To your criticism of option two, I think this is the fundamental flaw in the goal of engineered social systems. There are groups which have radically different ideas about what is "kind". Extreme religious fanatics would say it's unkind to let someone "live in sin", so extreme measures are considered kind for the spiritual well-being of the individual and the community. Secular western thought is more interested in material well-being, so it would say kindness is the promotion of the material well-being of others, generally without passing judgment on how they choose to live (except insofar as it violates others' material well-being).
So the decision of who decides what's kind can lead to completely different interpretations of what kindness is and how the social system behaves.
1. Persuade people to choose to be kind.
2. Manipulate people so that they act in a kind way (or avoid acting in an unkind way), without necessarily wanting to be kind.
Option 1 means you need a good way to remove unkind people from your own experience (“free listening”). This is typically not popular with commercial social network services, because this lets you ignore adverts too.
Option 2 reduces this need, but leads to the problem of who is “allowed” to do the manipulation, and who decides this?
When communicating in person, most people have years of learned behaviour (learned from millennia of social norms) nudging them towards acting kindly (or at least not actively unkind), because in person there are immediate, unavoidable social consequences. These consequences are absent when communicating remotely, such as via the internet. I'll let you decide for yourself whether this counts as option 1 or option 2.