Meta & Children's Online Safety

Documents / Chat: IGWB even explored private by default for teens but the growth impact was too high

Chat: IGWB even explored private by default for teens but the growth impact was too high

Internal document38 messages1 named participant, plus redacted or unnamed

Narratives:Profits over Safety

Open source PDF (3 pages)

A readable view of the chat, built from the transcribed text. Each row is one timestamped message in the record. Times are Pacific, as recorded. Check the Source PDF for the original layout.

████ name redacted. Different bars may be the same person or different people.

quoted in an excerpt in the archive.

Friday 30 October 2020

00:38Darius KilsteinOh hey, super sorry about that, definitely not my intention!! A < Looking back, wasn't a great choice of words for somebody I' ve never worked with! Thanks for the feedback, will be more careful making first impressions:-) @silent
00:41Darius KilsteinAlso just to be clear, I'm very supportive of this work stream. Actually an analyst on my team first fogged the problem about 1.5 years ago and we were surprised nobody picked it up in that time frame, so definitely want to make sure it works! @silent
00:42Darius KilsteinFlagged @silent
00:44Darius KilsteinThis could work. I wonder what % of grooming happens from people with at least 1 NFX action? Like maybe there's a solution for non mutual friending that does not require a prior action against them @silent Our plan S
00:44████to run a small experiment with these conditions, then run classifier estimated prevalence over that to measure the impact against projected impact in Topline engagement
00:45Darius KilsteinLastly, we have reasonable stated age coverage, so that's what I was eluding to for a different way, because then you could start looking at actual age brackets, happy to talk through that path with you @silent
00:45████The issue here is that people lie, and we know that' s quite prevalent
00:46████Especially groomers they lie to bypass our detection and even help them find minors
00:46████WIT WORK wich DS on this to size up
00:46████It's in progress now
00:47Darius KilsteinHave we considered settings to just let people turn off receiving messages from people that dot follow them ?
00:48Darius KilsteinDon't
00:49████We have, and IGWB even explored private by default for teens but the growth impact was too high and the decision was to explore more nuanced and less blunt solutions.
00:49████which is now we got herel
00:50Darius KilsteinYa PBD was my team:-p wasn't an elegant solution. Nobody looked at the messaging solution tho
00:50Darius KilsteinI think it's worth considering down the track
00:50████These restrictions on messaging and recommendations/search are what we have on FB today. We need to have some equivalency on Instagram.
00:51████eg: PYIR was responsible for 80% of violating adult/minor connections
00:52Darius KilsteinI do worry about the effectiveness of the 1 nfx action, I mean definitely better than what we have today and excited to see where it goes
00:52Darius KilsteinBut ppl can create new accounts etc
00:52████Yeah this is something that needs a lot of investment
00:52Darius KilsteinHappy to help you explore a way to make it work for 0
00:53Darius KilsteinFor example I could look at people who have actions taken against them and their stated age to confirm or deny your hypothesis about age lying
00:53████Is there anyone on your team who could spin a few cycles on this?
00:53████This would be great!!
00:53Darius KilsteinYa
00:54Darius KilsteinEg, if we took stated age > 30 and said those people cannot message teens (via model or stated age) if they aren't being followed
00:55Darius KilsteinIrrespective of nfx actions
00:56Darius KilsteinJust need to find the table for nfx actions and I could whip this up in a few hrs
00:56████I love something this neat
00:56████I do finu we have a bias towards overly complex solutions because we can before exploring simple fixes.
00:56Darius KilsteinGreat! I' 11 look into it
00:57Darius KilsteinYeh this place is weird like that
00:57Darius KilsteinCool it must be late where you are sorry to keep you up
00:57Darius KilsteinChat soon
01:35Darius KilsteinOh you're based in London I feel less bad about pinging you now lol
01:36Darius KilsteinTook a quick look at the data and the datasets for CI are pretty tricky to understand, so i want to suggest a partnership b/w my team and your DS team. who's the right DS mgr to reach out to?
03:11████That group 1 ve added you to is the working group putting all this together 1.