rejected; or
• A person may be subjected to unnecessary regulatory scrutiny.

Depending upon the circumstances, legal remedies may arise through:

• Statutory appeals;
• Administrative review;
• Writ jurisdiction;
• Judicial review;
• Constitutional remedies; or
• Other applicable legal mechanisms.

The availability and nature of a remedy depend upon the governing law and facts.

Advantages of AI in Administrative Decision-Making

Faster Processing

AI can process large datasets rapidly.

Consistency

Standardised systems can reduce certain forms of inconsistent decision-making.

Fraud Detection

AI can identify unusual patterns and potential irregularities.

Better Resource Allocation

Administrative resources can be directed towards areas requiring greater attention.

Predictive Analysis

AI can assist authorities in identifying trends and risks.

Record Management

Automated systems can organise large quantities of information.

Service Delivery

AI can potentially improve responsiveness and reduce administrative delays.

Risks of AI in Administrative Decision-Making

Algorithmic Bias

Biased data or design can produce unequal outcomes.

Lack of Explainability

Complex models may make decisions difficult to understand.

Privacy Risks

AI may encourage extensive collection and analysis of personal information.

Data Errors

Incorrect input data can produce incorrect outcomes.

Automation Bias

Officials may place excessive trust in machine-generated recommendations.

Accountability Gap

It may become unclear who is responsible for an erroneous decision.

Digital Exclusion

Technology may create barriers for vulnerable groups.

Excessive Delegation to Technology

Authorities may improperly allow an algorithm to determine matters requiring human statutory discretion.

Principles for Lawful Use of AI in Administration

A legally sound AI-based administrative framework should generally incorporate:

1. Legal authority — AI must operate within statutory and constitutional powers.
2. Purpose limitation — AI should be used for a legally justified administrative purpose.
3. Transparency — affected persons should receive appropriate information about significant AI-assisted decisions.
4. Accountability — a responsible public authority must remain identifiable.
5. Human oversight — significant decisions should receive meaningful human supervision where appropriate.
6. Fairness — legally required procedural safeguards must remain effective.
7. Non-discrimination — AI systems should not produce unlawful discriminatory outcomes.
8. Data accuracy — inaccurate information should be capable of correction where legally appropriate.
9. Privacy protection — personal data must be handled according to applicable law.
10. Security — administrative datasets and systems must be protected.
11. Reasoned decision-making — legally required reasons must remain available.
12. Effective remedies — affected persons must have meaningful avenues for review or challenge.

Important Cases

E.P. Royappa v. State of Tamil Nadu (1974)

Established an important connection between Article 14 and non-arbitrariness in State action.

Maneka Gandhi v. Union of India (1978)

Emphasised fairness and non-arbitrariness in governmental procedure and strengthened the relationship between Articles 14, 19 and 21.

S.N. Mukherjee v. Union of India (1990)

Recognised the importance of recording reasons in administrative and quasi-judicial decisions, subject to recognised exceptions.

Tata Cellular v. Union of India (1994)

Established important principles governing judicial review of government contracts and procurement decisions.

L. Chandra Kumar v. Union of India (1997)

Reaffirmed the constitutional importance of judicial review under Articles 226/227 and Article 32.

State of Tamil Nadu v. P. Krishnamurthy (2006)

Set out important principles concerning judicial review of subordinate legislation.

Justice K.S. Puttaswamy (Retd.) v. Union of India (2017)

Recognised privacy as a constitutionally protected right and established an important constitutional framework for government processing of personal information.

Justice K.S. Puttaswamy (Retd.) v. Union of India (2018)

Considered the Aadhaar framework and addressed questions involving digital identity, welfare delivery, privacy and proportionality.

Difference Between Traditional and AI-Assisted Administrative Decision-Making

Traditional Administration	AI-Assisted Administration
Primarily human decision-making	Human decision-making supported by AI
Manual examination of records	Automated data processing
Physical or paper records	Electronic datasets
Individual assessment	Algorithmic assessment may assist
Human reasoning is visible	Algorithmic reasoning may be difficult to understand
Errors generally arise from human processes	Errors may arise from data, models or human use
Accountability usually rests clearly with official	Accountability may become more complex
Traditional judicial review	Judicial review must also consider technology-assisted processes

Key Points for Examination

• AI in administrative decision-making means using AI to assist, recommend or determine administrative outcomes.
• AI cannot itself create governmental power.
• Administrative authorities must remain within statutory and constitutional limits.
• Article 14 is relevant to algorithmic arbitrariness and discriminatory outcomes.
• Article 21 and the right to privacy are important where AI processes personal information.
• Puttaswamy (2017) is a leading constitutional privacy case.
• Puttaswamy (2018) is important for Aadhaar and technology-enabled welfare administration.
• Natural justice remains relevant to AI-assisted administrative decisions where its requirements are attracted.
• The use of AI does not automatically remove the requirement to give legally required reasons.
• Algorithmic bias can create equality and fairness concerns.
• Incorrect data can produce incorrect administrative outcomes.
• AI should not unlawfully replace statutory administrative discretion.
• Meaningful human oversight is important in high-impact decisions.
• AI-assisted decisions remain subject to judicial review.
• Transparency, accountability, privacy and explainability are important elements of lawful AI administration.
• The use of technology cannot become a means of avoiding constitutional or administrative-law obligations.

Conclusion

Artificial Intelligence has the potential to transform administrative decision-making by enabling governments to process large quantities of information, identify patterns, detect irregularities and deliver public services more efficiently.

However, administrative efficiency cannot be the sole consideration. When AI is used to exercise or influence public power, traditional Administrative Law principles continue to apply. Legality, equality, natural justice, reasoned decision-making, privacy, proportionality, transparency and accountability remain important safeguards.

The central challenge is to ensure that AI remains a tool of lawful administration rather than an independent source of governmental power. Public authorities must retain responsibility for AI-assisted decisions and ensure that affected individuals have meaningful opportunities to understand, challenge and seek review of decisions that affect their rights and interests.

The future of AI in Administrative Law will therefore depend upon achieving a balance between technological efficiency and constitutional governance.