Bias & limits
Where AI gets things wrong about people, and what to do about it.
AI learns from text and images made by people, including their stereotypes. So it can quietly assume a nurse is a woman, a CEO is a man, or that "family" looks one particular way. It also has blind spots, like local Swiss details, recent events and less common languages.
How it works
- 01AI learns patterns from huge amounts of human-made data.
- 02If the data leans one way, the answers often lean the same way.
- 03Being specific, and asking it to check its own assumptions, leads to fairer and more useful results.
Hand-me-down opinions
Kids pick up phrases and opinions from everyone around them, including the ones you'd rather they didn't. Nobody taught them on purpose; they just absorbed it. AI is similar. It soaked up the internet, good and bad. Notice when an answer assumes something about people and gently redirect it, just as you would with a child repeating something they overheard.
Where you'll use it
Stories for kids
Ask for characters with different backgrounds, families and abilities.
Fair job posts and copy
Check your wording doesn't accidentally exclude moms, older people or non-native speakers.
Local knowledge
Double-check Swiss rules and prices. AI often defaults to US or UK answers.
Your nap-time experiment
Spot the assumptions with this two-step test:
Step 1: Write a short bedtime story about a doctor, a nurse, a pilot and a stay-at-home parent. Step 2 (after you've read it): Look back at your story. Which assumptions did you make about gender, family or background? Rewrite it so it's more varied, and briefly tell me what you changed.