Ravi is a senior police officer with vast experience in riot control and cyber-policing. Since one year, he has been the Superintendent of Police (SP) of a district with a history of frequent rioting.
Last year, Ravi had sought installation of an AI enabled software for predictive policing. This system has been operational for approximately six months. This new system employs advanced algorithms for capturing the biometric data of persons in a crowd and swiftly relating it to a data library. This has enabled the police to identify the persons involved in various crimes.
The system has identified an immigrant and low-income neighbourhood as a centre for gang violence and drug trafficking. Aided by this AI analysis, the local police has focused its patrolling, preventive detentions and establishing checkposts. Consequently, public order and law enforcement has visibly improved.
Last week, some community leaders, civil rights lawyers and human rights activists visited Ravi’s office. They submitted a memorandum that the new system is faulty as it is based on incorrect historical data caused by social biases and discriminatory policing. The memorandum also alleges that the increased surveillance has created a climate of tension amongst residents. This feeling is aggravated by the fact that the residents are not aware of the data noted against their names.
(a) What are the ethical issues including biases involved in the use of AI in data-driven policing? (b) Place yourself in Ravi’s role and discuss the alternatives available. Justify the action that optimises compliance with ethics.
Approach · directive: “what / discuss / justify”
What it asks · Identify the ethical problems and biases in AI-driven predictive policing and, as SP Ravi, weigh the alternatives and justify the most ethical course.
The question has 3 parts — answer each
- (a) Identify the ethical issues, including biases, in the use of AI for data-driven policing
- (b) As Ravi, discuss the alternatives available
- (b) Justify the action that best complies with ethics
Open with · Algorithms trained on skewed historical data can turn past discrimination into future 'predictions' — a feedback loop of over-policing.
Cover
- Bias: biased historical data creates a self-fulfilling loop against an immigrant, low-income area (Articles 14, 15); NITI's Responsible AI warns against deepening historic divisions.
- Privacy: mass biometric capture without clear legal basis fails Puttaswamy's legality, necessity and proportionality; the EU AI Act bans real-time public facial recognition, barring exceptions.
- Due process and transparency: residents do not know the data held against them; opaque algorithms drive preventive detentions.
- Accountability: who answers for wrong matches; officers' over-reliance on machine output; data security.
- Options: continue as is (order gains, rights and trust lost); scrap it (useful tool lost); suspend high-risk uses, audit and reform.
- Recommended: independent bias audit, cleaned data, human verification before any action, end blanket detentions, data access and grievance redress for residents.
- Also: community policing and dialogue with leaders, compliance with the DPDP Act, 2023, published SOPs and periodic review.
Close with · Technology should sharpen policing, not replace fairness; public order lasts only when the policed trust the police.
Add value (verified)
- MeitY lists bias, discrimination, exclusion and lack of transparency among the risks the AI governance guidelines seek to address. PIB — MeitY on India AI Governance Guidelines (19 December 2025) ↗“Some of these include bias, discrimination, unfair outcomes, exclusion, and lack of transparency.”
- NITI Aayog, Responsible AI (2021), Principle of Inclusivity and Non-discrimination: AI should not deepen historic and social divisions based on religion, race, caste, sex, descent, place of birth or residence. Responsible AI #AIForAll: Approach Document for India, Part 1 — NITI Aayog, February 2021 ↗“Principle of Inclusivity and Non-discrimination: AI systems should not deny opportunity to a qualified person on the basis of their identity. It should not deepen the harmful historic and social divisions based on religion, race, caste, sex, descent, place of birth or residence in matters of education, employment, access to public spaces, etc.”
- EU AI Act: real-time and remote biometric identification, such as facial recognition in public spaces, is among banned AI applications, with limited exceptions for law enforcement. EU AI Act: first regulation on artificial intelligence — European Parliament ↗“Banned AI applications in the EU include: Cognitive behavioural manipulation of people or specific vulnerable groups: for example voice-activated toys that encourage dangerous behaviour in children Social scoring AI: classifying people based on behaviour, socio-economic status or personal characteristics Biometric identification and categorisation of people Real-time and remote biometric identification systems, such as facial recognition in public spaces Some exceptions may be allowed for law enforcement purposes.”
- Kant's Humanity Formula: never treat humanity, in oneself or in others, as a means only but always as an end in itself. Kant's Moral Philosophy — Stanford Encyclopedia of Philosophy ↗“This formulation states that we should never act in such a way that we treat humanity, whether in ourselves or in others, as a means only but always as an end in itself.”
Question: UPSC's CS (Main) 2026, GS Paper IV — paper ↗. Approach: Minimalist IAS, checked 1 Oct 2026 (how we verify) — UPSC publishes no model answers. ·
Model answer · 349 words (UPSC limit 250) · Minimalist IAS
An algorithm trained on past policing records predicts where the police looked before, not where crime is. Ravi's system has bought visible order at a hidden cost in fairness, privacy and trust.
Stakeholders
- Residents of the neighbourhood; victims of gang crime; Ravi and his force; community leaders and rights groups; the state government; the vendor.
(a) Ethical issues and biases in data-driven policing
- Historical bias: skewed data marks one community as suspect; more patrols yield more records, and the loop confirms itself, offending Articles 14 and 15. NITI Aayog's Responsible AI principles say AI "should not deepen the harmful historic and social divisions".
- Privacy: mass biometric capture of crowds without a clear law fails the Puttaswamy tests of legality, legitimate aim and proportionality, and chills ordinary life. The EU's AI Act bans real-time facial recognition in public spaces, with narrow exceptions for law enforcement.
- Due process: detentions and checkposts driven by opaque scores; residents cannot see or contest data against them.
- Accountability: a false match is a machine error with a human cost; officers defer to the screen. MeitY itself lists bias, discrimination, exclusion and opacity among AI's risks.
- Ends and means: better crime figures do not justify treating a community as a suspect class.
(b) Alternatives before Ravi
| Option | Gain | Cost |
|---|---|---|
| Continue as is | Visible order | Discrimination, lost trust |
| Scrap the system | Harm ends | A useful tool lost |
| Pause, audit, reform | Order with fairness | Short-term slowdown |
The action that best complies with ethics
- Suspend high-risk uses: no detention or search on AI output alone; human verification and recorded reasons first.
- Independent bias audit: of data and algorithm; retrain on cleaned data; publish SOPs.
- Due process: residents may see and correct data held on them through a grievance officer, in line with the DPDP Act, 2023.
- Community policing: standing dialogue with the leaders who came to him, joint beat meetings and civic services in the area.
- Why: it keeps the legitimate aim of preventing riots while honouring equality, privacy and fairness.
Technology should sharpen policing, not decide it; order lasts only when the policed believe the police are fair.
Written by Minimalist IAS from facts checked at source (how we verify) — a little fuller than exam length, so every part of the question is covered; in the hall, keep the structure and trim the detail. UPSC publishes no model answers: compare your structure and coverage with this, then write your own.