Comment on Meta plans to use AI to automate up to 90% of its privacy and integrity risk assessments, including in sensitive areas like violent content

AstralPath@lemmy.ca ⁨5⁩ ⁨days⁩ ago

• HowAbt2day@futurology.today ⁨5⁩ ⁨days⁩ ago

  Not suitable for Lemmy?

  source

  • blargle@sh.itjust.works ⁨5⁩ ⁨days⁩ ago

    Not sufficiently fascist leaning. It’s coming, Palantir’s just waiting for the go-ahead…

    source
• ouch@lemmy.world ⁨5⁩ ⁨days⁩ ago

  What about false positives? Or a process to challenge them?

  But yes, I agree with the general idea.

  source

  • beejjorgensen@lemmy.sdf.org ⁨5⁩ ⁨days⁩ ago

    Or a process to challenge them?

    😂😂😂😔

    source
  • tarknassus@lemmy.world ⁨5⁩ ⁨days⁩ ago

    They will probably use the YouTube model - “you’re wrong and that’s it”.

    source
• brorodeo@lemmy.ca ⁨4⁩ ⁨days⁩ ago

  Bsky already does that.

  source
• head_socj@midwest.social ⁨5⁩ ⁨days⁩ ago

  Agreed. These jobs are overwhelmingly concentratedin developing nations and pay pathetic wages, too.

  source
• towerful@programming.dev ⁨4⁩ ⁨days⁩ ago

  Yup.
  It’s a traumatic job/task that gets farmed to the cheapest supplier which is extremely unlikely to have suitable safe guards and care for their employees.

  If I were implementing this, I would use a safer/stricter model with a human backed appeal system.
  I would then use some metrics to generate an account reputation (verified ID, interaction with friends network, previous posts/moderation/appeals), and use that to either: auto-approve AI actions with no appeals (low rep); auto-approve AI actions with human appeal (moderate rep); AI actions must be approved by humans (high rep).

  This way, high reputation accounts can still discuss & raise awareness of potentially moderatable topics as quickly as they happen (think breaking news kinda thing). Moderate reputation accounts can argue their case (in case of false positives). Low reputation accounts don’t traumatize the moderators.

  source