Mains

Open on the mismatch — a fairness principle on one side, 82% of working women in informal employment on the other. Then run the two sides honestly: principle-based fairness works where developers are few, visible and auditable, which is the logic of the EU AI Act and of India's own data-protection statute; it fails where informality dominates and nobody is checking. Bring Gender Responsive Budgeting in as the counter-instrument with its four questions — which women, which barrier, which metric, which budget line pays for corrections. Use AI literacy riding DAY-NRLM, DDU-GKY and Skill India through the Sakhi network as proof that delivery beats declaration, and Farmer.Chat as the evidence. Verdict must resolve, not hedge: principles set direction, budgets and procurement convert it into delivery.

This is a three-part answer and the marks follow the structure, so give each part its own block. Potential: run it through three concrete channels — agricultural advisory, credit access for women without collateral, and regional-language welfare interfaces — and anchor each in something real rather than asserting promise. Risks: algorithmic bias from male-dominated training data, the device-and-connectivity divide, technology-facilitated gender-based violence and its chilling effect, and opacity that hurts marginalised users most because they have the least capacity to contest. Governance: three operational moves — disaggregated gender impact assessments, AI literacy riding existing missions, and enforceable synthetic-content labelling with regional-language grievance. Close on the conditional: AI neither includes nor excludes automatically; procurement decides.

At 10 marks there is no room for a way-forward section, so the whole answer must be argument. Establish the old divide in one line — devices, connectivity, cost — then move the frontier: even where access exists, systems built without women's languages, mobility patterns and content needs do not reach them. The Farmer.Chat evidence is the hinge, because disproportionately high engagement proves the bottleneck was design and not demand. Close with the corollary that matters for governance: a design divide is closed by procurement standards and testing protocols, not by connectivity spending.

Decompose by ethical dimension rather than narrating examples, and name the asymmetry early — efficiency accrues to the state while exclusion falls on the vulnerable, which is what makes this an ethics question and not an efficiency one. Take four dimensions in turn: fairness threatened by historical bias in training data; explainability, which matters most to those least able to challenge a decision; privacy, which cuts differently for people with thinner legal recourse; and accountability, which decides whether harm can be corrected at all. At 10 marks a formal verdict is not required, but end on a usable test — can the person most affected understand the decision, challenge it, and be heard?

Name the stakeholders before the issues — the women denied credit, the SHG federations and Sakhi network as their collective voice, the district administration, the vendor, the Central ministry and the public. Then the four issues: algorithmic fairness and whether a 'neutral' model is reproducing historical exclusion; transparency, since a vendor's neutrality claim is unfalsifiable without disclosure; accountability, because no line of responsibility has been fixed; and justice, since those least able to challenge the system bear its cost. On options, weigh continuing, suspending the tool entirely, and the middle path of pausing automatic rejection while routing cases to human review. Choose the middle path — it protects applicants without halting disbursement — and close on procurement reform, because the real failure happened before the tool was ever deployed.

The directive is 'critically examine', so both sides must be argued and then resolved — an answer that only lists threats will not score. Open on the collision: the world's largest young population meeting a technology whose value proposition is more output with less human input. Opportunity side: productivity gains in agriculture advisory, healthcare reach and personalised learning, plus genuinely new job categories, backed by the national AI mission. Threat side: India's exposure is concentrated in exactly its middle-class employment base — business process work, IT services, back office, entry-level legal and accounting — and it lands on an employment elasticity already near 0.1. Then the framework in three layers: transition financing, employer obligation on mass separation, and sector-differentiated adoption. Close on the shelf life of the dividend, because that is what makes this urgent rather than merely important.

This is the 2024 Labour Codes question in disguise, so structure it as a genuine two-column evaluation and resist picking a side early. Establish the baseline first — twenty-nine statutes, many pre-Independence, whose complexity itself pushed employers towards informal hiring and left workers with no protection at all. Column one, business: single registration, retrenchment threshold raised from 100 to 300, fixed-term employment legalised. Column two, workers: social security extended to gig and platform workers for the first time, universal minimum wages, mandatory appointment letters, enforceable timely payment. Then the honest gaps: tighter strike conditions, misuse potential in fixed-term contracts, extended working-hour limits, and dependence on states framing rules. Conclude that balance is a process requiring monitoring and correction, not a one-time achievement.

Define the term in one line and give the trajectory, because the numbers are the argument: roughly 0.44 in the 1980s, about 0.19 in the 2000s, near 0.1 recently. Then establish that it is structural not cyclical — a cyclical fall reverses with the cycle, a fall sustained across three decades and multiple governments does not. Four structural reasons: the skipped manufacturing stage and the jump from agriculture to services, capital-intensity within manufacturing itself, MSME distress after successive shocks, and women's participation still below comparable economies. The growth model should follow directly from the four causes rather than being a generic list — labour-intensive manufacturing, MSME revival, a women's employment mission, and the rural non-farm economy.

The trap here is writing an ideological answer. Give each position its strongest form before criticising it. Growth-first: a small economy cannot fund large redistribution, so enlarge the base first — and this is not a fringe claim, it describes what South Korea, Singapore and China actually did. People-first: concentrated income hits a consumption ceiling, so mass consumption requires mass income, a point Adam Smith and Keynes reach independently. Then the failure mode of each: growth-first assumes jobs are an automatic by-product of output, which India's falling elasticity disproves; people-first taken too far suppresses the investment that funds everything. Resolve into growth-plus-equity and make the resolution concrete — investment-friendly reform paired with safety nets, employer accountability, and targeted missions.

At 10 marks, split the answer cleanly in two — lessons, then limits — and give the limits real weight, because the examiner is testing whether you can admire a policy without importing its politics. Lessons: treat the AI transition as a policy problem rather than a market outcome, place a reskilling obligation on employers who displace, and give workers and small entrepreneurs a formal seat alongside investors. Limits: India cannot replicate courts and directives that act on political instruction, change here must come through consultation and law, and China's own model is visibly under strain with high youth unemployment, a property crisis and an ageing population. Close on the distinction that carries the marks — the transferable thing is the principle, not the method.