Why in the News
India’s official statistics are set to move from figures observed in ground surveys towards figures that artificial intelligence (AI) models can estimate from incomplete data. AI and machine learning (ML) are core pillars of the newly approved Viksit Bharat Initiative for Statistical Transformation, Analytics & Research (VISTAR), the plan to overhaul India’s official data system. The risk is that model-made numbers start standing in for evidence collected on the ground.
What is VISTAR, and why does it lean on AI?
- What it is: VISTAR rebuilds how India’s official data is produced. The Ministry of Statistics and Programme Implementation (MoSPI) will use AI to collect, clean and combine large volumes of alternative digital data, digital records rather than survey responses.
- Why it was introduced: The main bottleneck is data silos, records held by separate bodies that are not linked. AI promises faster, more granular statistics for both the Centre and the States.
- More surveys too: VISTAR also proposes more major surveys during the 16th Finance Commission cycle, which strengthens data collected directly from people and firms.
- What the issue is: An estimate, however sophisticated, cannot replace evidence observed on the ground.
- The takeaway: AI can fill data gaps faster, but those gaps come from inefficiencies in the statistical system, and generating data does not fix them.
What is synthetic data, and why is it tempting?
- What it is: Synthetic data is artificially generated by statistical algorithms, mathematical models or AI. It copies the patterns of real datasets without reproducing the original observations, like a realistic sketch rather than a photograph.
- How it is made: Generative Adversarial Networks (GANs), AI models that learn patterns in existing data, can create new records that resemble real ones.
- Legitimate uses: It serves research, privacy protection and algorithm testing where real observations are scarce. The World Economic Forum (WEF) called it an essential tool in 2025.
- Why policymakers are tempted: Demand for district, block and village data is rising. Surveying millions of informal economic units, which sit outside formal accounts and digital reporting, is costly, and AI produces estimates quickly.
What goes wrong when models replace surveys?
- Inherited blind spots: Synthetic data inherits the limits of its training data. If informal employment, women’s unpaid work or farm distress is underestimated in the original, the synthetic version repeats the error.
- Urban bias: A model trained mainly on urban household data can miss how informal workers, migrant labourers and economically weaker sections spend.
- Filled-in districts: A thin district survey can be padded with records modelled on neighbouring districts, not on interviews. Local unemployment, seasonal migration or an industry’s collapse then goes unseen.
- Misdirected welfare: Distorted district data can send money to the wrong places under the Viksit Bharat Guarantee for Rozgar and Ajeevika Mission (Gramin) (VB-G RAM G), the rural jobs and livelihood mission, or direct cash transfers.
- Model collapse and liar’s dividend: The WEF warns of “model collapse”, as models trained on their own synthetic output get steadily worse. A “liar’s dividend” follows when synthetic content makes people doubt even genuine data.
Challenges
- Credible but wrong: The main danger is a model-made picture plausible enough to be accepted as correct, not obviously wrong data.
- No audit trail: Economists, international agencies and State governments may reject numbers that cannot be traced back to actual observations.
- Return to cash: Moves to levy a Merchant Discount Rate (MDR), a fee on digital payments, on the Unified Payments Interface (UPI) may push transactions back into cash, leaving fewer digital records.
Way Forward
- Responsible AI framework: Follow the United Nations Economic Commission for Europe (UNECE) 2025 Framework on Responsible AI for Official Statistics, which stresses fairness, accountability, transparency and validity.
- Human oversight: MoSPI should check model outputs against field data.
- Surveys as the bedrock: Keep primary surveys as the base of official statistics, with models only as supplements.
- Separate labels: Publish model-generated estimates apart from survey-based ones, naming each estimate’s source.
- Estimation standard: Hold every estimate to the Gauss-Markov BLUE (Best Linear Unbiased Estimator) goal: unbiased, reliable and backed by evidence.
Conclusion
AI can make India’s statistics faster, but an estimate is not an observation, and synthetic data repeats the blind spots it learns from. The test for VISTAR is whether it labels model-made numbers clearly and keeps ground surveys at its core.
Key numbers
- Major surveys under VISTAR: from 43 to 59 during the 16th Finance Commission cycle.
- Annual Survey of Unincorporated Sector Enterprises (ASUSE), 2025: more than 6.7 lakh informal establishments covered.
Matching Previous Year Question
“[2015] A decrease in tax to GDP ratio of a country indicates which of the following? (1) Slowing economic growth rate (2) Less equitable distribution of national income Select the correct answer using the code given below. (a) 1 only (b) 2 only (c) Both 1 and 2 (d) Neither 1 nor 2 ANSWER: (a)”