Can we build better AI off of the data these moderators have generated to make their lives easier?
Can you crowdsource the moderation task to double check what the ai is flagging?
Can you crowdsource the moderation task to double check what the ai is flagging?
I highly doubt that. These reported posts might contain private data or highly shocking content; putting that on a public platform, especially associated with the name of Facebook, is very unlikely to happen.
Is that not what the person who originally posted the content being moderated is doing?
The person that posted the content does not expect a random person from the internet to see a video of his children because it was accidentally flagged by an AI as child porn. :)
This would help with false positives, which is a noble goal. But it would not help with false negatives, which is imo a very disastrous consequence in this specific scenario.
Something that isn't CP got marked as CP? Sure, a moderator can manually doublecheck that flag and override if needed. Something that is CP didn't get marked as CP? You got a problem on your hands, because there is no way to manually doublecheck that, since that would mean manually checking every single post marked as "no issues here", which gets us back to square one of manual moderation of all posts.
Not even mentioning the exposure of FB users to CP that it would bring, which would mentally ruin quite a lot of people and bring very negative press to FB on the level of severity that we haven't seen before.
Or to restate, sites that rely purely on manual moderation of content that breaks rules still has content that breaks rules leak through all the time.
"Facebook sent me child porn" is probably not a headline Facebook wants out there.
> Can we build better AI
> Can you crowdsource the moderation task
Users don't want the back-stop to be "a critical mass of community members object;" for content like this, they want to see zero of it, ever, and will choose to use another site that satisfies that hard constraint if FB cannot.
> they want to see zero of it, ever
> When the goal is to prevent users from
> seeing this content, crowd-sourcing moderation
> defeats the purpose.
No system can ever stop 100% of users from seeing 100% of the material they find objectionable. Every user has their own idea of what material is beyond the pale, and even if we invent AI that is competent at blocking content, and capable of tailoring its filters per-user, it will still fail some of the time, because users are people, and people's tastes change over time.
Aside from that, the way Youtube, Reddit, Twitter et al, currently operate, a user who reports objectionable content typically waits days, weeks, or years (or forever) for the company to take action. If you give a user a little real agency, it goes a long way to mitigate their displeasure over occasional exposure to unwanted content.
Meanwhile, if your system requires users to view beheading videos regularly, people will just migrate to a site that doesn't require that.
> What real agency does the system you propose offer?
> if your system requires users to view beheading videos regularly
> people will just migrate to a site that doesn't require that.
Precisely. How often? Close to never.
https://www.bangkokpost.com/thailand/general/1853804/mass-sh...
I could swear there was also a Christchurch copy-cat attack in Europe, but I can't remember sufficient details about it to find an article. Perhaps that one wasn't streamed on FB.
I don't have encyclopedic knowledge of Facebook atrocities (I closed my account nearly a decade ago), but without vetting content before it goes live, I don't see how they will entirely prevent these videos from reaching some users.
(That's before we factor in unintended consequences such as the risk that a critical mass of moderators decide the garbage is signal and start passing it. It's more risk than FB wants to take on for a problem they already solve via paid employees).