By Sankalp Shukla, Founder, Development Leaders’ Alliance
From a QR code at a tea stall to a farmer’s unanswered query, the real test of technology is not just speed. It is whether someone remains accountable when things go wrong.
You scan a QR code at a roadside tea stall. Your screen says “payment successful”, but the shopkeeper’s phone has not buzzed. The transaction took two seconds; sorting out the confusion may take much longer. At that moment, you are not impressed by the sophistication of the technology. You simply want to know where your money went, and who will help.
This is not an artificial intelligence story. That is exactly why it matters. India’s UPI processed 24.07 billion transactions in September 2026, according to the National Payments Corporation of India.
Behind that everyday convenience sit standards, banks, regulation and a system in which errors can be raised and addressed. The lesson for AI is straightforward: scale may come from automation, but trust comes from accountability.
We are being told that AI will write our emails, answer our calls, prepare our reports and perhaps even make decisions for us. Much of that promise is real. Yet there is a world of difference between a machine helping someone make a decision and a machine having the final word on someone’s life. The companies and public institutions that understand this distinction may gain the most from AI.
When an answer is easy, but justice is not
Consider a farmer who wants to know why a PM-KISAN instalment has not reached the family account. Waiting in a queue or travelling to an office for a routine status update should not be necessary. The government’s Kisan e-Mitra AI chatbot already answers scheme-related questions in 11 languages, making information more accessible. That is the kind of automation worth welcoming. But suppose the underlying record has an incorrect name or a mismatched detail. A chatbot can explain the status; it should not be the last door the farmer can knock on. Someone must have the authority and the obligation to put the record right.
The same principle applies to a district hospital. AI may help flag an unusual scan or summarise a medical history before a consultation. It cannot replace a clinician’s duty to hear the patient, weigh symptoms that do not fit neatly into a template, and explain a difficult choice. In a classroom, software can identify a child who is falling behind in reading. A teacher may also know that the child missed weeks of school because the family moved for seasonal work. Data identifies a pattern; people understand its context.
And consider a loan applicant whose income arrives through gig work rather than a fixed monthly salary. A model trained primarily on conventional employment records may treat irregular deposits as a warning sign. That can help a bank ask better questions; it should not automatically turn an otherwise creditworthy person into a rejected application with no explanation and no appeal. Efficiency must not become a polite name for exclusion.
The everyday frustration behind the AI hype
Ask anyone who has spent twenty minutes arguing with an automated customer-care bot. The first instant reply feels helpful. By the fifth identical reply, “We understand your concern” begins to sound like an insult. If an online order is damaged, the customer wants a refund or replacement, not a beautifully worded paragraph. For an e-commerce company, the useful measure of AI is not how many conversations the bot handles. It is how many problems actually get solved—and how quickly a human can step in when the bot cannot.
There is an opportunity here for India’s smaller businesses, too. A neighbourhood retailer can use AI to catalogue products in Hindi, draft messages for customers and keep track of inventory. A small manufacturer can use it to translate an export enquiry or organise invoices. These are practical gains, especially for enterprises that cannot hire specialised teams. But no software should be mistaken for the business owner’s judgment about a trusted customer, an employee’s circumstances or a long-standing supplier relationship.
The temptation, in boardrooms as much as government departments, is to count what is easy to count: calls deflected, files processed, hours saved, chatbots launched. The better questions are harder. Did a citizen need fewer visits to an office? Was a customer’s complaint resolved the first time? Did the farmer get a usable answer in a familiar language? Did a worker become more productive, or simply easier to replace?
India can build a more human AI economy
India is right to invest in the foundations of AI. The IndiaAI Mission, approved with an outlay of ₹10,371.92 crore, is intended to widen access to computing infrastructure, datasets, skills and responsible AI tools. Government reporting in March 2026 said more than 38,000 GPUs had been onboarded for shared compute access. Making this capacity available beyond a few large technology companies matters. It gives startups, researchers and public institutions a chance to build solutions for Indian realities rather than merely import them.
But computing power is only the beginning. Poor records will not become accurate because an algorithm reads them faster. A service that is confusing offline can remain confusing on an app. And a system that cannot explain a decision will not inspire confidence simply because it speaks in fluent Hindi, Tamil or English. Frontline staff need training to question AI outputs. Citizens need a clear route to correct mistakes. Organisations need to test tools in real conditions, including on people with different languages, incomes and levels of digital access.
Before any institution automates a consequential decision, it should be able to answer three questions: Who checks the output? Who can challenge it? And who takes responsibility when it is wrong? If the answers are vague, the technology is not ready for that responsibility, however impressive the demonstration looks.
India does not have to choose between ambition and caution. We can automate the queue without automating away the right to be heard. We can use AI to help a doctor, teacher, farmer, shopkeeper or public servant do the job better without pretending that all of their judgment can be reduced to a prediction. The real AI advantage will not belong to those who remove the most humans from the process. It will belong to those who use machines to make human lives easier, and keep humans answerable when it matters most.