Ask HN: Real Karma?

I was wondering, with AI and stuff, why do we still rely on human likes and dislikes? (in general, not HN-specific)

It seems like an ideal application of text-based inference: to assign points depending on how someone interacts in a discussion (using correct factual arguments, not straying away, not misdirecting, being considerate and reasonable, and so on).

It surprises me that no one tried it before.

So, why we don't have any automated explicit karma-based moderator anywhere?