🧪 The AI Lab← All labs
Class 7 & 8 · Data, Fairness & Bias

⚖️ Bias & Fairness — why AI can be unfair

An AI only knows what you show it. It learns patterns from its training data — so if the data is lopsided or unfair, the AI becomes unfair too. Computers don't invent bias on their own; it sneaks in from the data, the design, or human choices. The rule to remember: Garbage In, Garbage Out. Play with the two demos below to see it happen!

Demo 1 · Balance the training data

Teach the AI: 🏏 cricket bat vs 🏸 badminton racket

Slide to choose what the AI sees while learning. Then watch how well it scores on each thing.

🏏 cricket bats 90% 10% badminton rackets 🏸
📦 The training pile the AI studies
🏏 cricket bat
–
🏸 badminton
–
🙂 Overall fairness
move the slider and test the AI

Demo 2 · Name that bias 🕵️

Read each real-life story, then tap the kind of bias it shows. Score 0 / 0

Loading…
🤖 In a real AI — bias comes from the data A hiring tool once learned from years of resumes that were mostly from men, so it started preferring men — that's historical bias. A famous study by researcher Joy Buolamwini found face-recognition systems were wrong on under 1% of light-skinned men but 30–35% of darker-skinned women — because those faces were barely in the training data (data bias). The fix is the same every time: diverse & balanced data, test before you launch, add human review, and keep checking.

Practice 🎯

← Back to all labs