AI Ethics: A Practical Handbook

Главные выводы

• Start with a ten-minute demo at the table, because a child who has watched a chatbot invent something understands every rule that follows.
• Before any AI app comes home, spend twenty minutes finding its age limit, what it collects, whether it learns from chats and how to delete.
• Make a homework help card with your child, subject by subject, and let the teacher's rule win whenever the card and the classroom disagree.
• Swap names for roles before pasting, and ask each recognisable person before their face or voice goes into any app.
• Run the name-swap test together: ask the same thing several times, change only the name or gender, and compare what comes back.

Внутри этого бита

1. A ten-minute demo that explains the machine

   Rules about AI make sense to children once they have seen what the tool actually is. An AI chatbot is a model: software trained on huge amounts of writing to predict which words usually come next.

2. The twenty-minute check before an app comes home

   Most apps explain themselves in two documents nobody reads: the terms of service, which set the rules of use, and the privacy policy, which says what happens to your information. You do not need to read either in full.

3. The homework help card

   Arguments about AI and homework usually start at nine at night, the worst time to invent a policy. The help card settles it in advance.

4. Pasting without exposing anyone

   Parents use chatbots for delicate jobs: an email to the school, a complaint to the landlord, a question about a relative's test results. The risk is that a chat can be stored and reviewed, and anything that identifies a real person is personal data that was never yours to share.

5. Asking before a face or voice goes in

   Apps now turn photos into cartoons, animate old portraits and copy voices for greetings. Faces and voices deserve more care than a password, because you cannot change them if they leak.

6. The name-swap test for hidden bias

   Bias in AI means a tool treating people differently for reasons that should not matter, such as gender, religion or where they live. It usually comes from training data that reflects old patterns, including stereotypes: fixed ideas about what a kind of person is like.