Six questions to ask before using any AI
Big idea: don’t only ask “can a machine do this?” Also ask: should it? Who does it help? Who could it hurt? And who is in charge?
The story in short
At the village Vigyan Mela, a salesman offered the school a camera that would watch every child and report who was “not paying attention”, straight to parents’ phones. “It is 90% accurate!” he said. Bunny asked six questions, and the village decided not to buy it.
Bunny’s six questions
- What did it learn from? Videos of children, mostly in big-city schools in other countries.
- Who might it be unfair to? Children who look, sit or dress differently from the training videos.
- How sure is it, and who checks? 90% accurate in a class of 40 means about 4 children marked wrong every lesson: around 24 wrong marks a day.
- What happens to our data? Videos of children’s faces stored far away and reused, without asking parents.
- Can we say no? “It only works if every child is watched.”
- Who is in charge, and can we switch it off? Dabba the robot had a big red button. The camera had none.
Key words
- Personal data: information about a person, like their face, voice or name.
- Privacy: the right to decide who sees your information.
- Consent: permission, freely given, that you can also take back.
- Accuracy: how often something is right.
- Accountability: being responsible for what happens.
Use them anywhere
These six questions work for a new app on your phone, an AI feature at your office, or a face-scan at a shop. Grown-ups can add two more from Volume 2: can it be tricked? and is it worth the cost?
Check yourself
An app is 95% accurate and checks 200 students a day. About how many students does it get wrong each day?
A new photo app asks for access to all your contacts, location and microphone just to add filters. Which of Bunny's questions fits best?
Which of these is real consent?
More for you: 🧒 Kids · 🧑🏫 Parents and teachers