RE: Understanding Steem's Economic Flaw, Its Effects on the Network, and How to Fix It.
I gave a more detailed explanation of the benefits of slight superlinear in my reply to smooth above, but in a nutshell: forces all profitable voting behavior into the light (wack a mole with 1 hole), disincnetivize profit based spamming and micro voting, make it more difficult for bid bots to accurate place a price on votes and hence increase the cost of content indifferent behavior, helps make good curation more competitive in terms of returns etc.
The superlinear proposed can be probably as low as n^1.2, and even have a linear tail, so in case the added downvote incentives (proposed 10% free downvotes) aren't sufficient in deterring all 'pile on' behavior, the damage will still be in check.
I honestly believe the benefits we can enjoy most of the benefits of n^2 with only a fraction of the detriments in terms of inequality under n^~1.2.
I share your concern with the detriments of all of these measures. I understand that all else being equal, favoring larger stakeholder's vote is certainly an undesirable feature. Similarly, additional downvote incentives can potentially greatly increase the toxic behavior on here. However, bad it is to face an aggressive whale as a smaller stakeholder is going to be compounded when the gloves come off and he's given free downvote stakes. Also, lowering author incentives isn't ideal either and we wouldn't consider it if the current economy worked.
But look at what we're fighting: an economic system that rewards content indifferent voting behavior roughly 4x more than 'desirable' voting behavior. To win, we need to not just shrink that gap but allow content reflective voting to provide similar or even higher returns than vote selling/farming.
The idea is to come up with an alternative system that does just that while doing the least harm to the system in terms of added inequality, toxicity and lower proportion of author rewards. I think our proposal of n^1.3, 50% curation and 10% free additional downvotes is perhaps JUST enough to do that while introducing the least costs.
Some say just the latter 2 measures should be sufficient. I strongly believe we're suffering from n^2 PTSD. Saying the slight superlinear wouldn't work because n^2 failed is like saying that inflation doesn't work because 100% hyperinflation failed. We should consider the benefits/cost ratio of slight superlinear more carefully.
I'm not saying I'm completely opposed to "voting stake based slight superlinearity" as a step beyond the existing "temporal superlinearity" for curation rewards that I mentioned above, but I'd rather try smaller steps first, especially as I'm not certain that the existing superlinearity isn't sufficient. One annoyance for me is that I'm not even sure what the curve looks like for the temporal superlinearity, so it's hard to guess its impact.
50% curation rewards + a change to put downvoting on an even footing with upvoting seems like a simple first try, although I'm not really a big fan of downvoting as I think it has emotional effects beyond the simple math itself.
Also, once any change is made, I think it will take a month or two for people to modify their behavior to the new economics. At this point, I doubt most users have even optimized their voting for HF20 effects yet.
But despite that, I would certainly be happy to move now to a new version of the blockchain rules that supported some of these changes (50% curation being the one that seems "safest" to me).
I think I've read all of your replies here. I share some of your concerns but not all of your conclusions
Firstly I want to clearly spell out the problem: Under our current economy of linear and 25% curation, it is roughly 4x more financially rewarding to participate in content indifferent voting behavior than content reflective voting behavior. This has lead to a complete failure in our ability to function as a content discovery and rewards platform.
The solution therefore is as follows: We need to introduce a new economic scheme that incentivizes content reflective voting by rewarding it at least as much as content indifferent voting.
Higher curation seems like it'll at least be part of the solution. Note that curation % is sort of a soft parameter because the market itself can in many ways circumvent it and set their own curation %. For example, you can view bid bots as offering 80-110% to sellers, undercutting the 25% we've set. That being said, I don't know if there are market incentives to negotiate a lower curation % than the official one (I need more thought here)
Having said that, 50% curation is probably insufficient by itself to compete platform wide against vote farming. Much higher and we risk removing too much financial incentive for content creators. I know there are arguments suggesting we can try curation as high as 100% but lets try leaving a healthy dose to content creators for now and see if we can make up the other 30-50% by other means. I do however view curation as a 'cost' because the $ that finds its way back to the voter is a pointless use of inflation in and of itself. It's the competition of this money that will bring value, and for that to happen it has to be significant, which is why we propose 50%.
I share your concern about downvoting, far more than smooth and a lot of others. However toxic it has been in the past, with something like 10% free downvotes, it will be a lot worse. There are some larger stakeholders here who are somewhat pre disposed to what I would consider needlessly adversarial behavior, and at times with much smaller steemians. However suffocating it must feel for them to be on the wrong end of a downvote chokehold for sometimes weeks on end, it will likely be much more painful. We're all human, this place can take a mental toll on me even if downvotes are financially negligible, maybe you can relate, so I can imagine what it's like to be a smaller account getting pinned down without reprieve. However, even knowing all this I still relunctantly support something like 10% free/separate downvotes. As we remove the lost opportunity cost to casting a downvote, I'd imagine most downvotes would be exercised in good faith. We need an extra force to bring down the rewards of vote farming so that 50% curation rewards can be the most competitive form of returns. Only then can our platform succeed.
This leaves superlinear. I understand the inequality that comes with superlinear, nevertheless I'm surprised at the resistance this is getting compared to greater downvote incentives. I would strongly wager that at the ground level, the detriments of downvotes outlined above would be felt far harder than the inequality of something like n^1.2-1.3.
But why have them at all? Well there are many benefits including making it far more costly to price a vote (as value is now dependent on popularity), making it no longer profitable to micro farm spam posts (100 1% votes are way less rewarding than 1 100% vote), basically it brings all profitable behavior into the light. Notice that while 10% of downvotes are free, unlike upvotes which has curation, there is no direct economic incentive to cast them carefully and precisely. Under linear, if people decide to spam and farm comments instead of playing the game fairly, there isn't an incentive to spend the effort to track them down with your downvotes. But this is impossible with a bit of superlinear.
It basically patches up a loophole that would otherwise exist, as well as reward curation more, facilitate downvotes by drawing easy targets around abuse and makes it difficult for bid bots to accurately price votes. As for its costs, I'd say that n^1.2 is no where near as bad as n^2. The collaborative effort of large players colluding to farm mega votes is greatly disincentivzed by a much more modest curve (compared to n^2), as well as the looming threat of downvotes also being proposed. If I'm right on this, then we get to patch up our loopholes in spam farming and enjoy other benefits of superlinear at the cost of pretty modest inequality. I think it's worth it.
The changes are somewhat drastic by necessity because we're replacing a system that rewards something we don't want 4x as much as something we do. The measures all have their costs, higher curation means superficially more redundant use of inflation, more downvotes will be painful, and superlinear, however mild, is unfair to smaller players. But I'm going for the minimum amount of evil necessary to get us to a functioning content curation system.
I know this has always been a concern of yours but at some point we need to look at the game theory reality of the situation. I'm pretty convinced that ONLY enhanced downvoting (at least a component, if not the only change) can possibly solve the problem when it comes to content agnostic reward-extraction voting. I know that is a strong statement but I'm very confident of it, and I will try to present the logical argument here.
First, we need to recognize that anything else other than downvoting can easily be turned into a vehicle for converting votes into profit in a content-agnostic manner. Curation rewards and superlinear curves don't really reward quality or value add. Rather, they reward concentration (voting for what other people vote for). By either being a very large stakeholder or working together with other stakeholders through trails, delegation, teams, games, side payments, or even reciprocal social convention, it is always possible to arrange for concentration that mutually benefits all of the participants on a content-agnostic basis. Since downvotes don't generate or direct rewards but merely scatter rewards to the rest of the community, they are essentially immune to this effect. There are really two reasons to downvote: 1) malicious/trolling, which is unfortunate but doesn't have a systemic effect on reward distribution; and 2) altruistic/community-motivated downvoting of over-rewarded content, which does have a (positive) systemic effect on reward distribution, but must be free or extremely cheap if you want it to happen on a significant scale (which, currently, it is not)
I think if we want a quality- or value-based voting system we have to take the bitter medicine of downvotes that are cheap, free, and even encouraged (because this is the only way you get people to contribute, at scale, to a social good that relies on altruism). Ideally, we should absolutely work to minimize the emotional cost of it however we can, of course, but that must be a secondary consideration if we want a rewarding system that works. Failing that, I believe Steem will need to just pivot away from voted rewards altogether at some point, or continue with a useless failed/failing system that does not contribute to its success in the crowded and competitive markets in which it operates.
It's a simple fact that superlinear curves reward concentration, as you say. And naturally this can be exploited. The obvious exploitation method is pre-agreements of one sort or another to vote on the same posts in a way that generates reciprocal shares of the resulting rewards over time. If a sufficient amount of stake becomes involved in such arrangements, then the reward system by itself no longer incentivizes upvoting based on content.
Note that this doesn't necessarily guarantee bad curation by such groups. The people acting in such group do still have a reasonable financial motivate for good curation if we assume that good curation makes the coin more attractive to holders. If the group is curating badly, I think it's reasonable that competing groups will arise that "do a better job".
And I think it's also reasonable to assume that a certain amount of people will decide that they would rather vote the way they like, rather than join such a group, in which case they will lessen the rewards of joining such groups.
I don't see how the mere existence of cheaper downvoting necessarily changes the economics above. You've already described it as "altruistic/community-motivated" downvoting. So altruism or like-mindendess is already assumed as a force that exists on the platform (I agree with this assessment, btw).
But altruistic "voting your conscience" also redistributes the rewards, just like downvoting does. It's true that voting "can" be directed towards one's own posts, but if it's altruistic, it certainly doesn't have to be.
The best argument for downvoting from a curation perspective is only that it allows you to actively distribute the rewards away from a particularly bad post that is being upvoted for selfish reasons.
In a system with lots of voters and posts and with a more distributed stake, I'm not sure downvoting would have much benefit at all.
I think in such system, the type of bad "self-voted" post I'm talking about would never get enough rshares behind it to get much value. The only way I could see it happening would be the rise of a mutual-voting group of the type I described at the beginning and the only real counter for such groups is a sufficient group of people who vote differently.
The difference is that this has a huge opportunity cost compared to voting to maximize your own reward (either directly or via some scheme). Since the cost is very high, it is unlikely to see such altruism persist. Sure, not everyone will maximize their own reward, but when the incentives are misaligned in this manner you expect the economy to shift more and more in that direction over time, which is exactly what has happened. I don't believe this has anything to do with stake distribution or the number of voters or any such thing. It is more that it is baked into economic design.
By contrast, once the high cost of downvotes themselves is removed, good downvoting vs. no downvoting vs. bad downvoting does not have a high opportunity cost (in effect virtually none), so it is far more likely that altrusim will succeed. If people had to pay money (or forgo a significant payment) every time they edited wikipedia, hardly anyone would ever do it (in fact, those who did would be more likely to be promoting some sort of manipulative agenda and often damaging the end product). But because editing wikipedia is very cheap/free and frictionless, you do get altruism on a large scale. Crowdsourcing works, but not if you expect people to pay a lot to participate. The only way to get this with Steem is by crowdsourcing downvotes.
This not just a question of downvoting being used actively to 'fight abuse', which is the traditional way we have looked at it. It is a profoundly different setup of the incentives, which can't be accomplished by upvoting, because the economic 'pull' of voting to direct rewards toward oneself and/or one or more collaborators is always there with upvotes. With downvotes it is not.
Such groups would certainly arise if there were significant money at stake, probably cleverly packaged into a game, challenge, service, meme, corporation, investment fund, etc. That is the nature of human creativity.
In any case, we can't simply wish ourselves into a system with evenly distributed stake and billions of users either, even if that would somehow manage to work. We have to need to have economic rules which are robust to real world conditions. Which means downvotes.
Well, no. A superlinear reward system never, in and of itself, incentivizes upvoting based on content. Structurally it only incentivizes concentration. And concentration that rewards the voter is incentivized most of all. Content is actually irrelevant unless there is a large enough portion of the stake voting on an altrustic basis, and as I've argued above, it is very unlikely to see altruistic upvoters persist on a scaleable, sustained basis. (Things might start out that way, but they will surely devolve over time.) The only real hope we have is altruistic downvoters.
Please give this some more thought. It took me the better part of two years to finally reach the conclusion that downvotes are essential (and not just theoretically-possible expensive downvotes that are hardly ever used, the way we have now), but I'm now quite confident it is correct.
We could expect bid bots to decline rewards, this still trends the post but leaves those of us that don't buy our trophies alone instead of subsidizing proof of wallet.
The bots get their curation rewards upfront in the form of payment for services.
This should lower the price of hitting trending, and clearly labels ads.
I think we need a better assessment of how much of this is still going on. I've heard that it has been significantly reduced already with the changes in HF20 (which did indeed introduce a slight degree of non-linearity on the low end, in addition to the bandwidth->RC change; both probably reduced or eliminated spam profit on the low end).
This will already happen with the switch to 50% curation. Curation rewards are non-linear and I doubt it will be feasible for bid bots to continue to ignore that when curation is 50%.
Narcissism of small differences is a name of a study, I extended it, perhaps too liberally, to apply here. Apologies for the confusion, I certainly didn't intend to accuse you of being a narcissist.
I agree that hf20 likely reduced micro vote farming. But I suggested it as a precautionary measure in anticipation of behavior that may come with the other proposed changes. Assuming things are otherwise working fairly well and curation rewards becomes the dominant form of profit maximization, consistent losers in curation may turn to such behavior to hide vote farming. It's not as much of an issue now partly due to hf20 and partly because under current economics, one can do it openly. If things start working and people start downvoting open farming, this spamming fuckery would likely increase. As I mentioned before, due to the lack of direct rewards for good downvoting, we shouldn't expect immense effort for it to be done very well. Forcing profitable votes into the light will be far more important in the future as it'll improve downvoting immensely. I want to close this exploit tightly if the price isn't too steep.
I admit overlooked the effects of non linear curation. I did however, factor into consideration that higher curation itself, irrespective of the curve, will directly introduce significant price uncertainty to bid bots. As they run on a timer dependent on voting power regen, and bid transactions are announced beforehand, they'll likely be at a considerable disadvantage with higher curation. Nevertheless the effects of superlinear on this will be additive and very strong.
We have a laughably poor economy that rewards content indifferent behavior 4 times more than content reflective behavior. We want to introduce changes to make the latter to be at least roughly as lucrative as the former and hopefully have value adding voting behavior be dominant on the system. Every measure I can think of (including the 3 proposed) has it's downsides. So the idea is to use just enough force to make the necessary behavioral change and no more.
If we cut something on one end, we'd likely need to up our dosage of another poison elsewhere. For example, if we dismiss slight superlinear, then stakeholders may attempt to circumvent the intended way the game is to be played by spamming more posts and self voting the ones with the least potential curation rewards 'stolen' by others after 15 minutes. Now downvote incentives will have to do more of the heavy lifting here and would likely need to be higher than if we took some of that superlinear poison. As mentioned, in many cases, downvotes are a less consistent and softer counter to undesired behavior than superlinear, and of course it has it's own downsides.
My prediction is that the benefits of some superlinear (maybe as low as n^1.2) would considerably outweigh its cost in inequality, and should be part of the changes. I can of course understand if, after careful consideration, others such as yourself do not think this is the case. What I can't understand is how such a fundamental problem in the economic of Steem rewards which has lead to this platform being completely undermined has not been addressed in over a year. I don't even think most people, including those at the top, have a correct diagnosis of the problem; the confusion and inefficiency is truly depressing, especially as I believe this problem can be remedied. This is why I am very grateful for your help smooth, despite our small differences in opinion.
I know the reference and was simply building on it to make a point! I was not personally offended but I appreciate the clarification.
I would agree with this, but we simply don't know the necessary dosage overall. If you have done some sort of research to support that, quantitatively, your prescriptions in each of the three dimensions you propose as well as the combined effect result in some optimal outcome, you haven't shown it. As far as I can tell, you along with everyone are just guessing when it comes to the question of 'dosage'.
In that sense I would also agree with @blocktrades that it may be sufficient to be conservative in making too many changes at once and try just increasing curation first (even without downvote changes, which as you know by now are my personal view on the most important dimension here, by very, very far) and see what happens. If that doesn't produce the desired effect then we can consider how to up the dose in the most effective manner.
HF20 made some very modest changes and it has apparently had an observable effect in terms of dust farming and spam. That's a good thing, and not necessarily a bad approach to continue (making incremental changes, one or a small number at a time, so we can observe their effects).
Likewise, and I absolutely agree that the lack of any sort of action on the clear failings (other than the very modest changes in HF20 which took an absurd 1.5 years to be deployed after being designed) is a travesty.
As a followup answer to my concern about superlinearity, please see my answer to @smooth in this thread. Super-linearity inherently increases the power of voting groups like the one I describe there, making it more difficult for them to be combatted by "dissenting voters" with lesser stake if those dissenters don't similarly band together. With a linear curve, such groups don't have to unify their voting to fight such tactics.