AI in Public Systems: Start With the Boring Problems

Spend a day in a district office and you will see where time goes. Officers read and sort hundreds of letters and complaints. Staff copy the same data from one format into another. Frontline workers call supervisors to ask what a guideline says about an unusual case. Families come back again and again because they did not understand what document they needed.

None of this makes headlines. All of it decides whether a scheme reaches people. And much of it is exactly the kind of work that today's AI tools are good at.

Where AI in governance can help now

1. Explaining schemes in plain language

Government guidelines are long, technical and usually written in formal Hindi or English. AI tools can turn them into short, clear explanations: who is eligible, which documents are needed and where to apply. With speech tools, the same explanation can be read aloud for people who cannot read the text. A family that understands what to bring makes one trip instead of three.

2. Making sense of grievances

Grievance systems receive large numbers of complaints, written in many styles and languages. AI can help sort them by topic, flag urgent cases and show patterns, for example that most complaints from one block are about delayed pensions. That turns a pile of individual complaints into evidence for fixing a process.

3. Supporting frontline workers

An Anganwadi worker or a block-level official often needs a quick answer to a practical question. Is this child eligible? What should I do if the certificate has expired? A well-built assistant, drawing only on official guidelines, can give an answer in seconds and show where in the guideline it comes from.

4. Cleaning and joining data

District data sits in many separate systems that do not talk to each other. AI tools can help match records, spot errors and prepare simple summaries, so that officers spend less time compiling data and more time using it.

Where AI can do harm

The same features that make AI useful also make it risky in public services.

• Confident mistakes. AI language tools can produce answers that sound right but are wrong. In public services, a wrong answer about eligibility can mean a family gives up on a benefit it is entitled to. Any tool that explains entitlements must be grounded in official sources, show those sources, and be checked regularly.
• Privacy. Public systems hold sensitive data about health, disability, income and family. The Digital Personal Data Protection Act, 2023 sets duties for how personal data is handled. Feeding such data into tools without clear safeguards puts people at risk.
• Automating exclusion. If AI is used to approve, reject or flag applications, it will learn from past decisions, including past mistakes and biases. People who were wrongly excluded before can be excluded again, faster.
• Language gaps. Many people in Rajasthan speak forms of Rajasthani at home, which is not one of India's 22 scheduled languages. Tools that only work well in standard Hindi or English will leave some of the people who most need help behind.
• Inaccessible design. A chatbot that does not work with a screen reader, or a voice line that does not understand atypical speech, shuts out persons with disabilities.

How to start well

The safest way to bring AI into public systems is to start small, keep people in charge and measure what happens.

1. Pick one real problem. Start with a specific, frequent pain point, such as explaining one scheme or sorting one type of complaint, and not a general "AI for the district" project.
2. Keep a human in the loop. AI should help officials and frontline workers decide. It should not decide for them, especially where a person's rights or benefits are at stake.
3. Test with the people it is meant to serve. Try the tool with real users, including persons with disabilities and people with limited literacy, before rolling it out.
4. Measure outcomes. Track whether the tool actually saves time, reduces errors or improves access, and for whom.
5. Be open about it. Tell people when they are dealing with an AI system, what it can and cannot do, and how to reach a human.

The opportunity

India is investing seriously in AI for public good. The IndiaAI Mission, approved in 2024, and the Bhashini language platform both aim to make AI tools available for public services in Indian languages. The opportunity is real. The question is whether it will be used to fix the everyday problems that block access, or spent on showcase projects that do little for the person waiting outside the office.

At Sutra, we are interested in AI as a practical tool for public systems in Rajasthan, used carefully, tested with real users and judged by whether more people get what they are entitled to. If you are a government team, a technologist or a non-profit exploring AI for public services, we would like to compare notes. Write to us at info@sutraindia.org.