4 ms·

> For example, if the #1 person believes something is being said that is really damaging in some way, they could put at stake the entirety of their karma and do

> For example, if the #1 person believes something is being said that is really damaging in some way, they could put at stake the entirety of their karma and downvote it, which would pretty much obliterate the comment. We are building a P2P network with subjective moderation [0], so while we don’t know what works yet (we’re experimenting) I can at least tell you why we’ve found this is suboptimal. The issue is that this creates an even more direct USD/upvote market in which users are incentivised to the max to harvest as much karma as possible (with cat posts) and then sell them to the highest bidder, which would invariably be people who have a lot to gain from a certain piece of information being hidden. In essence, if you make karma spendable by more than one per post, you up it’s monetary value 100x, and given that even at 100x it’s worth so much more to PR teams and to not many else, all you will ever see is going to be propaganda on your platform’s front page. [0] https://getaether.net https://getaether.net

Y_Y 5y ago
You've probably thought of this, but if you can associate votes to identities then a classic way to mitigate this behaviour is quadratic weighting, where e.g. your nth vote costs n^2, so a censor or propagandist needs to pay 1e4 counteract the single votes of 100 casual observers.
- rolleiflex 5y ago
  Yeah, quadratic voting is one of the potential solutions to this, but ultimately that removes the ‘fraud load’ by mostly moving it to another place, to number of accounts. That is arguably worse because it incentivises the malicious users to create as many accounts as there are votes so as to not be subject to quadratic voting (1 vote = 1 energy point equivalency only holding at one vote), and it’s substantially harder to deal with user account fraud than vote fraud. The goal is to minimise the fraud load and make whatever fraud load you have applied against the strongest parts of your system. It’s like designing a castle except with incentives - it has to fail gracefully and still be functionally successful even in failure.
  - mjburgess 5y ago
    Adding other factors into this system might help. What about: weighting a vote by the "degree of separation" of the voter? (ie., friends count more) hyperbolic discounting of karma acquisition limiting the number of voters per item new accounts can't vote until achived a network & karam status (eg., shares, by friends, are upvoted). accounts which primarily upvote/downvote outside of their network are ignored etc. I think this adds up to saying, "small networks of voters" are the primary type of community; and big-network effects are largely discounted. Where things are popular, they are because many different active communities appreciate them.
    - rolleiflex 5y ago
      Yes, these are all different kinds of logic that can be executed, and none of them are ‘right’, it just depends on personal preference and situation. We call that user-selectability ‘subjective moderation’, but it’s basically a client-side compiler for a people-based raw data feed.
  - naasking 5y ago
    > ultimately that removes the ‘fraud load’ by mostly moving it to another place, to number of accounts. That is arguably worse because it incentivises the malicious users to create as many account Arguably the common social media policy of not having your account tied to a unique identity needs to change. It permits all sorts of problematic incentives like you describe, and which lead to bot accounts and more. Of course, your identity doesn't need to be disclosed in your profile for when anonymity is important.

Y_Y 5y ago

You've probably thought of this, but if you can associate votes to identities then a classic way to mitigate this behaviour is quadratic weighting, where e.g. your nth vote costs n^2, so a censor or propagandist needs to pay 1e4 counteract the single votes of 100 casual observers.

rolleiflex 5y ago
Yeah, quadratic voting is one of the potential solutions to this, but ultimately that removes the ‘fraud load’ by mostly moving it to another place, to number of accounts. That is arguably worse because it incentivises the malicious users to create as many accounts as there are votes so as to not be subject to quadratic voting (1 vote = 1 energy point equivalency only holding at one vote), and it’s substantially harder to deal with user account fraud than vote fraud. The goal is to minimise the fraud load and make whatever fraud load you have applied against the strongest parts of your system. It’s like designing a castle except with incentives - it has to fail gracefully and still be functionally successful even in failure.
- mjburgess 5y ago
  Adding other factors into this system might help. What about: weighting a vote by the "degree of separation" of the voter? (ie., friends count more) hyperbolic discounting of karma acquisition limiting the number of voters per item new accounts can't vote until achived a network & karam status (eg., shares, by friends, are upvoted). accounts which primarily upvote/downvote outside of their network are ignored etc. I think this adds up to saying, "small networks of voters" are the primary type of community; and big-network effects are largely discounted. Where things are popular, they are because many different active communities appreciate them.
  - rolleiflex 5y ago
    Yes, these are all different kinds of logic that can be executed, and none of them are ‘right’, it just depends on personal preference and situation. We call that user-selectability ‘subjective moderation’, but it’s basically a client-side compiler for a people-based raw data feed.
- naasking 5y ago
  > ultimately that removes the ‘fraud load’ by mostly moving it to another place, to number of accounts. That is arguably worse because it incentivises the malicious users to create as many account Arguably the common social media policy of not having your account tied to a unique identity needs to change. It permits all sorts of problematic incentives like you describe, and which lead to bot accounts and more. Of course, your identity doesn't need to be disclosed in your profile for when anonymity is important.