We had a really interesting discussion yesterday about voting on Lemmy/PieFed/Mbin and whether they should be private or not, whether they are already public and to what degree, if another way was possible. There was a widely held belief that votes should be private yet it was repeatedly pointed out that a quick visit to an Mbin instance was enough to see all the upvotes and that Lemmy admins already have a quick and easy UI for upvotes and downvotes (with predictable results ). Some thought that using ActivityPub automatically means any privacy is impossible (spoiler: it doesn’t).

As a response, I’m trying this out: PieFed accounts now have two profiles within them - one used for posting content and another (with no name, profile photo or bio, etc) for voting. PieFed federates content using the main profile most of the time but when sending votes to Mbin and Lemmy it uses the anonymous profile. The anonymous profile cannot be associated with its controlling account by anyone other than your PieFed instance admin(s). There is one and only one anonymous profile per account so it will still be possible to analyze voting patterns for abuse or manipulation.

ActivityPub geeks: the anonymous profile is a separate Actor with a different url. The Activity for the vote has its “actor” field set to the anonymous Actor url instead of the main Actor. PieFed provides all the usual url endpoints, WebFinger, etc for both actors but only provides user-provided PII for the main one.

That’s all it is. Pretty simple, really.

To enable the anonymous profile, go to https://piefed.social/user/settings and tick the ‘Vote privately’ checkbox. If you make a new account now it will have this ticked already.

This will be a bit controversial, for some. I’ll be listening to your feedback and here to answer any questions. Remember this is just an experiment which could be removed if it turns out to make things worse rather than better. I’ve done my best to think through the implications and side-effects but there could be things I missed. Let’s see how it goes.

  • echolalia@lemmy.ml
    link
    fedilink
    English
    arrow-up
    1
    ·
    edit-2
    4 months ago

    While not a perfect solution, this seems very smart. It’s a great mitigation tactic to try to keep user’s privacy intact.

    Seems to me there’s still routes to deanonymization:

    1. Pull posts that a user has posted or commented in
    2. Do an analysis of all actors in these posts. The poster’s voting actor will be over represented (if they act like I assume most users do. I upvote people I reply to etc)
    3. if the results aren’t immediately obvious, statistical analysis might reveal your target.

    Piefed is smaller than lemmy, right? So if only one targeted posting account is voting somewhat consistently in posts where few piefed users vote/post/view, you got your guy.

    Obviously this is way harder than just viewing votes. Not sure who would go to the trouble. But a deanonymization attack is still possible. Perhaps rotate the ids of the voting accounts periodically?

    • cabbage@piefed.social
      link
      fedilink
      English
      arrow-up
      1
      ·
      4 months ago

      It will never be foolproof for users coming from smaller instances, even with changing IDs. If you see a downvote coming from PieFed.social you already have it narrowed down to not too many users, and the rest you can probably infer based on who contributes to a given discussion.

      Still, I think it’s enough to be effective most of the time.