Using #AI generated training data for training newer AI models seems like the computer equivalent of inbreeding.

Isn't this obvious to AI developers?

What's the solution?