Skip to Content
Join the Network with Us — Join Membership


AI in Official Statistics: Why India Must Balance VISTAR’s Technology with Ground-Level Evidence

November 18, 2026

India’s efforts to modernise its official statistical system through Artificial Intelligence (AI) and Machine Learning (ML) are opening new possibilities for faster data processing and more detailed economic analysis. However, experts warn that technology-generated estimates cannot automatically replace information collected through surveys and direct field observations.

The newly approved Viksit Bharat Initiative for Statistical Transformation, Analytics & Research (VISTAR) places AI and ML at the centre of an overhaul of India’s official data architecture. The initiative aims to help the Ministry of Statistics and Programme Implementation (MoSPI) process, clean and combine large volumes of alternative digital data.

The central challenge is ensuring that faster and more comprehensive statistical estimates remain representative of actual economic and social conditions. A sophisticated model can process millions of records quickly, but that does not necessarily mean the resulting figures accurately reflect what is happening on the ground.

VISTAR Seeks to Address Gaps in India's Statistical System

One of the major problems facing India's statistical infrastructure is the existence of data silos, where information held by different government departments and institutions is not adequately connected.

AI and ML could help integrate these datasets, identify patterns and produce more frequent and geographically detailed statistics for both central and state governments.

However, the concern is that technology may be used to fill gaps in official information without addressing the reasons those gaps exist in the first place.

VISTAR also envisages an expansion in major surveys during the 16th Finance Commission cycle, indicating that direct data collection will continue to play an important role in the statistical system.

What Is Synthetic Data and Why Does It Matter?

Synthetic data refers to information artificially generated through statistical algorithms, mathematical models or AI techniques. It is designed to reproduce patterns and relationships found in real datasets without necessarily representing actual individual observations.

For example, if survey information about household consumption or employment is available for a limited number of locations, statistical models can use those observations to generate additional simulated records.

This approach can be useful for research, privacy protection, software testing and situations where real observations are limited. Generative Adversarial Networks (GANs), for instance, can learn patterns from existing datasets and produce artificial records resembling the original information.

However, synthetic data also carries an important limitation: it can reproduce the weaknesses and biases present in the information used to train the model.

How Biased Data Can Distort Economic Indicators

The risks become more significant when statistical models are used to generate information for districts, population groups or economic sectors where actual survey observations are insufficient.

A model trained primarily on urban household information, for example, may fail to capture the consumption patterns of informal workers, migrant labourers and economically weaker communities. Similarly, if the original datasets underestimate informal employment, unpaid work performed by women or agricultural distress, synthetic data may reproduce those same gaps.

The concern is not limited to statistical accuracy. Incorrect estimates could influence government planning, welfare allocations and assessments of economic performance.

Consider a district with insufficient employment survey information. A model might generate additional household records using employment, income, age and demographic patterns from neighbouring districts or state-level datasets. Although the resulting figures may appear comprehensive, they may not reflect local unemployment, seasonal migration or the decline of a particular industry.

If such estimates are used to guide programmes such as the Viksit Bharat Guarantee for Rozgar and Ajeevika Mission (Gramin), or direct cash transfers, public resources could be directed towards the wrong areas.

India's Informal Economy Creates an Additional Challenge

A substantial part of India's economic activity takes place outside organised corporate structures, formal accounting systems and regular digital reporting. This makes collecting reliable information about informal businesses and workers particularly difficult.

The Annual Survey of Unincorporated Sector Enterprises (ASUSE) illustrates the scale of the exercise. In 2025, it collected information from more than 6.7 lakh establishments across 24,153 surveyed first-stage units.

The article also highlights concerns that introducing a Merchant Discount Rate (MDR) on Unified Payments Interface (UPI) transactions could encourage some users and businesses to return to cash payments. This could reduce the availability of digital transaction records and make economic measurement more challenging.


Model Collapse and the Risk to Public Trust

Another concern associated with synthetic data is known as model collapse. This can occur when AI models are repeatedly trained on their own generated outputs, potentially reducing the diversity and quality of the information they produce.

The World Economic Forum has also highlighted a related problem called the liar's dividend. As synthetic content becomes more widespread, people may become increasingly sceptical about the authenticity of information, including genuine data.

For official statistics, this creates a serious credibility challenge. Economists, international agencies and state governments may question estimates that cannot be traced back to actual observations or supported by a clear empirical audit trail.

The greatest risk is therefore not simply that an AI system might produce an incorrect number. It is that it could produce a convincing-looking estimate that is accepted as accurate despite lacking sufficient ground-level evidence.


How India Can Build a More Reliable AI-Based Statistical System

The article points towards a balanced approach in which AI strengthens statistical capabilities without replacing direct evidence.

Several safeguards are important:

  • Retain primary surveys: Direct surveys and field-level observations should remain the foundation of official statistics.

  • Introduce human oversight: Statistical experts should review AI-generated estimates and verify important findings against ground-level information.

  • Ensure source transparency: Every published estimate should clearly identify whether it comes from an actual survey, administrative records or a model-generated dataset.

  • Maintain independent verification: Models should be tested against reliable observations before their outputs are used for important policy decisions.

  • Protect statistical independence: Technology adoption should not weaken the credibility and autonomy of official statistical institutions.

The United Nations Economic Commission for Europe (UNECE) 2025 Framework on Responsible AI for Official Statistics offers an international reference point, emphasising fairness, accountability, transparency and validity.

The article also refers to the Gauss-Markov BLUE principle — Best Linear Unbiased Estimator — as a benchmark for reliable statistical estimation under specified assumptions. Although every modern statistical problem may not meet all the conditions required by BLUE, the broader objective should remain the production of unbiased, reliable estimates supported by evidence.


The Wider Debate Around AI Safety

The discussion also connects statistical transformation with the broader debate about AI governance.

A statement warning that mitigating the risk of AI-related extinction should be a global priority alongside other major societal risks was signed by 350 technology and industry leaders. Among the signatories named in the article were Geoffrey Hinton, Yoshua Bengio, Demis Hassabis, Sam Altman, Dario Amodei and Bill Gates.

The statement highlighted concerns about scenarios in which humans could lose control over machine optimisation goals. While this debate extends beyond official statistics, it reinforces the importance of accountability and human oversight when AI systems are used in areas affecting society.

Conclusion: AI Should Strengthen Evidence, Not Replace It

VISTAR presents an opportunity to modernise India's statistical infrastructure, improve data integration and make economic information more timely and detailed. However, technological advancement alone cannot guarantee accurate representation of the country's diverse economic realities.

Synthetic data can support research and fill certain analytical gaps, but it must not be treated as a substitute for actual observations without proper validation.

For organisations adopting AI-driven analytics and digital decision-making, the same principle applies: technology should complement reliable information, not replace the need for verification and sound judgement. Professional Other Services can provide a starting point for exploring relevant business and financial support as organisations adapt to changing technology and data requirements.

Ultimately, the success of India's statistical transformation will depend not only on how quickly data can be processed, but also on whether the resulting information remains transparent, representative and trustworthy.

in News
Share this post
Archive