Current Events > Jesus, Facebook's algorithm censorship is a hilarious mess

(+) Yeah! (+)

My friend was suspended because it detected a photo for nudity, even though it was just a selfie with no skin shown besides her face. She got it overturned, but the profile had to be put into warning mode and never went away.

It autoflagged her post ranting about her abusive ex as hate speech, and she was suspended for 30 days. A week after she came back, she posted a story talking about her son and her friend climbing trees, and falling off and laughing about it. She said in her post "boys are so crazy", and then THAT sentence was flagged as hate speech as well, and she was suspended again.

---

People wanted Facebook to be more aggressive in censoring itself, so this is kind of the result. Given the sheer volume of content, autonomy is the only way to respond fast enough, and when your model is trained to detect things like "<group> are <insult>" without the human ability to understand context, stuff is going to get caught in the net.

Really, though, nobody should be using facebook. It's just a cesspool of pointless drama and gossip.

---

s0nicfan posted...

People wanted Facebook to be more aggressive in censoring itself, so this is kind of the result. Given the sheer volume of content, autonomy is the only way to respond fast enough, and when your model is trained to detect things like " are " without the human ability to understand context, stuff is going to get caught in the net.

They could, y'kno, autoflag rather than outright automod stuff that the AI throws up. And if they feed the human responses back into the AI (including human responses to manually marked posts), that could even train it so that over time it becomes more accurate - though likely never perfect.

---

I got a text from Facebook asking me to come back since I haven't logged in for almost a year. I'm not sure if it was a scam text, but I'm not going back regardless lol.
---

Hayame Zero posted...

My friend was suspended because it detected a photo for nudity, even though it was just a selfie with no skin shown besides her face. She got it overturned, but the profile had to be put into warning mode and never went away.

It autoflagged her post ranting about her abusive ex as hate speech, and she was suspended for 30 days. A week after she came back, she posted a story talking about her son and her friend climbing trees, and falling off and laughing about it. She said in her post "boys are so crazy", and then THAT sentence was flagged as hate speech as well, and she was suspended again.

She is being targeted by facebook. I am not sure sure how or why but it happens, I hear stories like this all the time, facebook is messed up.

Trumble posted...

They could, y'kno, autoflag rather than outright automod stuff that the AI throws up. And if they feed the human responses back into the AI (including human responses to manually marked posts), that could even train it so that over time it becomes more accurate - though likely never perfect.

It could autoflag, but that still requires human intervention, and when people are counting the hours/days between a post being up and it being deleted to try and argue that Facebook isn't doing a good enough job, their only line of defense is more aggressive moderation.

And yea, there are techniques I'm sure they're actively using to try and improve their algorithms, but like you said, they'll never be perfect. They also have to be extremely careful of things like response-based training because that can lead to both increased bias in the model and also opens it up to certain types of adversarial attacks that can be used to effectively train the model away from certain targets.

---

matchboxsantana posted...

She is being targeted by facebook. I am not sure sure how or why but it happens, I hear stories like this all the time, facebook is messed up.

are we sure she isn't being targeted by someone reading and reporting her posts?

---

I had a post in a group flagged for "sexual solicitation".

The post was me talking about a song on Rock Band called Sugarbaby. It was delisted and made available to download for people who previously purchased it

Medussa posted...

are we sure she isn't being targeted by someone reading and reporting her posts?

it depends. If her posts can be read by everyone, yes. If she has them set up so only her friends can read them then either one of her friends actually hates her or facebook is targetting her. Maybe facebook has some sort of rule where after being reported once they scrutinize an account more or something.

With a username like Matchboxsantana I have to wonder if you like it smooth

Badmotorfinger posted...

With a username like Matchboxsantana I have to wonder if you like it smooth

In every breath and every word

matchboxsantana posted...

it depends. If her posts can be read by everyone, yes. If she has them set up so only her friends can read them then either one of her friends actually hates her or facebook is targetting her. Maybe facebook has some sort of rule where after being reported once they scrutinize an account more or something.

No, it's set to Friends only. I asked if there was someone reporting her, and it's all been automods since the first issue. It's in a Warning state - which is like probation - and it pretty much fixates on these profiles and flags terms and content more frequently, and the warning won't go away.

She's been having to use "m*n" and "b*ys" when simply describing the situation.

---

Hayame Zero posted...

No, it's set to Friends only. I asked if there was someone reporting her, and it's all been automods since the first issue. It's in a Warning state - which is like probation - and it pretty much fixates on these profiles and flags terms and content more frequently, and the warning won't go away.

She's been having to use "m*n" and "b*ys" when simply describing the situation.

I meant that human intervention would be required to actually look at the autoflag to determine if it needs to be deleted or not. I did not mean to imply that a person would need to also flag the post.

---

Hayame Zero posted...

asked if there was someone reporting her, and it's all been automods since the first issue.

ok. i didn't know you could tell the difference

---

Reminds me of that site which determined an image of Aladdin and Jasmine on a flying carpet was 80% likely to be porn.