Off topic, writing code/articles using AI is cool and all, but it suddently occured to me, what if we want to study pure-human generated content and distinguish

Off topic, writing code/articles using AI is cool and all, but it suddently occured to me, what if we want to study pure-human generated content and distinguish AI generated ones? Is there a reason why gpt-3.5 and gpt-4 all using pre 2021-09 data? Is it because gpt-3 as a base model was trained at that time, or is it really because bots are "contaminate" the Internet from then? I guess we can never go back now. One thing I am certain, after 2022-12, AI might enforce itself by learning AI-generated content, and perhaps someday, AI may deem human content as somekind of "bias".