I feel the same whenever the article relates to neuroscience...its just not the discipline of many in the HN crowd, but because everyone has a brain, people frequently think they have something insightful to contribute about how the brain works.
Contrast this with computer related ropics and the HN expertise shines.
Not my field, but perhaps fear of embarrassment is important? Feynman's snark that (something vaguely like) good cocktail party topics are those where no one present has expertise.
I've seen a biology professor, writing a biology children's picture book, with unusual care, who seemed in part motivated by not being seen by their peers as having gotten it wrong.
I've recently seen comments that since it's HN, voluminous uninformed conversation is fine, and without harm.
So absent ego threat, or clear social norms, the next line of defense for quality is... comment votes? Which seems more a mechanism for norm enforcement than creation.
The HN comment guidelines don't actually recommend "thoughtful and substantive", only that that should increase as topics get divisive. The emphasis seems respectful and curious discussion. That combination might tend towards quality, but perhaps has failure modes. Like concentrations of poorly informed discussion, perhaps less than ideally respectful and curious, either discouraging quality, or taking too long to shift towards it?
One argument for discussion fora supporting personalized filtering, is the speculation that virtual subpopulations might create and sustain norms more easily than an unfiltered population can filter comments.
Brainstorming, perhaps highly upvoted comments might be darkened or otherwise highlighted, analogous to how the downvoted are lightened? Upvotes moving a comment upward, within its parent's node, is a quality signal that is weakened when there are few siblings. So outside of root and big nodes, it's hard to scan/skim for high-quality outliers. Which incentivizes making bulk judgments about whether chunks of the hierarchy, or even entire article discussions, are of a quality to be worth reading. And reduces the incentive for creating quality outliers in parts of the hierarchy where they will remain low visibility.
Contrast this with computer related ropics and the HN expertise shines.