"Quality" is a really hard thing to define algorithmically. Even if you had great journalists on staff, I don't know if their heuristic expertise could translate to an algorithm.

Most signals of trust/quality come via backlink profiles, which may be a poor proxy but are still the best that Google has been able to come up with.

The human raters are used to assess the performance of the algorithms (i.e. do human opinions align with algorithmic ones), not to rank content directly.

A recurring theme with Google is that they always want to solve problems algorithmically.

But leaning on automated algorithms seems like the only way to realistically scale for day-to-day needs? I was going to posit that Google seems like the company most open to mixing human and automated processes. But that's not necessarily fair to Facebook, for which moderating user content and interactions is a much different problem space than search rankings; and of course, YouTube seems to have the same issues as FB does.

And what do they do with these human assessments then? Of course they feed them back to the algorithm so that it can improve. In effect it is still humans influencing the search rankings albeit indirectly. So it is a bit duplicitous to say that "it is the algorithm that decides" when there are humans providing learning data for it.

It's still algorithm-first. It's not "learning data" in the sense that human ratings are used to train the algorithm.

It's more of a check step. So, for example, if humans think one website is vastly more (or less) authoritative than human raters, the engineers might dig in to see which aspects are causing the algorithm to evaluate it differently and, potentially, test tweaking the algorithm accordingly.