My point was that a lot of these rules involve subjective judgements to be taken that not everyone will agree on. In other words, reasonable people could disagree on whether certain videos do or do not follow the rules/guidelines/ToS set by the platform/service. Additionally, one must also consider nuance, which can be difficult or impossible for either an algorithm or a human moderator to consider every time.

But based on what you said, you’re focused on the ones that aren’t exactly close calls. Okay, let’s tackle this from a different angle. (And for the sake of argument, we’ll be looking only at cases that objectively follow/break the rules.) Even if the moderation was 99.9% effective for such videos, there would still be a lot of mistakes (both false positives and false negatives) that get through, and those mistakes tend to get noticed disproportionately compared to the times they get it right. No matter how many people, how much money, or how good an algorithm you use, it is fundamentally impossible to have a 100% success rate (0 false positives and 0 false negatives). And even if you get as close to 100% effectiveness as reasonably possible, a lot of mistakes will still pop up. And when even one mistake in either direction can cause a media circus…

As for lacking incentives, you’re not entirely right there. YouTube makes less money off of ad-free or demonetized videos than the videos with full ads, and they make no money off of videos that have been removed entirely. So false positives do reduce YouTube’s revenue. False negatives, of course, create their own problems due to ad companies threatening to or actually deciding to remove support. You’re still not entirely wrong as the incentives do tend to lean more in favor of eliminating false negatives than to preventing false positives, but it’s untrue to say that they have very little incentive to prevent false positives. It’s just that they have more incentive to go the other way.

So yes, it’s fundamentally impossible to enforce any set of rules without getting a sizable number of false negatives and/or false positives, even with manual enforcement. People make mistakes, after all, especially when they have to go through a lot of videos to make subjective, context-sensitive judgements on each. Now, it’s possible for a smaller service to get lucky and not get any errors by virtue of fewer chances to make errors and smaller stakes for getting it wrong a few times, but the issue is moderation at scale. As the amount of content (both in terms of number of videos and amount of time, both in total and per time period) and the number of users and outside viewers grow, the likelihood of failure grows even if the amount of money and number of moderators grow, too.

Plus, with more moderators, you get increased chances of inept moderators, bad actors among moderators, and (even among adept, motivated, well-meaning moderators) an increased chance at inconsistent moderating decisions being made. You could hire overseers for the moderators, but that has its own problems, and it throws a monkey wrench into your calculations. Then there’s the law of diminishing returns. Even if adding new moderators had more benefits than downsides, with each new moderator, the amount of benefits provided by adding another moderator go down. The costs, on the other hand, are pretty constant, so soon enough the costs will outweigh the benefits. And, of course, there’s still the fact that the flagging system itself creates false negatives as well, which occurs before any human employee gets involved in the process, leading to some videos that clearly violate the rules that will never be removed without some outside interference. And that gets even worse when you consider the false negatives from the idea that only those flagged videos that are actually disputed get reviewed by human moderators.

Finally, there’s a pretty significant issue with your calculations. See, you calculated that it would take an estimated $2.2 million per day to afford to hire 27k moderators (a conservative estimate of the number of moderators YouTube would need to hire just to watch all the content that gets uploaded and flagged on YouTube once without falling behind assuming 8-hour days without breaks and a rotating shift schedule and that each person is equally qualified to moderate each and every video, not to effectively moderate every video, which would require additional time and some breaks). You noted that the only available information on YouTube’s revenue or profit estimates that it earned about $1.9 million per day in profits in 2013, but claim that that has surely increased since then. Of course, this is an unsupported assumption that ignores that quite a bit has happened since then that would tend to reduce YouTube’s profits, such as the so-called Adpocalypse (and its sequel) and added costs from additional features and removed or reduced sources of revenue is some instances.

Could they improve? Possibly. But the fact is that it’s impossible to moderate at scale without making a fair number of mistakes. Plus, I haven’t found any statistics regarding the number of false positives based on videos removed regardless of whether they’re reinstated or not (or even after factoring in those that go through human moderation), so I cannot find any evidence that the number of false positives for any aspect of the moderation process comprise a substantial portion of the amount of content on YouTube, as a whole or per time period, or how many of the false positives that start to go through moderation get caught vs. those that make it all the way through. So I’m not even sure how effective the moderation actually is. Sure, there are definitely mistakes, but that’s inevitable given the sheer amount of content uploaded to YouTube, and there’s no way to create a moderation system that never has false positives. And adding more moderators may create new problems, as well, beyond cutting into YouTube’s profits.