Somebody chose this
Suppose an AI is unfair to one group. Which choice is behind it?
Step 1 of 4
- Whose examples did it learn from?
- What did it reward as a good answer?
- Who answers for a wrong result?
- A person made each of these choices.
The rule
People choose the examples and the goal of an AI. People must answer for it.
The examples are the data it learned from. If they come from one place, the AI may fail elsewhere.
The goal is what the AI counts as a good result. A goal can favour some people by accident.
The last question is who answers. Someone should be able to explain a result and fix it.
| Case | The choice behind it |
|---|---|
| A plant app learned from one garden only. | The examples |
| An app picks readers by books borrowed. | The goal |
| A bank says "the computer decided". | Nobody answers |
Take the reading app. It picks the best readers by books borrowed. Children with no library nearby borrow few books. The goal counts books borrowed. So the app ranks them low.
The goal is the choice behind it. It favours children near a library.
Find the choice behind each case.
Loading lab…