Storybooks › Jugnupur Tales, Volume 1 › Chapter 2

The mango fair: bunting, stalls full of mangoes, and Mitthu standing proudly on the judge's stand beside Dadi, Happy and Bunny.Illustrated with AI

Jugnupur Tales · Volume 1 · Chapter 2

The Judge Who Never Crossed the River

Big idea: AI is only as fair as its examplesClasses 5–7

Part 1 · The Mango Mela

Every year, when the mangoes are at their sweetest, Jugnupur holds its Aam Mela, the mango fair. Farmers come from villages near and far. There are bright flags, sugarcane juice, a giant jhoola, and mangoes piled so high that even the air smells sweet.

This year there was something new. The story of the mango mystery had spread all over the village, so the Mela committee made an announcement: “Our judge for ripe mangoes this year will be… Mitthu the parrot!”

Mitthu sat on a wooden stand marked JUDGE, his feathers fluffed up, looking very important indeed. Bunny and Dadi stood close by, feeling proud.

Farmer after farmer brought their baskets. Mitthu sniffed near the stems, pressed gently, and looked at colour and shape. “Pakka! Kachcha! Pakka!” He was quick, and he was right. The crowd cheered every time.

Part 2 · The basket from the hills

Bhola kaka, in an orange turban, stands sadly behind his basket of red mangoes while Mitthu squawks from the judge's stand and Bunny looks worried.Illustrated with AI

Mitthu had never seen a mango like this before. He guessed anyway.

In the afternoon, an old farmer walked slowly up to the stand. He wore a bright orange pagdi, with a red gamchha over his shoulder, and he had been walking since sunrise.

“Namaste. I am Bhola, from Pahadpur, up in the hills across the river,” he said, setting down his basket. “These are our Lal Moti mangoes.”

The mangoes were small, round and deep red. Mitthu had never seen anything like them. He pressed one. Firm. He sniffed it. Not the sweet smell he knew.

“Raw!” he announced. “All raw! Next!”

Bhola kaka’s face fell. “But I picked them this morning, at exactly the right time,” he said quietly. “In Pahadpur, everyone knows that Lal Moti stays firm even when it is ripe.”

But the judge had spoken. The committee put his basket aside. Without a “ripe” mark, no shopkeeper would buy his mangoes. Bhola kaka sat down under a tree, his long walk wasted.

Bunny frowned. Something did not feel right.

Part 3 · Bunny counts

That evening, Bunny opened his notebook. In it he had drawn every one of the fifty mangoes Mitthu practised with.

He counted. Dasheri, Langra, Totapuri, Chausa… every single mango came from Dadi’s orchard. From Jugnupur: fifty. From the hills: zero.

“Mitthu,” he said slowly, “you didn’t get Bhola kaka’s mangoes wrong because you are a bad judge. You got them wrong because you had never seen one.”

Mitthu hung his head. “My clues only work for the mangoes I learned from. For hill mangoes, my clues were wrong.”

“And it wasn’t fair,” said Bunny. “Bhola kaka lost his whole day, and maybe his money, because of what was missing from your practice pile.”

Dadi put down her cup of tea. “A judge who learns from only one kind will be unfair to every other kind,” she said. “It is not because he has a bad heart. It is because of the examples he never saw. But a mistake like this must still be put right.”

Part 4 · Across the river

Bhola kaka rows Bunny and Mitthu across the river towards the green terraced hills of Pahadpur.Illustrated with AI

Early next morning, Bunny, Mitthu and Dadi found Bhola kaka at the ghat, getting his boat ready to go home. “Kaka,” said Bunny, “will you teach Mitthu about your mangoes?”

For the first time since the Mela, Bhola kaka smiled. They climbed into his little boat and crossed the river to Pahadpur.

There, among the green terraced hills, the farmers showed Mitthu forty Lal Moti mangoes: ripe ones and raw ones, big ones and small ones. Mitthu learned new clues. A ripe Lal Moti is red all over, with no green patch near the stem. Its skin is shiny. And it has a light, flowery smell, very different from Dadi’s mangoes.

Then Mitthu practised, round after round, just as he had with Bunny before. Guess. Check. Adjust. Again.

Part 5 · Judging again

Back at the Mela, the committee agreed to judge Bhola kaka’s basket again. This time, Dadi and two Pahadpur farmers stood beside Mitthu to check his answers.

Bunny also had a new idea. “Let’s test Mitthu on each kind of mango separately,” he said. “Otherwise a good score on Jugnupur mangoes could hide a bad score on hill mangoes.”

On Jugnupur mangoes, Mitthu still got nine out of ten right. On Pahadpur mangoes, he had got only three right at the Mela. Now he got eight.

“Pakka. Pakka. Pakka.” Most of Bhola kaka’s basket got its ripe mark, and a shopkeeper bought every mango by evening. And when the Mela ended, Lal Moti won the prize for the sweetest mango of the year.

“I am sorry, kaka,” said Mitthu. “Next time I meet something new, I won’t pretend. I’ll say, ‘I have not seen this kind before.’”

Part 6 · A call from Bengaluru

At night, Bunny talks to Kabir chachu on a video call, with a small smart speaker on the table.Illustrated with AI

That night, Bunny told Kabir chachu everything.

“Bunny, what happened to Bhola kaka happens with AI too,” he said. “Engineers call it bias. An AI learns only from its examples. If most of its examples come from one kind of people, one place, or one way of speaking, it can work well for them and badly for everyone else.”

“Like what?”

“Like a voice assistant that learned mostly from one accent. It may find it harder to understand people who speak differently, even when they are saying exactly the same words.”

“That’s not fair!”

“No, it isn’t. So good engineers do what you did. They count what is in the data. They look for who is missing. They collect examples from them. They test the AI on each group separately. And for important decisions, they make sure a person checks the answer.”

After the call, Bunny looked out at the fireflies over the orchard. Some blinked fast, some slow, some bright, some soft. A fair judge, he thought, has to learn from all of them.

Moral of the story

A judge who has seen only one kind will be unfair to every other kind. Before you judge, ask yourself: have I seen enough of this kind?

Who learns from just one side of the stream
sees only half of what things seem.
So cross the river, meet the rest:
to learn from all is to learn best.

Mitthu’s notebook

What really happened?

Here is Chapter 2, translated into the language of AI.

In the storyIn AI, we call itWhat it means
Mitthu’s 50 practice mangoes, all from one orchardUnbalanced dataExamples that come mostly from one group, place or kind.
Calling ripe hill mangoes “raw”BiasAn AI that is regularly wrong, or unfair, for one group.
Bhola kaka losing his saleReal-world harmAI mistakes can hurt real people, not just scores.
Bunny counting his notebookChecking the dataLooking at what is in the examples, and what is missing.
Learning from 40 Lal Moti mangoesDiverse dataAdding examples from every group the AI will meet.
Testing each kind separatelyFairness testingChecking results for every group, so one good score can’t hide a bad one.
Dadi and the farmers checking answersHuman reviewPeople checking an AI’s important decisions.
“I have not seen this kind before”Knowing its limitsA good AI should warn you when something is new to it.

Did you know?

Fun facts

A famous studyIn 2018, researchers Joy Buolamwini and Timnit Gebru tested face-analysis programs from big companies. The programs made far more mistakes on darker-skinned women than on lighter-skinned men. After the study, the companies worked to improve them.
So many ways to speakIndia’s Constitution lists 22 official languages, and hundreds more are spoken across the country. The Government of India’s Bhashini project is building AI tools for Indian languages, so more people can use technology in their own language.
An old sayingComputer scientists have a saying from the early days of computing: “Garbage in, garbage out.” If the examples going in are poor or one-sided, the answers coming out will be too.
Same fruit, different placeThe soil, rain and weather where a mango grows can change its colour, size and taste. That is why examples from many places matter.

Try this · no computer needed

Draw a Shoe

You need: 5 or more friends, paper and pencils. Time: 20 minutes.

  1. Everyone gets one minute to draw “a shoe”. Don’t look at each other’s drawings!
  2. Put all the drawings together. This pile is your training data.
  3. Count the kinds: how many sports shoes, school shoes, chappals, juttis, Kolhapuris, sandals, gumboots?
  4. Imagine a machine that learned “shoe” only from your pile. Which kinds of shoes might it fail to recognise?
  5. Who would be treated unfairly if that machine decided who was “wearing shoes” at a school gate?
  6. Now fix it: draw the missing kinds and add them to the pile. Count again. Is it more balanced now?

Quiz time

How well did you learn?

Score: 0 / 6

Think and discuss

No wrong answers here

  1. Was it only Mitthu’s fault? Who else could have noticed the problem before the Mela?
  2. If a learning app learned only from students in big cities, whom might it misunderstand?
  3. Think of one thing about you, like your language, handwriting or festivals, that an AI might not have seen enough examples of.
  4. Should a machine alone decide important things, like who wins a prize or who gets a scholarship? Why should a person check?

Be curious

Ask a teacher, parent or engineer

Who decides which examples an AI learns from? How do engineers check whether an AI is fair? Does the AI on our phone understand our language? What should I do if an app gets something wrong about me? Can bias ever be removed completely?

Word bank

New words from this chapter

Bias: when an AI is regularly wrong or unfair for one group.
Dataset: the whole collection of examples an AI learns from.
Balanced data: examples that fairly cover every group.
Fairness testing: checking an AI’s results for each group separately.
Human review: a person checking an AI’s important decisions.
Aam Mela: a mango fair.
Ghat: steps leading down to a river.
Pagdi / Gamchha: a turban / a thin cotton towel worn over the shoulder.

For teachers and parents: this chapter fits two 35-minute periods. Period 1: read Parts 1–6 aloud (15 min), go through Mitthu’s Notebook and the balance-scale picture (10 min), do the quiz (10 min). Period 2: “Draw a Shoe” and the Think and Discuss questions. The key idea to leave with children is that AI reflects the examples it learns from, so missing groups lead to unfair results. People can find and fix this by checking the data, adding what is missing, and testing for each group. The story illustrations marked “Illustrated with AI” were made with an AI image generator; the diagrams, charts and labelled scenes were drawn by hand.