# Why is an expectation used instead of simple sum in GANs?

Why do the GAN's loss functions use an expectation (sum + division) instead of a simple sum?