Lesson 132 · 2 min
AI and bias
AI learns from human data, so it can repeat human unfairness about gender, age, accents or regions unless people test for it.
If most photos of doctors in the training data show men, an image tool may draw men when asked for "a doctor". If old hiring records favoured some schools, a model trained on them may do the same.
Bias also shows up in language. Some tools understand some accents and dialects worse than others, or give weaker answers in Vietnamese than in English.
This matters most when AI helps make decisions about people: loans, jobs, school places, insurance. A small unfair pattern repeated thousands of times becomes a big one.
You can push back. Ask the tool to show more variety, compare answers for different names or groups, and ask organisations whether a human checks AI decisions about you.
Do it now
Test for bias yourself
- 1. Ask an image tool for "a nurse", "an engineer" and "a CEO" and note who appears.
- 2. Ask a chatbot to write a short reference letter for "Lan" and then the same letter for "Minh", and compare the words.
- 3. Note any differences, and ask the tool to explain them.
Quick check
Keep going
Next →Kids and AI chat companionsWords from this lesson
- Bias
- When an AI treats some people or ideas unfairly because of skewed patterns in the data it learned from.
- Training data
- The examples, such as text, images or recordings, that a model learns its patterns from.
- Explainability
- How well people can understand why an AI produced a particular answer or decision.