I'm increasingly convinced that a lot of smart academics are confused about AI because most of their worst experiences with it come from the classroom. But professors play two roles and are also supposed to spread knowledge.
In class, we test whether students learned something themselves. Sure, test them and do oral exams. That's what I'm planning to do as well. But in journals, we should publish true, original, important claims, whether AI-assisted or not.
There is no easy solution for AI plagiarism. The modus operandi of the journals will have to change, One possibility is that refereeing not be done as now by reading and commenting on the article but by arranging a video conversation between the referees and the author(s). In a conversation, it would become clear if the article is an AI plagiarism or is genuine.
Generally speaking, we have to move to oral examinations.