Scalability, Content Quality and Centralization - An open discussion on an incentivized downvote system

in steemit •  2 years ago

With the influx of new users in the past month, it has become progressively more difficult for new content to be seen by curators of significant weight. Some of it is to be expected: with growing userbase comes increased competition. In this post, I highlight some of the issues new users might be facing and propose a possible implementation of an incentivized downvote system, which I believe directly improve the situation.


The issue

In current system an “upvote only” culture is promoted, due to the lack of rewards and the negative stigma behind downvoting (flagging). This creates an environment in which more visibility almost certainly equates to higher payout. It can seem like a win-win situation, but taking a step back can help us to see how it objectively affects the Steemit ecosystem. The following statement can’t be emphasised enough:

In a zero-sum game in which a fixed amount of Steem is created every day, an upvote for one article is equivalent to downvoting every other article (of lower weight).

As a result, on the present platform, each upvote shouldn’t be merely viewed as a user’s appreciation for a given post, but rather as a statement signifying their preference for said content over all other. This is a glaring issue, as no curator should be able to pre-judge and lower access of content they haven’t seen to others on the website.

Since the money used as rewards isn’t taken directly from the voter's wallet, it may feel to them as though they’re handing out free cash with minimal effort and no consequences for unwise curation. Looking at the bigger picture, however, it is clear that everyone pays for such actions in several ways: lower content quality, limited scalability and centralization (content/payout/curation power) come to mind. All of these can contribute to the decline of the platform, or at least impact its growth in a significant way.

Content quality
Inflated payout for content A entails that some other content B is underpaid. This means that lower quality content can make it to the trending/hot page for everyone to see. Similarly, this leads to a ranking system in which the creator of content B doesn’t get their deserved visibility, one that accurately reflects the quality of their content.

Visibility and the high disparity in payouts has been a hot topic in the past couple of weeks. These are real problems that need our attention as a community. Here’s a discussion between @teamsteem, @fyrstikken and @stellabelle in which they briefly talk about the issue. Here’s @dan acknowledging the visibility issue.

If you’re not convinced yet, there are two very good examples of articles that were posted twice. After going unnoticed at first, they were reposted by another user who recognised their value. The first one was featured in @stellabelle’s secret writer (she discusses it here) and the second one is mentioned in this post by @coinbitgold. These examples either demonstrate the high dependence on luck or the poor performance of the ranking system in assessing value. Either way, the situation is far from optimal.

Centralization (of content, payout and curation power)

The current system promotes name recognition more than content quality, which has a direct impact on centralization. Those who make it to the top trending page are much more likely to get there again, even with subsequent lower quality articles. Lower diversity leads to content centralization and an increase in their SP holdings. In turn, this increases their influence in selecting new content. For instance, a community based around cryptocurrency is more likely to upvote cryptocurrency content, leading to a much higher percentage of cryptocurrency related articles than there is on any other social media. This can be positive for building and reinforcing ideas in a small community of experts, but not when trying to appeal to a larger population of diverse interests.

Scalability

How important exactly is platform growth? If you read the Whitepaper, you'll recall Metcalfe's law: the value of a telecommunications network is proportional to the square of the number of connected users of the system (n^2). This was shown to hold for large social networks like Facebook and Tencent. Read more here..

Metcalfe's law doesn't quite accurately describe Steemit - it's still in it's infancy- but here’s what it looked like as of last week. (Data taken from Steemd and here)

Assuming those numbers hold in the long run, we have

Market Cap ≈ 0.22 * n^2

This means that a 10% increase in number of users should lead to an increase in Market capital of approximately 21% (110%^2). Looking at it this way, large SP holders should spend most of their efforts trying to make the platform more welcoming to new users. Reducing the payout gap would certainly help towards that.

With the current voting system, most of the daily mined Steem for content creation and curation gets distributed among users on the top trending/hot pages. This is to the detriment of the majority of posts (close to 4000/day) that get close to nothing. I invite you to listen to this excerpt from Free Talk Live in which @ftlian discusses his first experience and perception of the trending page as a new user.

At the time of this post, 87% of all articles have received less than $1 for their contribution to the site. A good percentage of it is likely spam, but it still shows that, for most new users, Steemit is far from profitable. I’m afraid that if it is to grow exponentially, so will this issue, driving potential new users to other, more lucrative, alternatives.

Platforms like Akasha and Yours will soon be competing with Steemit. They're likely going to be more than 60k users behind it at their launch, but they'll catch up quickly if newcomers realize they have more influence over there. That influence could be drastically transformed by reducing inflated payouts.


The proposed solution

I hope the above remarks have convinced you that changes are needed. It should be mentioned that considerable progress has already been made aimed at solving some of these issues. For example, the recent modifications to the payout delay have been successful in reducing the average time a post stays on the trending page. This still only addresses part of the problem. After taking time a reading a post, curators currently have two choices to influence the payout of an article: upvoting, if content is worth more than the estimated payout, or flagging, for spam or dishonest content. But that leaves a place for a third option: one that would reduce the payout when it’s inflated, independent on the voter’s liking or disliking of the content.

A user’s choice to upvote or downvote an article should be based on its content, in addition to what they believe its value to be compared to the estimated payout. The option to flag should remain as a mean to decrease someone’s reputation and prevent spamming or dishonest content.

This might all sound like a stretch, and maybe you’re not aware of the influence that downvoting can have, but some downvoting whale action has already been happening. See @smooth’s comment on this announcement post:

To be clear, I’m not merely suggesting here that downvotes should be used to discourage bad content, but rather advocating the use of downvotes on good content that may have an inflated reward.
There’s nothing inherently bad with promoting an upvote only culture on a platform of complete information, but the problem is just that: the medium we use to access that information favors certain posts, namely, those that were heavily upvoted early after their creation.
In a context in which an important disparity exists in exposure to different blogs, users shouldn’t simply upvote what they find interesting, they should vote in a way that brings the payout closer to what they deem to be the real value of a certain post. As such, increased exposure would equate to a fairer payout, as opposed to an inflated one as discussed above.

The problem with simply adding a downvote button is that it’ll have little use if it isn’t incentivized in the same way that upvoting is. An ideal system would incentivize both positive and negative votes, so long as their net effect is to pay the content creator a fair price for their contribution. The question then becomes: in which situations should downvotes be rewarded?


The implementation

Assuming a free market of positive and negative votes: the more people partake in the curation process, the more representative it is of its quality. Going by this, we refer to the fair value of a given post by its payout after the first (12h) or second delay (30 days) depending on the period in which voting takes place. The curation rewards would then be attributed to those who correctly predicted the final payout by voting accordingly. Here’s an example:

@dan posts an update, announcing a new feature on Steemit.com. Browsing through the “new” section, @wingz upvotes only a few minutes after the creation of the post when it’s still at $9.40, expecting it to go higher. A few hours later, @ned wakes up and looks at the top trending posts. It’s a nice, sunny day and he sees @dan’s post at $6,245.33. Feeling generous, he upvotes his partner’s article. Finally, @smooth finds the post at $7,129.21. He evaluates that its current payout estimate is higher than its fair value, so he downvotes it. 12 hours after its creation, the payout (fair value) settles at $4,124.33.

Here, @wingz and @smooth stand to receive a curation reward for correctly assessing the fair value of the post. On the other hand, @ned would receive no reward for upvoting purely based on his mood and the author’s name, not considering the actual content quality.

For the proposition described above to be implemented, no major changes in the weighting system would be necessary. Only the estimated payout before (EPB) and after (EPA) a vote would need to be recorded. The current rules for calculating curation rewards (based on weights) can be summarized as the following:

  • Curation rewards scale with VESTS (similar to SP, see the conversion ratio)
  • Rewards scale with voting power, which decreases after each upvote before slowly growing back towards 100%
  • Curation rewards scale with the weight given to your vote if it is greater than 0 (0 to 100%) - No reward is given for a negative weight (downvote).
  • Given 2 users with the same amount of Steem Power, the earlier one appears on the voting list (ordered by time of vote), the larger their reward will be.
  • Votes casted t minutes after the creation of a post where t<30 minutes will have their rewards reduced by (t-30)/30*100 percent.

EPB and EPA would come into play in situations where the fair value lies between the two. In those cases, the voter’s share of the curation rewards could be adjusted according to their proportional contribution in correctly valuing content. The reward per vote would then be given out according to those relations:




And then

This type of payout structure would have the additional advantage of rapidly counterbalancing the effect of bots that upvote users uniquely based on the payout received for their past content. Curators who read the post could better assess value and benefit from downvoting a horde of bots that inflated a payout past its fair value.

While this doesn’t directly address problems due to visibility, I believe it would help in 2 ways. First, an adjustment in prices on the trending page would automatically grant more relative curating power to those who do not participate in inflating payouts. A price distribution that reflects the comparatively narrow range of content quality would allow for small price variations to change the ranking order of a post significantly. Secondly, posts that suffer from low visibility would have their curation rewards increased by redistribution of money that is currently being given to more visible posts for no valid reason.


I’m sure that lots of thought has been put behind the economics of Steem by its creators. I hope I was successful in highlighting the current issues on the platform and showing how an incentivized downvoting system could make the situation better. Steem is still in beta and I think it’s important for the community to challenge established ideas if other alternatives could improve the overall user experience.

Please let me know me know what you think about any of the points raised by replying below.

Special thanks to @wingz, @pfunk, @complexring and @jesta for their input and for helping me in gathering my thoughts on the topic.

Authors get paid when people like you upvote their post.
If you enjoyed what you read here, create your account today and start earning FREE STEEM!
Sort Order:  

I proposed something very similar if not identical in response to an early proposal by Dan in how to revamp the curation rewards. (Neither Dan's proposal nor mine was adopted).

Both your suggestion and mine are based on the fallacy that the eventual payout is a "fair value" that can be predicted. This is not the case due to the circularity, especially in the case of whale votes (since it is the votes that determine the eventual payout). There were some other issues with it but that is the key conceptual error that undermines the viability of it entirely.

Also, some sort of superlinear payoff is needed otherwise the incentives favor voting for your own posts regardless of popularity or merit, either directly or via shill voting.

In thinking about this quite a bit over the past few months I've also come to believe that superlinear payoff is economically correct for a platform like Steemit, because only posts that are of interest to a large number of people benefit from being shared in this manner. If something is only of interest to a few people (dinner plans for example), it is better to share it privately via email (or private Steemit messages once that is implemented!) and not impose both the resource and clutter costs of sharing it with thousands or millions. So something of interest to 100 or 1000 is worth more than 10x or 100x something of interest to only 10; the latter probably has close to zero or possibly negative value (once costs are considered) on Steemit.

All of this is not to rule out that tweaks to the current model might be helpful, but getting all the incentives right after considering all the relevant factors is much harder (and I would suggest the current system is much better than many of its detractors claim) than just wanting to make them "more fair". It is also not always so easy to see the various perspectives on how the current system performs multiple useful (and even essential) functions, when focusing in narrowly on one aspect of it such as perceived fairness.

I do like the idea of providing some sort of incentive for valued downvotes; perhaps some version of these ideas could be applied in a more limited manner to do that. But even there, getting all the incentives right and avoiding opening up vectors for abuse is not easy.

·

Both your suggestion and mine are based on the fallacy that the eventual payout is a "fair value" that can be predicted. This is not the case due to the circularity, especially in the case of whale votes (since it is the votes that determine the eventual payout). There were some other issues with it but that is the key conceptual error that undermines the viability of it entirely.

Not 100% sure I know what you mean by circularity.

"since it is the votes that determine the eventual payout" - Isn't that exactly how markets work though? AFAIK they're the best tool we have at assessing value. Why would it fail here?

One counter-argument could be that content quality changes over time, while the supply of Steem used to pay content creation/curation remains constant. I still think that an incentivized downvote is better in those cases than upvotes only, even if it doesn't accurately find a fair value for all posts. That fair value is also likely platform dependent, but it would find some sort of equilibrium or "standard" over time.

Again, I might've misinterpreted/misunderstood your point, please let me know.

Also, some sort of superlinear payoff is needed otherwise the incentives favor voting for your own posts regardless of popularity or merit, either directly or via shill voting.

I agree. My feeling is that n^2 is too aggressive, but I don't have much to support that at the moment. Maybe we can address the self-voting effect by using a weaker form (n^1.2 ?) and not have the same blowout effect.

In thinking about this quite a bit over the past few months I've also come to believe that superlinear payoff is economically correct for a platform like Steemit, because only posts that are of interest to a large number of people benefit from being shared in this manner. If something is only of interest to a few people (dinner plans for example), it is better to share it privately via email (or private Steemit messages once that is implemented!) and not impose both the resource and clutter costs of sharing it with thousands or millions. So something of interest to 100 or 1000 is worth more than 10x or 100x something of interest to only 10; the latter probably has close to zero or possibly negative value (once costs are considered) on Steemit.

I agree, but I don't see why we couldn't have that and an incentivized downvote?

Also, we're not quite in that "Dinner plans" vs "President Obama AMA on Steemit" kind of situation when a post gets reposted and gets 1000x payout the second time around. Not quite your point, but when talking about non-optimal distribution of wealth, that's a prime example.

Thanks for taking the time to read this and for joining the conversation, I'm glad you got to see it and that it didn't get 100% lost in the abyss of Steemit.

This suggestion definitely creates an interesting dynamic in the reward system.

It's an interesting scheme you've developed after our conversation.

I'm glad that the scheme attempts to incentivize downvotes.

Why don't we remove the Lottery effect? Currently if I where to up-vote an 3K post that post could gain 1$-2$ while if I where to up-vote a new post with a few cents(0.5$) . I'm luck if that post would move 1 or 2 cents at most. In my opinion that 3K post should be getting the same amount of my votes affect (or at least similar) than the newly created 30min to 1h old post.

·

I think you're right. From the whitepaper:

The economic effect of this is similar to a lottery where people over-estimate their
probability of getting votes and thus do more work than the expected value of their reward
and thereby maximize the total amount of work performed in service of the community.
The fact that everyone “wins something” plays on the same psychology that casinos use to
keep people gambling. In other words, small rewards help reinforce the idea that it is
possible to earn bigger rewards.

I'm sadly not sure if the "lottery effect" is really something they'd like to see removed... It seemed strange to me after a first read that this is what they were going for. Not sure that having a system that "plays on the same psychology that casinos use to keep people gambling" is really something we should strive for. Aiming to have fairer economy in which its users don't feel "tricked" would seem more reasonable to me.

·

Why are you voting on the 3K post at all??? It doesn't benefit you. Even though the post reward may go up $1-2, your curation reward will still be tiny because you are a very late voter. I don't think people understand how things actually work and that is a large part of the problem

·
·

For what it's worth, I sometimes vote on a $3k-type post so I can actually see my vote make a difference for the author. It's cool that even as a small fish I can participate and help them get a couple bucks.

At first I was really trying to maximize the value of my upvotes. I was all focused on getting in early on the posts destined for the top. At my levels of steem power though, all of my curation rewards are way less than ten cents. So for now, I get enough satisfaction from seeing an author's post get a boost that I go for it sometimes.

·
·
·

Makes sense. Thanks for the explanation!

·
·
·
·
·
·

its much more encouraging if you can change the payout with some cents then voting for something new and see no effect at all (except if you have tons of steem power).
Therefore a more linear payout algorithm would encourage more users to vote for articles they like, even if they have not much votes yet. Self votes I don't see that much as a problem. With a transparent blockchain you should easily find out the black sheep's that selfvote all the time.

I suggest that you yourself can decide where your inflated coins are going. If you don't use your coins in lets say 30 days they can be distributed the same way as now. With this in place the coins could be instantly paid out with every vote. If you have the feeling that it is your coins that you distribute you would also take much more care how to use them.

For example you have 10 dollars worth of steem power. With an inflation of currently lets say 1% per day and 5% of that for blogging rewards you have 10 dollar * 100%*5% = 0.5 cents a day to distribute. If you don't distribute for 30 days you have 15 cents to distribute. If you still dont distribute them every day 0.5 cents are distributed trough all participants.

جيد جدا ، ولكن لا يوجد

·

:(

·
·

Hmm.. I hope he's ok? :D

Really solid write up. You've highlighted some of the major potential barriers to mass adoption. There's certainly a bandwagon effect with posts.

There are lots of things going on under the hood and you've provided a good perspective on things people may not have even known were going on.

Thanks @owdy

Great article. in my post I threw a half idea out that curation and author allocation should be switched. total rewards all shown solely for the post (not individualized) and votes hidden. The bulk of the reward then going to comments upvoted. I agree it is good to expand and toss idea out, I like your proposal, I bristled a bit at seeing one whales vote essentially undo the communities opinion on it based only on that reason.

·

Thanks for the feedback, I left some of my thoughts as a reply on your post.

·

Giving a 25% portion of post rewards to comments as a handout was proposed for Steem v0.8.x but it was almost unanimously agreed by witnesses that it was a bad idea. It would have invited incredible amounts of no-value spam comments everywhere. Even without those incentives, a ton of bots showed up making generic comments in a ton of posts until people started countering them with downvotes.

I personally think if that change had happened it would have been a disaster for Steem(it). It would have allowed abusive/garbage bot accounts to scoop up a large amount of SP. Giving the bulk of rewards to comments is redundant and a handout. Comments can get upvotes already.

Interesting incentive you have planned out. While I can see your conversions, I still wonder what is "fair value." On Steemit it's pretty subjective. A system like this could most definitely benefit some users, but it would also hurt some more powerful users (although not very much), so I wonder what a whale's incentive would be.

·

Isn't value always subjective? Its objective component comes from evaluating value assessment across a group of people. In a free market that's done automatically where supply and demand meets, that's pretty much what I'm proposing here.

A system like this could most definitely benefit some users, but it would also hurt some more powerful users (although not very much), so I wonder what a whale's incentive would be.

How? There seems to be a lot of similar comments, but I don't understand how you reach that conclusion. It would "hurt" users that would've otherwise been overpaid, but I see everyone else as winning.

@owdy This falls apart because it incentivizes downvoting. Downvoting is taking money from someone without the need to give any explanation.

I've only ever seen @smooth downvote whales like @dantheman and only because every time they post, they suck up 100% of the money in the system cuz everyone wants to be liked by a whale.

The far simpler solution would be to cause a popup to occur when upvoting which presents a series of stars. 5 stars causes you to vote with 100% power, 0 stars causes you to cast a 0% vote. Each star is 20%.

A zero or null vote would be almost the same thing as a downvote. It wouldn't remove money from the system or the poster, but it would put a slight drag on the upward pressure on their future earnings growth for the post.

In otherwords, the UI should just reflect the existing capabilities of the platform instead of trying to make everything a binary or adding yet another incentive that can backfire horribly.

·

This falls apart because it incentivizes downvoting. Downvoting is taking money from someone without the need to give any explanation.

I agree with that second sentence, but I don't think in falls apart. In one case you have a runaway situation, in which more upvotes->more visibility->more upvotes->more visibility->... In the other you have a self-correcting one. If you downvoted a post from the trending page and it were to lose visibility, another one would take its place for people to see and evaluate, instead of always having the same ~8 posts showing.

I've only ever seen @smooth downvote whales like @dantheman and only because every time they post, they suck up 100% of the money in the system cuz everyone wants to be liked by a whale.

A zero or null vote would be almost the same thing as a downvote. It wouldn't remove money from the system or the poster, but it would put a slight drag on the upward pressure on their future earnings growth for the post.

I like the idea of putting a drag on upvote pressure. And yes, it'd be nice to have a weight scale or stars for our votes, especially for whales to use.

·
·

@owdy giving curation rewards for guessing the fair value seems interesting. In away it would be like a prediction market :) I dont see whales votes as a problem as @smooth sees it. Because currently whale votes help to find out what the vision of steem is.

To balance voting itself more even and be more welcome to newcomers I would suggest to give some kind of universal dividend to all people that fully verified their account.

I also like @williambanks idea of having different vote options.

I would suggest to allow 3 different kind of votes:

  • If you press the first time up-vote you only rank the post higher, no extra payout.
  • If you press the second time up-vote your vote is also considered for the payout.
  • If you press the third time up-vote you indicate that this post is very very important for you.

Important posts could be valued higher lets say 10x your voting power, but should be more limited then normal up-votes, lets say max 30 in the last 30 days.

the same could be done for downvotes.

To the visibility / trending problem, why not use some kind of lottery:

Currently posts are not only rewarded exponentially, but also they get a lot more votes if they become visible on the trending page. Why not draw the posts which are displayed on the main site as default in a lottery like style. Every time you reload the page the displayed posts on the top site could be drawn through a lottery. The more votes they have already the higher the chance to be selected. This would give all participants a fair chance to be listed at the top and therefore attract more votes.
(or in your proposals case downvotes :))

If you want more details about these suggestions, in this post I looked deeper into the topic how we could evolve the voting / payouts, feel welcome to share your thoughts and link similar ideas:
https://steemit.com/steemit/@arcurus/tagging-and-flagging-hidden-by-a-whale-how-to-evolve-further

·
·
·

In away it would be like a prediction market

Exactly!

I would suggest to allow 3 different kind of votes:
...

Yep! There's definitely something that could be done with voting weight. It's already there in the background on Steem, just not on Steemit, so integrating that shouldn't be too difficult.

I agree about your second point, but I don't see it as a real solution to the problem as Steemit scales up. The number of spots on trending is just too limited for that implementation to have a large impact. Even right now, we're talking about 4000 posts/day, there's just no way a lottery system can cycle through enough of those to make a major difference. - What do you think?

·
·
·
·

a the lottery system from trending topics can scale up perfectly, because the articles are chosen for every user each time he reloads the site. So it can even scale with billions of users and articles :)

·
·
·
·

@arcurus You're right! I misunderstood the concept, if it's done on an individual basis it could certainly work. I like the idea and, right now, I can't really think of a better way to easily address visibility issues.

·

@owdy This falls apart because it incentivizes downvoting. Downvoting is taking money from someone without the need to give any explanation.

I mostly agree. I don't personally see a way to incentivize legitimate (anti-spam/scam) downvotes that wouldn't lead to more abuse angles. I'd like to hear a solution that does however, so I wouldn't discourage someone from thinking about it. I disagree with incentivizing subjective value-reduction downvotes though. If someone downvotes based on a subjective opinion on a post's value, it should cost them voting power and not give them anything back.

The far simpler solution would be to cause a popup to occur when upvoting which presents a series of stars. 5 stars causes you to vote with 100% power, 0 stars causes you to cast a 0% vote. Each star is 20%.

People with higher amounts of SP have been asking for an easy UI on Steemit for controlling vote power. Even though the majority of users don't need this, the whale accounts have enough stake that it's still an important thing. I would not like to see a two-step process to vote everywhere however, especially not for all accounts. Some kind of toggleable option tucked away to enable some kind of tiered % voting might be ideal for whales and casual users alike.

·
·

If someone downvotes based on a subjective opinion on a post's value, it should cost them voting power and not give them anything back.

How and why does this differ from upvoting based on a subjective opinion?

IMO the adjustable voting power thing isn't that important, even for whales, because the system already adjusts your vote power to how much you vote. Whales might fine tune this a bit, but the incentives favor, overall, using as much of your vote power as possible, which is close to what the system already does.