Fairness is equal opportunity

Course Q&A Generative Adversarial Networks (GANS) Build Better Generative Adversarial Networks
1

In the quiz, it asks

Another measure of fairness is equal opportunity, which means that those members with a property should have equal probability of a positive prediction, but not necessarily those without that property. Suppose that the ability to fly is meaningful to the prediction, and other properties are not considered. Your model predicts a positive result for 10 of the 80 Elephants which can’t fly, and 15 of the 20 Elephants which can fly. If there are 40 Flamingos which can’t fly and 60 which can fly, which of these are valid numbers of positive predictions under equal opportunity:

The answer explains

As 3 of 4 flying Elephants had a positive prediction, 3 of 4 flying Flamingos should have one too.

Why are flightless flamingos not considered here ?

2

@mc04xkf, equal opportunity is defined specifically to apply to members that have a property. In this case, the property is the ability to fly.

We use equal opportunity to help ensure that among members with a property, we want all those members to get an equal opportunity, regardless of which group they are in. So, if we have a positive prediction for 15 of the 20 elephants that can fly, that’s 3 out of 4, so we would want to see approximately 3 out of 4 of the flamingos who can fly also having a positive prediction.