An 11-Year-Old Girl Died Because a Machine Couldn’t Read Her Mother’s Fingerprints

Part of “The AI You Don’t See” series by Akshay A. Walimbe

Santoshi Kumari was eleven years old. She lived in Karimati village, Simdega district, Jharkhand  one of the poorest regions in one of India’s poorest states. In September 2017, her family and independent fact-finding teams say she died of starvation. District officials later attributed her death to malaria, though no medical report was produced to support that claim.

Not because there was no food. Not because the government had no scheme to feed her. There was a scheme. There was food. The ration shop had grain sitting on its shelves.

Santoshi died because a machine said no.

Here is what happened, step by step. I want you to follow the chain because at no point will you find a human being who made the decision to let this child starve.

In February 2017, the Indian government made Aadhaar linkage compulsory for accessing the Public Distribution System the network of ration shops that provides subsidised food to hundreds of millions of Indians. To collect your monthly ration of rice and wheat, you now had to scan your fingerprint on a biometric machine at the ration shop. The machine would check your fingerprint against the Aadhaar database. If it matched, you got food. If it did not match, you did not.

Santoshi’s family’s ration card was cancelled because it was not linked to their Aadhaar number. The local ration dealer refused to give them food. For six months.

Six months without rations. In a district where the next meal is not a given.

Santoshi did not survive those six months.

Now, you might think this was a one-off failure. An edge case. A tragic exception in an otherwise well-functioning system.

It was not.

The Right to Food Campaign investigated 57 starvation deaths between 2015 and 2018. They found at least 19 were directly tied to Aadhaar problems. In the twelve months after Santoshi’s death alone, at least 37 more starvation deaths were recorded. Thirteen of those were linked to Aadhaar authentication failures.

Ruplal Marandi, an elderly Adivasi man from Deoghar, Jharkhand reported as between 60 and 64 years old across different sources  died in October 2017 after his family was denied rations for two months because biometric authentication failed. Two children Munni, age 2, and Govinda, age 5 died of hunger in Buxar, Bihar in 2018 after their family could not get an Aadhaar card made and was denied rations for eight months.

These are not statistics. These were people. Children. Elderly men. Families who had every right to eat and were turned away by a fingerprint scanner.

Let me tell you why the machine kept saying no.

Biometric authentication  the fancy term for “scan your finger and we will verify you are who you say you are” has a failure rate. That failure rate is not evenly distributed. According to a World Privacy Forum study published in the journal Health and Technology, the biometric failure to match rate in Jharkhand was 49 per cent. Nearly half. In Rajasthan, it was 37 per cent. Compare that to states like Andhra Pradesh, where the statewide average was closer to 2.5 per cent according to an Indian School of Business study.

See the pattern? The poorest states had the highest failure rates. The states where people needed rations the most were the states where the machine failed the most.

And there is a reason for that. Fingerprints wear out. If you spend your life laying bricks, mixing cement, handling salt, working with chemicals, your fingerprint ridges erode. Manual labourers  the very people the Public Distribution System exists to feed  are the people whose fingerprints the machine cannot read.

The elderly are next. Fingerprint ridges fade with age. Dry hands cause scan failures. Eye conditions affect iris scans.

Researchers Jean Dreze, Nazar Khalid, Reetika Khera, and Anmol Somanchi studied the Aadhaar-Based Biometric Authentication system in Jharkhand. Their conclusion was devastating: the system caused “pain without gain?” It did almost nothing to reduce the leakages it was supposed to fix. But it created massive exclusion, disproportionately hitting widows, the elderly, and manual labourers. At least half of the households they surveyed experienced a problem with the point of sale machine.

Pain without gain. That was the academic phrasing. The human phrasing is: a system designed to prevent fraud ended up preventing the poorest Indians from eating.

But here is the part that haunts me.

When biometric authentication fails, there is supposed to be a backup. The prescribed exception handling mechanism is an OTP  a one-time password sent to your mobile phone. You type in the OTP, and you get your rations without a fingerprint scan.

Think about that for a moment. The backup system for when the fingerprint scanner fails is a text message to your phone. In rural Jharkhand. Where the elderly often do not own mobile phones. Where network connectivity is unreliable at best. Where 500 million Indians in rural areas are still offline.

The exception handling mechanism was itself exclusionary. The safety net had a hole in it, and the hole was shaped exactly like the people who needed the net most.

According to research by Muralidharan, Niehaus, and Sukhtankar published in The Review of Economics and Statistics, approximately 23 per cent of beneficiaries in Jharkhand had not linked their Aadhaar at baseline, and a significant number suffered reductions in benefits. Across Andhra Pradesh, Rajasthan, and West Bengal combined, the IDinsight “State of Aadhaar” report found that roughly 2 million people were denied PDS rations every month due to Aadhaar-related factors.

Now I want you to ask yourself a question. Not a philosophical one. A very practical one.

Who was responsible for Santoshi Kumari’s death?

Was it the ration shop dealer who refused to give her family food? He was following the rules. The system told him the family was not verified. He did what the system said.

Was it the Aadhaar system? A system does not have intent. It does not decide to starve a child. It simply processes inputs and returns outputs. Input: fingerprint. Output: match or no match. No match. No food.

Was it the policymakers who made Aadhaar linkage compulsory? They were trying to prevent fraud and eliminate ghost beneficiaries, which are legitimate goals. The Aadhaar system has saved the government thousands of crores by plugging leakages. Even the researchers most critical of its implementation acknowledge that. But the Supreme Court had issued multiple interim orders since 2013 directing that no person should be denied services for lack of Aadhaar. The government proceeded with the mandate regardless, arguing that the Aadhaar Act of 2016 provided the legal basis. But policymakers did not design the biometric system. They did not set the failure threshold.

Was it the technology designers who built the biometric authentication system? They built a system that works perfectly in controlled environments. Clean fingers, good scanners, reliable connectivity. They may never have tested it on the hands of a woman who has spent thirty years washing clothes by hand.

Was it the Public Accounts Committee of Parliament, which heard evidence about high biometric failure rates but could not prevent the deaths that were already happening?

This is the accountability gap. Not a single person in this chain made a conscious decision to deny food to an eleven-year-old girl. The ration dealer followed the system. The system followed its algorithm. The algorithm followed its design parameters. The designers followed their brief. The policymakers followed their policy goals.

Everyone did what they were supposed to do. And a child died.

When we talk about AI and technology making decisions about people’s lives, we like to talk about it in abstract terms. Algorithmic bias. System failures. Edge cases. Error rates.

But every error rate is a person. A 49 per cent failure-to-match rate in Jharkhand means that roughly every other person who walks up to that fingerprint scanner gets turned away. Some of them walk away hungry. Some of them walk away and do not come back.

The machine does not know this. The machine processes the next fingerprint and the next. It has no concept of the eleven-year-old girl who is not eating tonight.

And that is precisely the problem. The machine has no concept of consequences. But the consequences are real. The consequences are measured in lives.

Who was accountable for Santoshi Kumari?

The honest answer  the uncomfortable answer is that in the current system, nobody was. The technology made a decision. The decision killed a child. And the chain of responsibility dissolved into procedures and protocols and system specifications until there was nobody left to hold accountable.

To be fair, the government has since introduced fallback mechanisms  iris scanning, manual overrides, extended deadlines for Aadhaar linkage. The Supreme Court’s 2018 Puttaswamy judgment upheld Aadhaar for welfare delivery while striking down its mandatory use for bank accounts and mobile phones. Things have improved. But improved is not the same as fixed. And the people who died between 2017 and 2018 did not get the benefit of those improvements.

This is not an edge case. This is what happens when systems designed in air-conditioned offices are deployed on the bodies and fingerprints of the most vulnerable people in the country, with no meaningful accountability for what happens when they fail.

Who was accountable?

I’m have written a book about exactly this  how AI and automated systems make decisions about your life, where accountability disappears, and what we can do about it. If you want to know morea about this book or order a copy, you can do it here: https://akshaywalimbe.com/beyond-bias/