Any academic is familiar with the tendency of new students, fresh to reading papers, to come in basically two flavors:
a) It's in a paper, that was published!, it must be good, and correct
b) Everything is crap
Reading academic papers, understanding them, and judging their quality, is a skill. It takes practice to get good at it.
Over time, we naturally raise - and lower - our cultures' competency in certain skills as we practice them more - or less.
If/as we adopt reviewing practices that use AI to replace parts of the work, the scientific community's ability to do those things will naturally atrophy.
In particular, if AI is used as a judge of the quality of work, we, as a scientific community will get worse at distinguishing good work from bad - it's inevitable.