Once kids come into HS/Middle School at least now in 2019/2020, most if not all use personal emails and external chatting apps (most likely Messenger if not Instagram Messages or Snapchat). Compounded with psuedonyms, it's hard for any AI to determine who's chatting who... and even if FB/Snap knew who it was, I'm somewhat certain they wouldn't do anything.
However, in Elementary school, email and Hangouts was the way to go (experience from siblings and myself around 5 years ago), so I guess Gaggle's "AI" can determine if elementary and some middle school kids are a "threat".
Regardless, kids nowadays are really into "self-deprecating jokes": "I want to KMS" or other abusive/harmful messages which a good majority use as jokes (in my experience) although they don't have any wrong intentions. I think any AI or NLP model to determine which is a credible threat and which isn't will be extremely hard to tell, even for humans who aren't teens.
It's hard for a human to determine who's chatting who, but not impossible. It'd just take too much attention away from other more important things like teaching.
Getting an AI to determine who's chatting who? It's a lot easier than you think. I recommend you read everything Snowden's released. Or watch Joe Rogan's podcast-interview with him.
Your phone goes with you when you go places. That is literally all that is needed to determine who's chatting who, particularly with using Facebook or Snapchat or TikTok or whatever spyware kids mistake for trendy social apps these days.
> even if FB/Snap knew who it was, I'm somewhat certain they wouldn't do anything.
Maybe, maybe not. You don't know the laws in your country. Even if you did know the laws in your country, people and businesses break laws all the time.
You might not yet see the danger is selling an advertisement to you. I highly recommend you pay attention in any psychology or statistics classes you have.
Sometimes you are joking, but the person you are joking with isn't. Think about that next time you jokingly tell someone to KYS because they got a 65 on a quiz.
Just so you know, they don’t use the AI to determine who is who, just the context of the text. They know exactly who is who despite pseudonyms because of device IDs and tracking cookies and all the rest. The “who is who” is the easy part.
NLP stuff is pretty advanced with things like BERT and T5, but even then, I imagine that doing accurate threat detection will be hard for nns