You actually could judge a video to a surprising degree. Consider a system that 1) performs object recognition on videos, 2) uses natural language processing (NLP) on the video speech, 3) evaluates the broader cluster created by the video poster's other content and 4) uses these as inputs in a classification algorithm that predicts a probability that the flagged video represents -- for lack of a better phrase -- "hateful content". If the poster were a journalist conducting an interview or reporting from the field, that will be revealed in the NLP algorithm and the cluster the video poster is in. If the poster is Aunt Polly at a picnic and the "gun" is a toy waved by a kid, that also can reveal itself in the NLP data. This is doable today, right now, with current machine learning engineering. It may not be able to be 100% automated, but it could help sift the triggering content and make fair-minded human supervision a bit easier. And it would continually adapt and improve. But nothing can fix bad faith. If the tech giants want to be political censors for the Washington, DC power elite (aka "the Swamp"), there's not much we as general public can do about it other than not use their services.