It's become disconnected so much that metrics of quality are seen as arbitrary, unrelated attributes.
There is literally zero reward for hard work. It's actually more likely to hurt than help you because original scholarship or novel insights is definitionally more dissimilar to existing things than low effort derivative spam.
There needs to be a fundamental redress that combines with categorical human-based systems. Categorical not only in what they "like" but in their modalities of engagement and what the styles of flow mean.
For example, if a 4 year old watches a video 150 times over 2 weeks, they're being a 4 year old. If a 40 year old does that exact same thing, then it's a completely different meaning.
There's other modalities, do people complete it but take longer then the duration (maybe to look things up) or do they set it at 1.5 speed and scrub through it. There's ways of characterizing these interactions.
Choosing to be naive about such things in our search and recommendation systems is at our own peril and it's a big part of the SEO spam and low quality avalanche that we are all awash in.
Digging ourselves out of it will require us to somehow capture a human part of human beings we haven't appeared to digitally organize yet.