A lot has been made of the "pre-AI" internet as a sort of barrier for training sets as it will flood the internet with AI content that will be difficult to pars

A lot has been made of the "pre-AI" internet as a sort of barrier for training sets as it will flood the internet with AI content that will be difficult to parse out later. For the bullish-AI types, is there a prevailing hope that "human content" may itself be an artificial barrier and a properly trained AI could generate better training content for future AI generations? Is this strictly in "AGI" territory?