In the end, scientist themselves will decide which future research papers are interesting based on what they themselves can build on, and spot problems caused by overreliance on AI, just like they historically spotted problems with overreliance on other approaches.
A big issue is the training of future scientists. AI can be used to make it easier for them to learn (which is good) or to make it easier for them to avoid learning (which is bad). With today's students, the latter is becoming a MASSIVE problem, calling for big changes in how learning is assessed.