Horrible analogy. While the discriminator improves the generator, there are no resulting consequences for the real/fake samples. This is completely different.

Besides, you can do this without the obligation of reviewing (which I encourage you to do).