There has been much discussion of these issues recently. It can become very heated, especially when behaviours are given a moral interpretation. What I seek to do here is to clarify a few things and then concentrate on encoding rules that encourage behaviour that is beneficial to the whole reward pool and that reward people who have different strategies.
The Steemit Game
Steemit is a game; it has rules that reward certain types of behaviour. Because there is real money at the end, the game can become very serious. But it is still a game and therefore can be analysed using game theory. And recent game theory is quite clear on three important matters: that absolute self-interest from all players does not achieve the best outcome for everybody, that cooperation can lead to higher growth than self-interest; secondly, that there is no way to guarantee that everybody follows a path of cooperation, that there is no way to eradicate behaviour that may appear anti-social; and lastly, that the best that can be done to achieve an optimal ecosystem is to limit behaviour that is considered sub-optimal. This must be achieved through the rules of the game and by changing those rules if it is deemed necessary.
Now, in the language of Steemit, we are talking about rampant self-voting and voting within a small clique, what I will call here voting-circles. These are all currently allowed within the rules of the game. However, a very simple thought experiment is all that is required to see if this is optimal behaviour or not: what would happen if everybody, absolutely everybody, did exactly the same thing? It is important to think about this; if these strategies are so profitable why doesn't everybody use them?
The calls for banning things such as self-voting are understandable but ultimately futile. Anybody can open two accounts and cross-vote between the two; a ban on self-voting is so easy to circumvent that the system would then need an array of security-bots to analyse voting patterns and discover voting-rings. All of which can be done, but it is wasteful energy; indeed, even any punishment would add to the waste. I think it is better to grow together but with a new-found balance.
Some New Voting Rules
The tables below show the effects of some new rules that target both self-voting and voting-circles. Because of the way self-voting bans can be got around, the two classes of voting behaviour are put together under the same rules. I think these are easy to encode if there is the will to implement them. I present a range of options just to show how the mathematics works; sometimes what appear to be large changes are not so large in the long term, and compounding is a very powerful multiplier that must be taken into account.
These proposed rules are in addition to the standard voting power depletion at 2% per vote. I have not, in this article, factored the two together. This is, firstly, not to confuse two issues and, secondly, it is a simple matter of multiplying together the numbers in the Rewards column with any of the numbers in the same row that correspond to each new rule. I can write a further article if this level of analysis is required.
| Vote | Reward | Sum Rewards | Reduce 10% | Sum R-10% | Reduce 20% | Sum R-20% | Inv V | Sum Inv-V | |||
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 1.00 | 1.00 | 1.00 |
1.00 | 1.00 | 1.00 | 1.0000 | 1.0000 | |||
| 2 | 0.98 | 1.98 | 0.90 | 1.90 | 0.80 | 1.80 | 0.5000 | 1.5000 | |||
| 3 | 0.96 | 2.94 | 0.81 | 2.71 | 0.64 | 2.44 | 0.3333 | 1.8333 | |||
| 4 | 0.94 | 3.88 | 0.73 | 3.44 | 0.51 | 2.95 | 0.2500 | 2.0833 | |||
| 5 | 0.92 | 4.80 | 0.66 | 4.10 | 0.41 | 3.36 | 0.2000 | 2.2833 | |||
| 6 | 0.90 | 5.71 | 0.59 | 4.69 | 0.33 | 3.69 | 0.1667 | 2.4500 | |||
| 7 | 0.89 | 6.59 | 0.53 | 5.22 | 0.26 | 3.95 | 0.1429 | 2.5929 | |||
| 8 | 0.87 | 7.46 | 0.48 | 5.70 | 0.21 | 4.16 | 0.1250 | 2.7179 | |||
| 9 | 0.85 | 8.31 | 0.43 | 6.13 | 0.17 | 4.33 | 0.1111 | 2.8290 | |||
| 10 | 0.83 | 9.15 | 0.39 | 6.51 | 0.13 | 4.46 | 0.1000 | 2.9290 | |||
| 11 | 0.82 | 9.96 | 0.35 | 6.86 | 0.11 | 4.57 | 0.0909 | 3.0199 | |||
| daily % | 0.1993 | 0.1372 | 0.0914 | 0.0604 | |||||||
| year % | 72.73 | 50.09 | 33.37 | 22.05 | |||||||
| CAR % | 206.8 | 165.0 | 139.6 | 124.7 |
Reading the Table
The first column shows the number of votes. At a depletion rate of 2% per vote and a replenishing rate of 20% per 24 hours, one can cast 11 votes per day. The "10 votes per day" meme is merely an approximation; the exact number is 11.
The next two columns labelled "Rewards" amd "Sum Rewards" shows the current algorithm and the effect of casting 11 self-votes starting at 100% voting power. I have set the first vote as yielding a reward of $1.00 purely to make the numbers easy to read. At the moment, using 100% voting power and 100% voting effectiveness, a vote is worth approximately 0.02%. This can change and does change, but at the moment it is a valid approximation for the purposes of making comparisons.
But what this means is that a $1.00 reward on a single vote requires approximately 5,000 SP. Bear this in mind when scaling up or down to your own Steem Power. Note that 11 self-votes yields almost exactly 0.2% per day. If all such rewards are cashed out this yields almost 73% per annum; however, if all rewards are powered up (and thereby compounded) the compounded annual return (CAR) is 207%.
New Algorithms
The next columns show three other algorithms that could be implemented to reduce the effects of self-voting and voting-rings. The first one shows a decrease of 10% per self-vote ("Reduce 10" or "R-10") and the next one shows a decrease of 20% per self-vote ("Reduce 20" or "R-20"). Notice how the mathematics works; changing from the current R-2% to R-20% does not drop the daily reward by a factor of 10 but only by a factor of about 2 - dropping from 0.2% daily to 0.09%, with an annual compounded return of 140%.
This is still not a bad return on investment and precisely the aim of adding such a rule to the current algorithm so that those who wish to use Steem as a resource for mining can do so without jeopardising the whole pool for those who wish to be rewarded for their social content.
The last column deviates from a percentage-drop algorithm and is an inverse-proportion algorithm so that the Rth self-vote is worth 100%/R. Notice that the rewards seem to drop much more steeply, but the power of compounding means that the daily rate for 11 votes is still 0.06%, which is 22% annualised and with a CAR of 125%.
As I already said, one thing I haven't done, which I could do in another article, is to combine the current R-2% rule with one or more of the above rules. I don't think we should change the current rule as that works globally. One of the above rules should be an additional change in voting effectiveness.
How Would a New Algorithm be Triggered
Any new rule should be as simple and efficient as possible to achieve its aims. Turning mathematical formulas into computer codes requires looking at the speed of such codes. This is just a proposal, but I have thought through the various ways that other triggers could be circumvented. Remember that the overall aim is to treat self-votes and voting-rings as essentially the same phenomenon - it is the voting for the same person repeatedly, whether it is oneself or a sock-puppet or another ring-member. We do not need different rules for each of these scenarios - one should be enough.
So, imagine you are Alice and click on an upvote to a post or comment by Bob. The code scans all the votes you have cast in the previous 24 hours and counts how many times you have vote for Bob. It then uses that number to calculate your voting power for that vote. For example, if this is Alice's 4th vote for Bob within 24 hours, and we are using the R-20% algorithm, the vote is now worth 51% of whatever it would have been had it been the first vote. Exactly the same calculation would be made if this was Alice's 4th vote for herself.
Yes, this takes a few more steps, but we are talking about a system designed to be scalable to thousands of times what it is now. This simple counting algorithm also resolves one strategy whereby Alice votes for herself at 100%, then votes for 20 other people at 1% each, then votes for herself again at 100%. Only bots can vote many hundreds of times a day - a human will rarely go beyond 60 or so.
Discussion
I welcome all discussions about these proposals. These have more details than when I first mentioned them in paragraph 13 of Ideas for Future Rule Changes - Voting, Earnings, Maximum Social Benefits - a Discussion Document. That document is still worth reading as it sets out some ideas as to how I have arrived at these new rules, especially what game theory tells us is and isn't possible to achieve. To reiterate, it is not possible to completely eradicate those who see Steemit as a mining operation; what we can do is ensure that their mining activities do not suck out all the resources.
Mining towns were traditionally boom towns. The wealth generated from mining meant that new jobs were created, new businesses were born servicing the workers, who then gave rise to families and an enlarged community. Some of the wealth was sold outside the community, but there was enough left to make many others either wealthy themselves or at least comfortable. When a mine runs dry, we are left with a ghost-town.
