Abuse mitigation plans for Blurt: A discussion

I have been asked to post this double-header by @jacobgadikian as it is an important issue to be discussed and is possibly easier to compare methods on one post than having a discussion take place in parallel across two or more articles.

This does include my own edits, in the hope of making both positions as clear as possible, so if any further amendments are needed, or I have accidentally mis-represented a position, then it shall be edited as soon as possible.

To make further discussions easier, I have just labelled them as Plan R and Plan J. So, without further rambling, here are the current ideas.

---

Plan R

Firstly, this plan is the first of a 3-stage process leading to the development of a fully-automated algorithm that aims to reduce rewards to a range of posters and voters who abuse the reward system. Some of this work has been done but much of it involves on-going research to develop metrics that are both sensitive and appropriate to mitigating such behaviour.

I have seen this work, albeit with a final human check for false signals, but if the effect is gradual instead of binary, this gives users the chance to change their behaviour and hence avoid even the appearance of being bad actors.

On a more general note, we really need to move away from the idea that there is “one reward curve”. That idea has not worked at all and will never work, the reason being that one formula will have an obvious optimisation strategy and it is not at all clear that such a strategy is good for the whole economy. We should look forward to possibly two or three different curves that affect user behaviour in different ways. If designed properly, then we can find an optimal user strategy that is in line with the growth of a good economy.

We already have such a mixture of curves within the 5-minute “reverse-auction” but it is a mere microcosm of the whole system. However, it is proof that two curves are better than one! (I shall write about this in the near future but don’t wish to go off on a long tangent.)

Enough of the preamble, so what is Plan R? Given that Blurt is an open system, with pseudo-anonymous accounts and a command line interface (CLI), any abuse mitigation has to be done at the blockchain level and not within the front-ends or apps that most users engage with. The key aim is to make abuse less and less profitable and hence have a diminishing effect on the whole Blurt ecosystem. By the necessity of time, this is a human-led system with algorithmic assistance.

The fundamental aim is to use the existing Warden roles to trigger a vote from a dedicated account, such as @abusecontrol, so that the rewards algorithm picks this up and decreases that poster’s rewards. This can quite easily be done by inserting some extra lines at the point at which the claims of a post are calculated from the total rshares from voters. In general, this is also the point where any future “curves” can be added.

Wardens are already heavily involved in finding bad actors, for both content-abuse and vote-rings, so it strikes me that they are in the best position to carry on such work. Also, having some automated assistance will relieve them of some of the repetitive work of finding the same accounts every day.

One or more Wardens would have access to the @abusecontrol account, which can be used both manually and as an auto-voter. This effectively means that a “blacklist” is created within that auto-voter so that persistent abusers will have their rewards reduced for every post. The voter will also be able to scan both posts and comments. The blockchain then just needs to check whether that abuse-account has voted on a post and adjust the rewards accordingly.

The actual level of reward-reduction is something up for discussion. The principle here is that rewards should be reduced to a level whereby the user can earn more by being a good actor than by being a bad actor. I therefore suggest a reduction of 60%, so they would still get 40% of their ill-gotten gains! I have had some discussions with @megadrive about this and there are possible refinements, such as a reduction to even 0% for persistent abusers. Such numbers can be decided as and when this becomes coded. I think the aim is to not necessarily drive stakeholders away but rather to change their behaviour.

There are further refinements needed so that the automated trigger makes intelligent votes, but these are issues outside the blockchain itself. Just to repeat, this human-triggered semi-automated system is designed to be fairly quick to implement, after which I can work on designing a fully-automated system that works within the blockchain.

---

Plan J

This proposal is for the defense of the reward pool. I believe that such a defence should be the responsibility of Witnesses.

The idea here is that accounts that abuse the reward pool shall have their post rewards disabled - somewhat like the “Decline Payout” option currently available in the post creation Advanced Settings, but enforced by the blockchain.

The witnesses shall keep their own blacklist of such abusers and an algorithm written so that a username is added to a master blacklist once 2/3rds of witnesses include it in their personal list.

Note that this targets the posters rather than the voters, so that the stakeholders that actually create the rewards are able to change their behaviour. The curation rewards are now relatively high so that a large stake can earn significant BLURT by voting on better content rather than aiming to accumulate both author and curation rewards from what can only be described as shit-posts.

The broader community of Blurters is obviously sick of seeing garbage posts pumped into Trending by a circle-jerking clique of stakeholders. I am not suggesting the ability to remove the right to vote, and hence earn, but I do want to see the right to abuse terminated. I thus think it may even be possible to include the wider community in this process, with a direct line to be able to add a username to a warning-list so that Wardens and Witnesses can be alerted more quickly.

Placing the onus on Witnesses gives a clear indication that any abuse of the economy will not be tolerated at both social and management levels. In the Blurt DPOS system, Witnesses have both network and governance roles. I appreciate this will mean some extra work, but it is also entirely possible to form witness-teams so that one individual may focus on the network while another is assigned governance responsibilities. The potential issue of Witnesses themselves abusing this responsibility can be solved by voting for better Witnesses.

I believe that once in place, the social network effect will deter users from bad behaviour. I also believe that the list is easier to maintain than people might think. I accept there may be refinements to the process of blacklisting, but what I want to see is the social expectation of quality and a mechanism for enforcing quality.

---

So, let's open this to further discussion...

PS. I was going to add @jacobgadikian as a beneficiary but see this isn't available - looks like another topic for discussion!