You Are the Teacher
In every lesson in this module, we have talked about examples — good ones, bad ones, lots of them, tested ones. But here is a question we have not asked yet: WHO decides which examples to use? The answer is: a person does. Maybe a scientist. Maybe an engineer. Maybe — in some tools you can already use today — YOU. That means the human who chooses the examples is the real teacher. And teachers have real responsibility.
The Human Is Always in Charge of the Examples
A machine cannot go out and collect its own examples. It cannot decide that the examples are fair or unfair. It cannot notice that all the photos of "doctors" only show one kind of person. It just trains on whatever it is given. That means the person who PROVIDES the examples is making powerful decisions: Which things get included? Which get left out? Are the labels honest and accurate? Are there enough examples of every kind of situation? Are the examples fair to all kinds of people? These choices shape everything the machine will ever learn. The machine will reflect the values, care — and mistakes — of the person who gathered the examples. That is an exciting responsibility and a serious one.
YOU are the teacher. The machine learns what you show it. If you choose examples carefully and fairly, it learns carefully and fairly. If you are careless or unfair, the machine will be too.
Here is a story that shows this beautifully. Two students, Jordan and Casey, each use a Teachable Machine tool (you will get to try one of those in this module!) to teach a computer to recognize their hand signals. Jordan teaches three signals: thumbs up for 'yes,' thumbs down for 'no,' and an open hand for 'stop.' Jordan is careful and collects 30 examples of each signal, from different angles and lighting. The machine learns all three perfectly. Casey collects 30 examples of 'yes' but only 3 examples of 'no.' The machine learns 'yes' well but keeps guessing 'yes' even when Casey shows 'no.' The machine is not being stubborn — it just never got enough examples of 'no' to learn it well. Jordan's machine works well because Jordan was a thorough teacher. Casey's machine has gaps because Casey's teaching had gaps.
Prompt Challenge
You are the teacher! Write a prompt asking an AI to help you think of fair, varied examples to train a machine.
Your prompt should…
- Tell the AI what you want examples of
- Ask for examples that are varied and different
- Ask the AI to keep the examples fair and balanced
Being a good teacher for a machine means asking yourself some important questions before you start: Am I including enough variety? Do my examples cover all the situations the machine might face in real life? Are my labels honest? Did I label things correctly, or did I rush and make mistakes? Are my examples fair? Do they represent all kinds of people, places, and situations — or only some? These questions are not always easy to answer. Even professional AI researchers get them wrong sometimes. But asking the questions is always the right first step.
If your examples leave out certain people, places, or situations, the machine will not know they exist. Always think about who or what your examples might be missing.
Who decides which examples a machine learns from?
Jordan collected 30 examples of each signal. Casey collected 30 of 'yes' and 3 of 'no.' Why did Casey's machine struggle with 'no'?
Design Your Training Set
- Imagine you want to train a machine to recognize 'things that belong in a kitchen.'
- On a piece of paper, list at least 8 examples you would include. Try to include different types: tools, food, appliances, containers.
- Now think: what have you left out? Write down 2 things a kitchen might have that you forgot.
- Check your labels: are all 8 examples truly kitchen things, or did any sneaky non-kitchen items slip in?
- Finally, think about fairness: would your kitchen examples only show one kind of kitchen, or many kinds from different families and cultures?
- Talk about your design with someone at home.