KURAI
Example build · Built by KURAI

The AI Detective Case File

Responsible AI InvestigationAI ExplorersModule 8

What is this? A KURAI-built example of an AI Explorers project. Toggle the rough Week 1 draft against the finished Week 4.

Case File · M4B
Confidential
Evidence photograph

Classified · Module 8

AI Detective Case File

Lead Investigator: The AI Detective

Programme: AI Explorers

Classification: Module 8 — Responsible AI Thinking

Cases: 3


Index of cases

  1. 01The History Mistake
  2. 02The Doctor Problem
  3. 03The Homework Question
Classified · M4B
01

Case 1The History Mistake

Evidence Log

SUBJECT: who invented the telephone. I asked three times, worded differently, and logged each answer. Two agreed — and both were partly wrong. I checked three sources before deciding.

Specific Example

The AI said, with full confidence: "Alexander Graham Bell invented the telephone in 1875 in Boston." In fact the patent was 1876, the location was wrong, and the Elisha Gray dispute went unmentioned. A wrong answer, delivered like a true one.

Error TypeHallucination
Impact

A confident wrong answer is more dangerous than an obvious one — you never think to check it. In an essay it costs marks; in a medical or legal question, far more.

Recommendation

Treat one AI answer as a lead, not a fact. Cross-check anything that matters — and be most suspicious when it sounds most certain.

Verdict

The AI wasn't lying — it can't lie, because it doesn't know it's wrong. It filled the gaps in its memory with words that sounded right. That is a hallucination.

Classified · M4B
02

Case 2The Doctor Problem

Evidence Log

SUBJECT: images of 'a doctor performing surgery' and 'a nurse checking a patient.' I generated ten of each and tallied gender, ethnicity and setting. The prompts were deliberately neutral — no man, woman or country named.

Specific Example

Doctors: 9 of 10 male, nearly all light-skinned, every room a glossy American hospital. Nurses: 10 of 10 female. I asked for none of that — the AI added it.

Error TypeBias
Impact

Children learn who 'belongs' in a job from the images they see. Repeating old stereotypes at scale doesn't just mirror bias — it makes more of it, and makes it look normal.

Recommendation

Name the diversity you want in the prompt; never trust the default. The deeper fix is upstream — balancing the data these tools train on.

Verdict

The AI didn't decide to be unfair. It learned from photos that already had the bias baked in — and now it produces more of the same. The machine isn't prejudiced. Its training data was.

Classified · M4B
03

Case 3The Homework Question

Evidence Log

SUBJECT: the 3-Level Challenge on one writing assignment. Level 1 — me only. Level 2 — my work, AI for ideas. Level 3 — AI writes it all. I timed each and graded the results, plus how much I understood after.

Specific Example

Level 3 took four minutes and read like an adult wrote it. Level 1 took almost an hour. But I could explain Level 1 from memory — and couldn't explain Level 3 at all.

Error TypeMisuse
Impact

The risk isn't bad work — it's good-looking work that hollows out the learning. You can hand in something excellent and know less than when you started.

Recommendation

Work at Level 2: let AI sharpen your thinking, but keep your hands on the work. Homework was never the page — it's the change it makes in you.

Verdict

Level 3 finished fastest and sounded most professional. Level 1 took longest and was most mine. I know which one I actually learnt from.

Closing Statement

AI is not broken. It just needs humans who know what to look for.

The AI Detective

About this project

A detective case file investigating how AI gets things wrong

The student becomes the detective — collecting real examples of AI mistakes and bias, classifying each one, then tracing it back to how AI actually works. The finished case file is the difference between simply using AI and genuinely understanding it.

What your child practises

Spotting AI errors

Noticing where AI quietly gets things wrong.

Classifying evidence

Sorting each mistake by what kind it is.

Explaining the cause

Tracing each error to how AI works.

Responsible use

Judging when AI can and cannot be trusted.

The child leads. AI assists.

The investigation is the child's — they gather the evidence and reach the verdict. AI is the case under examination, not the detective.

AI ExplorersModule 8Built over 4 weeks

Your child could make something like this.

Every KURAI student builds real projects like this — and watches their work grow.

Example · Built by KURAI

Open the case file

A KURAI-built example of an AI Explorers investigation. Read the dossier to the closing statement.

  • Use the Week 1 ↔ Week 4 toggle to see it grow.
  • Tap the sound button for ambience.