I understand what you mean, but that's not how it would be supposed to work. The idea is pre-filtering by that AI, such that it only shows you posts that are most likely shitposts and / or spam. We recognize those at first glance, so a computer program trained with enough data should be able to roughly categorize posts, as well. Only the ones most likely to be useless to the forum would be displayed, and even then, a human will still need to decide whether to report them or not.

It is not meant to 'decide what we see' algorithmically, rather just use a software to facilitate people's job to find posts that are to be considered report-'worthy'.

Exactly like that, just a bit... 'smarter'? But maybe that's not needed, after all. I'm all about keeping things simple and stupid, so if your method works fine (low false positive rate), that's great!