AI for Students · Class 7 · Age 11–12 · Lesson 7 of 12

Why Does AI Sometimes Get It Wrong? 🤔

Understand the four main ways AI makes mistakes — and learn the simple habits that keep you safe, informed, and in control when using AI tools.

📘 Class 7 · Lesson 7 🕐 40–50 min 🧠 Critical thinking 🆓 Free lesson
Illustrated scene: Indian student looking at a tablet with a puzzled expression, question marks floating around, a magnifying glass nearby
Watch first · 2–3 minutes

Class 7 Lesson 7 — Why Does AI Sometimes Get It Wrong?

No sign-in needed · English narration · Safe for all school ages

Story · Arun's History Surprise

The Confident Answer That Was Completely Made Up 😮

Arun, 12, from Pune, was preparing for a History quiz at school. He asked AI: "When did the Battle of Plassey take place and who fought in it?" AI answered confidently: "The Battle of Plassey took place in 1757 between the British East India Company and Siraj ud-Daulah, the Nawab of Bengal."

So far, so good — that answer was correct. Encouraged, Arun asked more questions about lesser-known historical figures. For one question, AI gave a detailed, confident, specific answer — including a name, year, and description of a battle — that Arun's teacher later told him was completely invented. The name did not exist. The battle did not happen.

The AI was not lying. It had no way to know it was wrong. It was doing the only thing it can do: generating the most plausible-sounding text based on patterns — even when it had no real knowledge to draw on.

👉 This lesson explains exactly why this happens — and gives you the tools to catch it before it causes a problem.
Section 1 of 8

🤖 How AI Actually Works (Simple Version)

Before we talk about AI mistakes, it helps to understand what AI is actually doing when it generates text. This is not complicated — and once you understand it, AI errors will make complete sense.

What AI does: An AI language model is trained on billions of sentences from the internet, books, and articles. It learns the patterns of language — which words and ideas tend to appear near each other. When you ask a question, it generates text that fits the pattern of a likely answer. It is not looking anything up. It is not checking a database. It is predicting what a good answer would sound like — based on patterns.

Think of it like this: if someone asked you to finish the sentence "Two plus two equals ___", you would say "four" — not because you calculated it at that moment, but because the pattern is completely familiar. AI works the same way, but with vastly more complex patterns — including many patterns for wrong answers.

The important consequence: AI does not know what it does not know. When you ask about something obscure or unusual, AI may generate a confident-sounding answer that is entirely fabricated — because the pattern "answer this type of question" is strong, but the actual knowledge is absent. This is called a hallucination.
Section 2 of 8

🔍 The Four Main Types of AI Mistakes

Not all AI errors are the same. Understanding the type of error helps you know where to be most careful.

🌀
Hallucination
AI invents facts, names, dates, or events that do not exist — with complete confidence. Most dangerous for specific historical facts, statistics, and people's names.
📅
Outdated information
AI's knowledge has a training cutoff. It does not know about events, laws, discoveries, or current affairs after that date. Always check for recent information.
⚖️
Bias
AI learned from internet text — which reflects human biases. It may give answers that favour certain views, cultures, or perspectives over others, often without realising it.
💬
Overconfidence
AI rarely says "I don't know." It tends to give a confident answer even when uncertain. Confident language does not mean the answer is correct.
Which errors matter most for Class 7 students? Hallucinations are the biggest risk for school projects and quizzes. Outdated information matters for current affairs. Bias matters when researching topics about culture, religion, or society. Overconfidence affects every answer AI gives — the confident tone can mislead you into trusting an incorrect answer.
Section 3 of 8

🌀 Hallucinations — When AI Makes Things Up

A hallucination is when AI confidently states something that is factually incorrect — often an invented person, date, statistic, or event. This is the most alarming type of AI error for students because it looks exactly like a correct answer.

When is hallucination most likely?

✅ Hallucination Check Prompt
You said [state the specific fact AI gave you]. How confident are you in this? Can you tell me the specific source this comes from? If you are not certain, please say so clearly — I would rather know you are uncertain than risk using an incorrect fact in my school project.
Explicitly asking AI to admit uncertainty works — it will often qualify its answer when directly asked. This prompt forces that conversation.
Most at risk: Specific numbers, names of people (especially Indian historical figures who are less represented in AI training data), statistics about India, and any fact that begins with "the first", "the only", or "the largest".
Section 4 of 8

📅 Outdated Information

Every AI model has a training cutoff — the date when it stopped learning new information. An AI trained in early 2024, for example, knows nothing about anything that happened after that date. For most school topics, this does not matter. But for some subjects, it matters a great deal.

Topics where outdated information is a risk:

✅ Currency Check Prompt
I'm asking about [topic]. This is for a Class 7 school project in India in 2026. What is your knowledge cutoff date? Are there parts of your answer that might be outdated? What should I look up in a current source to verify?
Asking about the cutoff date directly often gets AI to flag which parts of its answer you should verify elsewhere.
For current affairs: Never rely on AI alone. Use credible news sources (The Hindu, PIB, or your school's recommended newspaper) for any information about events from the past two years.
Section 5 of 8

⚖️ Bias in AI

AI learned from text written by humans — and humans have biases. AI can reflect those biases without being aware of it. For a student in India, this is especially relevant for topics about Indian culture, history, religion, and society, where training data may be skewed towards non-Indian perspectives.

Common forms of bias to watch for:

✅ Indian Perspective Prompt
You've explained [topic] from a general perspective. Now give me the Indian or South Asian context specifically. Are there aspects of this topic that are viewed differently in India? Are there important Indian contributions or viewpoints that are often missed?
This prompt actively asks AI to check its own blind spots — and often produces much more relevant answers for Indian students.
The good news: You don't need to be afraid of bias — just aware of it. When a topic is about India, always ask AI to give you the Indian angle specifically. And always cross-check with your NCERT textbook, which is written by Indian educators for the Indian curriculum.
Section 6 of 8

💬 Overconfidence — The Most Invisible Problem

Of all AI's error types, overconfidence is the hardest to spot — because it is invisible in the text. AI says "The capital of Maharashtra is Mumbai" with exactly the same confident tone as "The Battle of Khandwa was fought in 1543 between..." — even if the second fact was invented.

Red flag signals — when to be extra careful:

✅ Confidence Level Prompt
For each key fact in your last answer, rate your confidence: - High: this is well-established knowledge likely to be correct - Medium: I believe this is correct but you should verify - Low: I am uncertain — please check before using List each fact with its confidence level.
This prompt forces AI to audit its own output. It is surprisingly effective — AI will often flag the uncertain parts when asked directly.
Section 7 of 8

🛡️ Your Everyday Protection Habits

You cannot verify everything AI tells you — that would take more time than just reading the textbook. But you can build five simple habits that protect you from the most common and damaging errors:

  1. Verify any specific number or statistic — never use an AI statistic in a school project without checking it in a textbook or official source
  2. Names of people = always check — especially for Indian regional historical figures
  3. Ask for the source — if AI cannot name a source, treat the claim as unverified
  4. Check for currency — anything about current events, recent data, or "as of now" needs verification from a live source
  5. Trust your textbook over AI — for school subjects, NCERT is more reliable than AI for facts, definitions, and explanations
The overall mindset: AI is a brilliant study partner who has read everything — but has a terrible memory for specific details and sometimes confuses things they half-remember with things they fully know. Use their big-picture knowledge. Verify the specific details yourself.
Section 8 of 8

📋 AI Error Spotter — Quick Reference

Keep this table in mind every time you use AI for school work:

Error typeMost dangerous forHow to catch it
🌀 Hallucination Specific facts, names, dates, statistics, obscure topics Ask AI for its source; verify in textbook or official site
📅 Outdated info Current affairs, recent events, recent data Ask AI for its knowledge cutoff; use a current news source
⚖️ Bias Indian history, culture, society topics Ask for the Indian/South Asian angle specifically; cross-check with NCERT
💬 Overconfidence Every answer — confidence ≠ correctness Ask AI to rate its confidence on each fact; red-flag specific numbers and superlatives
Arun's lesson: After the history incident, Arun added one rule to his AI use: "I ask AI for context and understanding — never for specific facts about specific people or events without checking." His History marks improved — not because he trusted AI less, but because he used it more wisely.

🧠 Quick Quiz — Test Yourself!

10 questions · Click your answer · Check your score at the end

1. What is an AI "hallucination"?
2. Why doesn't AI "look up" the answer when you ask a question?
3. You are writing a project about recent government schemes in India (2025). AI gives a detailed answer. What should you do?
4. Which type of AI error is hardest to spot?
5. You ask AI about an important Indian freedom fighter from your district. AI gives a detailed, confident answer. You should:
6. What is the best way to get AI to admit when it is uncertain?
7. Which of the following is a red-flag signal that an AI answer might be a hallucination?
8. AI bias towards Western perspectives is a concern for Indian students mainly when researching which type of topic?
9. Which of these is a safe, reliable fact to take from AI without extra verification for a Class 7 project?
10. What is the best overall mindset for using AI at Class 7 level?

📝 Worksheet — AI Error Spotter Log

Tip: in the print dialog, choose "Save as PDF" to download.

Next time you use AI for school, keep this log. When you verify an AI answer, note what you found. Copy this table into your notebook.

AI's answer (brief) Error type I suspected How I checked it Was AI correct? ✅/❌
1.
2.
3.
4.
5.

Use this table in your notebook today, or print this page directly if helpful.

📋 Note for Parents and Teachers

What this lesson teaches: Students learn the four main types of AI error — hallucination, outdated information, bias, and overconfidence — and develop practical habits to verify AI outputs. This builds critical thinking and media literacy skills that apply far beyond AI use.

How to check at home:

For teachers: The AI Error Spotter Log worksheet makes an excellent ongoing class activity. Students log AI verifications over a week, then share patterns as a class: "What types of AI mistakes did we find? Which subjects had the most errors?" This builds both critical thinking and collaborative inquiry skills.

← Lesson 6: Research and Reading with AI Lesson 8 — AI Images and Creative Tools →