I'm fairly sure that most of the content watched on youtube was uploaded by someone other than the copyright holder, at least historically. Having an algo people have to actively thwart, can be unaware that the particular content is copyrighted, and/or generates a significant enough number of false positives that google can claim that they're trying hard enough to damage a portion of their legal business is a tool in and of itself.
Not using an algo makes it clear you're doing editorial, and once you're doing editorial, oversight follows. Anyway, even with the algo you can do as much editorial you want through training and dispute resolution. By slow-walking some disputes and fast-tracking others, you can change the shape of your business to hopefully be more independent of legally fragile stolen content.
The Trump/Russia conspiracy was a boon to all of these algorithmically filtering DMCA carriers, now they can actually explicitly filter points of view. I decided that point was inevitable anyway after I heard of the first facebook censoring of an anti-wingsuit post. IMO humans + that degree of editorial = worst case scenario, at least laundering it through the AI means that editorial requires some cost and degree of engineering skill.
https://www.nationalgeographic.com/adventure/features/why-ar...
> “The post got deleted by Facebook, and I pissed off the entire community,” says Lewis. “My message was followed up by the most fatal month in BASE. I just painfully sat back and watched my friends die one after the other.”