Heavenbanning: Using AI models to commit banishment through praise

The SF/VC crowd, for all of it's feel good bullshit, essentially subscribe to a brand of technocratic authoritarianism. Stuff like heaven banning sounds fine in principle, it game theoretically solves a problem if you acknowledge that certain speech is a problem and that a group of engineers or executives in SF can identify it.

Have you ever listened to those people talk though? If I had to categorize them as a group they are maybe 2 levels more coherent than the Alex Jones crowd with the exception of they can rattle off a bunch of coherent thoughts regarding programming languages and systems architecture. I hear as much dumb/crazy shit out of that group as any other.

You've got documented cases like elder statesmen Tony Hsieh who would be the kind of people who would be in positions to make broad decisions on the acceptable speech of millions who was spouting off ridiculous utopian ideology, had zero grasp on reality, and died huffing nitrous in a shed.

People support stuff like this because they see themselves as the kind of person who would be setting the rules. But the rules are much more likely to be applied against them then by them. The coders mostly aren't writing or enforcing the rules, they are writing the implementation.

Soon enough, it will be easy enough to automate the people out entirely and have computers deciding who's speech is acceptable based on datasets some researcher chose and creating false realities (heaven banning fits this description) for real people who don't communicate in the way the computer see's fit.

This isn't something anyone should be celebrating

The idea is to show banned users what they want to see, in the form of LM-generated responses to their posts, relegating them to a "pocket dimension" of sorts where all other users are fake bots. I can see a softer form of this being implemented, where a user's posts are still visible to all, but in a form rewritten by the LM. And, likewise, the user sees their original posts as well as LM-rewritten posts of other real users. This softer form wouldn't just be applied to banned users -- everybody is in their own bubble. It won't be used to improve the quality of the content on the platform; it will be used to drive addiction (ahem, engagement). So, naturally, I see this being implemented by Facebook.

This seems at its core, like shadowbanning, actively hostile at some level? It would be more altruistic to just be honest with the banned user; this is manipulative and wasting their time.

Seen from that perspective someone has to wonder why a company would do this, actively cause harm to a customer?

It drifts into something resembling sabotage, and labeled that way would lead to a host of questions and criticisms.

Do you need agreeing replies? You can do this with upvotes without ML. I wonder if HN does something like this (shadow modding is already used).