When someone tells you "It's not hard to automate content moderation" they are: 1. Not an engineer 2. An engineer that has never actually tried to do this 3. Trying to grift someone There is no fourth option.
-
The “creeps that watch kids” thing was driven by a small group that focused heavily on a small number of otherwise-unpopular videos. They basically announced what they were doing in the comments with timestamps etc.
Once google knew the context, they could have built an (approximate) list of these users and blacklisted their views from the recommendation algorithm (which they probably already do for suspected bot accounts). This would have mostly prevented recommending these vids, I think.