The concept of Generative Adversarial Networks (GAN) in AI is fascinating.
GAN is a class of machine learning that puts 2 neural networks together.
It's setup as a game, where one generates an output based on the training set it has been provided (eg create a new picture of a cat based on cat pictures), and the other knows what a good output looks like (eg if the picture does indeed ressemble a cat).
"doesn't look good enough, keep trying"
It could be seen as a similar dynamic to that between a human trainer and its trainee (or a parent and child).
The latest and greatest AI models (eg DALL-E) use another approach though - "diffusion models" - though it's harder to get my head around how these work at this stage.
Illustration: 1st result generated by Stable Diffusion with prompt "an artificial intelligence training another"
