Citrus Ocean

(Since tumblr won't let me post more than 10 pictures in a post, I will just copy and paste the thread after the girls)

.

.

Coco × Flammulated Owl

Agott × Lesser Sooty Owl

Tetia × Maned Owl

Riche × Unspotted Saw-Whet Owl

.

.

The rest of the characters on twitter! 🦉

.

.

.

(Oh, come on, baby)

(Oh, come on, darlin')

(Let me steal this moment from you now)

(Oh, come on, angel)

(Come on, come on, darlin')

(Let's exchange the experience)

.

.

.

And it's not always because some executive made a decision to favor a TERF. It's because the technology is broken in a way TERFs have learned to use to their advantage.

So content moderation is a HUGE task. It is incredibly inefficient to manually monitor all user reports-- when you "flag" something as spam or inappropriate or 'inciting harm' (what Blueskye is saying the banned users had in their posts). So what actually happens is that the bulk of it is reviewed by machine. The precise "how" is a bit of a black box (if you say exactly, it makes it more vulnerable to being bypasses), but we generally know it's a bit of machine learning ("ai", but not he chatgpt kind) and a bit of automation. Generically this gets referred to as an "algorithm" (which is a meaningless word I don't have time to unpack here).

The TERFs have a formula. A battle plan, if you will. I've been watching them use it for almost a decade. And they play social media sites like a fiddle every single time.

The way it works is that they coordinate attacks on targets they dislike. (Usually trans people, but ace discourse is a notable example of targets not necessarily fitting that demographic.) They send links to each other of posts to report. They'll flag it for whatever the nearest thing they can is. The blueskye users were wishing ill to JK Rowling, so "inciting harm" was the excuse selected. "Inappropriate or sexually explicit content" is a common selection for other trans content.

Then what happens is the content moderation "algorithm" sees a huge influx of reports for a certain post or user. And instead of making a human manually review all of that, it automatically initiates a ban or takedown. And if the "algorithm" has word censorship built in, it will begin to auto-flag for things containing words in posts frequently flagged, so it begins to auto-flag the word trans itself.

This has been happening for almost a decade. There was a trans youtuber who proved TERF hater group chats flagged his videos so much that it started auto-demonitizing identical videos with the word trans in the title, but leaving ones without the word alone. I strongly suspect this is why we see huge amounts of trans women banned for no reason on tumblr. And this is clearly what is happening now.

How do we fix it? There's no easy solution. All the things about the "algorithm" that make moderation easier for the social media platforms are the exact reasons it's so difficult to defeat. It's hard to teach a machine what's a genuine flag and a malicious one. Again, it's been almost a decade since the issue was raised with youtube, and there has yet to be a solution for rooting out malicious coordinated flagging.