You are viewing a single comment's thread from:

RE: Why I Advise Against Linear Reward

in #steem6 years ago

"In order to realign incentives and discourage individuals from simply voting for themselves, money must be distributed in a nonlinear manner."

I agree with the above statement; however, I don't feel that is the correct question to be asking here. The problem we are trying to solve here is not people "voting for themselves", it's how to properly incentivize real curation efforts as compared to automated "gaming" of the reward system.

In my opinion, no change in numbers or algorithm will achieve that goal. No matter what we set the reward curve to, or the curation percentage, it will always be possible to take advantage of the algorithm to gain optimal rewards via automated voting over manual curation.

The only way I see as a realistic possibility to effectively combat that is a crowdsourcing approach (as compared to an algorithmic approach). Currently the crowdsourcing aspect of Steem rewards is clearly broken due to the extreme disincentive to downvote, which is why I am strongly in favor of something like a separate downvote pool which is being discussed.

Once that is in place, I think the rest of the reward system should be made as simple as possible for people to understand, as this will help the crowdsourcing effort. That is why I am against changing both the reward curve and the curation percentage. I feel those changes will end up being gamed just the same as the current system is now, and on top of that they make things more complicated for users to understand which will hurt the effective crowdsourcing of rewards which I hope we can improve significantly.

I should probably make a more detailed post about my thoughts on this topic when I have some time.

Sort:  

The first thing that popped to mind as well to me was that it was not the relevant question. Interesting.

I have a doubt when it comes to downvoting.

I think downvoting must be part of the system, however, I am worried about a culture of downvoting. Downvotes are perceived much much stronger than upvotes, so if everyone downvotes a lot this may backlash.

It works on youtube though.

I think that's mostly because downvoting is so rare nowadays, and so financially disincentivized, that when it happens it's a really big deal. It will likely take some time for people to adjust to the change, but after a while if it becomes a commonplace thing (like on youtube as you mentioned) then it won't be a big deal anymore.

I downvote over-rewarded posts all the time. I leave a little comment giving my reasons. I sometimes a get a downvote back on the comment. I use decline payout on my comments for protection. It works pretty well.

If they downvote my comment, they are using Voting Power on me instead of themselves. They are inflicting damage on themselves without even thinking about it.

there is no downvoting on youtube, there is only like and dislike, and report - here on Steemit any whale or deepstate sponsored individual can ruin someone's rep by big downvotes, seen it when @berniesanders did it to me

The problem I have with the incentivized flags is that rather than be used to crowd-source for quality etc, they will become ridiculous vendetta tools. Can't really claim that the average maturity of the platform is high as if it was, a lot of the problems wouldn't exist in the first place.

I agree that the reward system should be as simple as possible, at least for the beginner stages as the current barriers to entry are already so high due to the complexity of the entire process. The complexity can still exist of course but the average user need not feel they have to worry about it at sign up unless thy are investing in.

I think that's all part of the crowdsourcing. If someone is using downvotes as an attack against otherwise valuable content, then the community can step in and upvote it. Often I've found that people who are the target of unfair downvoting can appeal to the community and come out ahead of where they started in terms of support and votes.

Also, i'm not necessarily saying that what i'm proposing will work, i don't think anyone knows that for sure. But my guess is that it would work better than adjusting numbers in some algorithm, and i think the experiment of the Steem platform as a whole is about crowdsourcing rewards more than algorithmic rewards, so I would like to give it a real shot.

Often I've found that people who are the target of unfair downvoting can appeal to the community and come out ahead of where they started in terms of support and votes.

It becomes a strategy to get upvotes as I think we have seen utilized enough.

and i think the experiment of the Steem platform as a whole is about crowdsourcing rewards more than algorithmic rewards, so I would like to give it a real shot.

Don't get me wrong, I am pretty open to experimentation here as long as there is the understanding that it is going to keep getting experimented on to find a point that is best for the community as a whole. No solution is ever going to be perfect so in my eyes, the experimentation doesn't end.

Flags aren't really incentivized in this proposal (we don't know of a good method to do that even if we wanted to), they just stop being so costly. You still earn essentially nothing by downvoting. Your other content does go up every so slightly slightly in payout all else being equal, but it is quite negligible (and your other content too can be downvoted if don't have sufficient merit)

Bad choice of words. Less-disincentivized. Don't get me wrong, I am in support of trying it as I would like to see how it plays out. I understand that people fear flags and that should change (my first flag was from you a very long time ago :D ) and I would hope that it would make people think twice or up their standards before boosting nonsense.

There is no point discussing forever something that can be trialed quite cheaply. Work or not, either way, we would have better knowledge going forward instead of only predictions.

Yeah, as I've seen the back and forth on this topic, it seems to me that every proposal needs a downvoting component for it (and I do like that you say 'crowd-sourced' here, that does seem apt) to work, so I am pretty much in agreement with you here.

Superlinear reward encourages some curation as opposed to the current system that incentivizes exclusive self-upvotes. Good or bad can only be determined by the person doing the curation.

Superlinear reward encourages some curation as

It doesn't and if you read the above section of the whitepaper carefully, no such claim is made:

money must be distributed in a nonlinear manner

This states a necessary but not sufficient condition. The white paper does not claim that superlinear alone is sufficient to achieve curation, in fact it states precisely the opposite because it explains the need for downvotes to prevent collusion (a somewhat technical term, but in this context, it means more SP working together to increase the payout of content above its merits; somewhat different from the implied English meaning since there is no requirement for multiple actors, only an increased amount of SP working to self-reward).

Unfortunately the white paper got some of this wrong. It describes the crab bucket model where stakeholders are supposed to pull down others who try to collude to extract rewards. However, in the real world, we have seen that most stakeholders would rather see to their own reward earnings rather than 'waste' vote power pulling others down. The psychology may be valid, but the cost of downvoting is just too high for psychology to override economics here.

Crowdsourced downvotes as described by @yabapmatt also distribute rewards in a non-linear fashion, as follows. If 100% upvote and no one downvotes, the reward is 100% relative to the amount of voting power used. If 80% upvote and 20% downvote, the reward is 75% relative to the amount of voting power used. This is non-linear and satisfies the necessary condition of the white paper. This assumes that downvotes don't deplete upvote power as in the @cervantes/@kevonwong/etc. proposal. If they do then the reward is 60%, but either way, still non-linear.

We think that is a better approach. I wouldn't flat out rule out some superlinear (indeed we have one at my urging in HF20 with the new treatment of dust votes at the very low end), but n^2 doesn't look like a good solution and other solutions are underdeveloped (and, quite possibly, unnecessary).

I really appreciate your inputs!

it explains the need for downvotes to prevent collusion

I implied this as it's implied by the whitepaper.

that most stakeholders would rather see to their own reward earngs rather than 'waste' vote power pulling others down.

This isn't a binary thing. It's part of some very complex ever-changing dynamics. At some point, the bulk of the stakeholders even agreed to refrain from voting to a large extent forgoing a lot of rewards.

Crowdsourced downvotes as described by @yabapmatt also distribute rewards in a non-linear fashion, as follows. If 100% upvote and no one downvotes, the reward is 100% relative to the amount of voting power used. If 80% upvote and 20% downvote, the reward is 75% relative to the amount of voting power used.

Very interesting solution. I hadn't understood it correctly so I'm very grateful for the way you've described it. My first impression is that it's a solution I could very be in favor of.

This isn't a binary thing. It's part of some very complex ever-changing dynamics.

We have two and half years of experience across various other tweaks in the reward system to see basically the same result. Stakeholders, by and large, don't want to waste their valuable earning potential on downvotes or not voting (there are always exceptions but they are precisely that).

At some point, the bulk of the stakeholders even agreed to refrain from voting to a large extent forgoing a lot of rewards.

Yes it was me (along with abit) who organized and promoted that so I'm quite aware of it. Having been there, I can tell you it wasn't sustainable. Even if abit and I were willing to continue doing what we were doing indefinitely, the other major stakeholders were not, and in the place of the largest stakeholders, the next largest were starting to do the same thing.

That sort of crab bucket approach does not work if downvotes are too expensive. Other than that, it could very well work. I've even said that with cheap downvotes n^2 could work. It might produce different results than what many in the community want in terms of reward distribution, but it might be (probably would be) a whole lot better than what we saw with n^2 and no cheap downvotes.

I agree with most of the points you've made and our disagreements aren't fundamental disagreements.

I wanted to state my truth as clear as possible. The selfless are losing some ROI to the selfish in a flawed system.

Selfishness is not a long term successful approach. Since it's not sustainable. The selfless are losing in the short term frame yes. But they can build up stronger relationships that lasts for a life time. In the end the goal is to use Tokens to produce higher leverage by investing in human brains that think big and long term.

This space needs to evolve into digital jobs eventually if you truly want to maximise ROI. At the moment since it's such a new field everyone is just trying to do some blogging with random stuff. More organisation is needed. Niches + Community systems. And moderators that own various Niches would be an interesting experiment. People focusing on niches that they love. Therefore everyone is getting more of what they want.

Thank you for the discussion.

the only get rich here is the bots. that's the reality, just saying

I do not understand how it encourages curation.

If we have nonlinear rewards my upvotes are worthless unless I upvote the most popular content. So it supresses curation even more so, as my voice is suppressed unless I follow the crowd.

Right now my vote as an example is worth $2. I get to give this to anyone I like. Much easier for me to actually curate and reward what I find valuable.

This is what I want to see answered. Superlinear isn't magically going to make curation happen.

i think the assumption is that people will converge into what's valuable... but as the old saying says: beautify is in the eye of the beholder and to apply it to steem...

One man's shitpost is another man's gold..

When you say "people will converge", you mean when things are superlinear? This isn't going to be true, unless you mean to say "a few top whales will converge". And that's the troubling part for me.

Posted using Partiko Android

I've long come to grips with the fact that trending was never amazing... it was different because of how young the project was.

Look at whaleshares, look at the amount of rewards being given... early adopters are potentially killing it... and its linear!

watch a year from now people over there complaining about the same things they do here.

Nonlinear rewards can be gamed by dividing your SP up into a bunch of different accounts. you're not going to stop self upvoting. What we really need is account based voting with accounts verified by oracles, which is coming anyway I believe.

Nonlinear rewards can be gamed by dividing your SP up into a bunch of different accounts.

Incorrect as pointed out by @smooth.

It may, but my point is that I would prefer to keep the reward algorithm as simple as possible, and allow the community to regulate it from there. So if the community decides that excessive self-upvoting is bad, they can downvote people who do that, which then removes the incentive that is there now.

Flagged rewards are simply returned to the reward pool and as stated in my post those flagged rewards are up for grab by selfish and selfless upvoters alike, leaving selfish upvoters at an advantage.

Flagged rewards are simply returned to the reward pool and as stated in my post those flagged rewards selfish and selfless upvoters alike, leaving selfish upvoters at an advantage

Not true. Those second round of selfish voters are also more likely to be downvoted. You have to look at this as a large economy, not individual steps. The first downvote of a selfish voter puts the rewards back in the pool but that doesn't mean selfish and selfless voters are on then equal footing. The downvotes can continue to follow selfish voters indefinitely.

The result is likely to be an equilibrium with far less selfish voting, not only because selfish voters get downvoted, returning those rewards to the pool, but because crowdsourced downvotes change the incentives on selfish voting in the first place. It will not be perfect (nohting is) but it will likely be far better, with more for the pool going to value adding activities.

Those second round of selfish voters are also more likely to be downvoted.

True but under linear reward, the rewards returned would represent a growing % of the reward pool with ever-growing incentive to defect. If downvotes are crowdsourced my statement is irrelevant.

That entire statement makes no sense. The returned rewards represent either the same % of the reward pool (if all returned) or a smaller % (if only a portion are returned). There is no way for it to increase. I have no idea what sort of convoluted reasoning has led you to conclude this.

I've cut corners while explaining my point.

Let's say someone's flag or vote was first worth an infinitesimal part of the reward pool well after the most extreme case imaginable where every single vote is counteracted, well that infinitesimal vote would now control 100% of the pool.

The incentives to defect become greater as more flags are given because flagers become "fewer" or spent and the potential rewards greater.

It's a very convoluted way to make my point and I'm not sure how sound my explanation of it is but hopefully, it's clear enough.

It's good that you are calling me on my half-baked explanation that I had given. I knew when I wrote my answer it wasn't really clear at best and possibly didn't mean much at worst.

well that infinitesimal vote would now control 100% of the pool

Aside from the fact that this situation is contrived and would never happen....you haven't demonstrated anything here. It is just as easy to downvote that vote as all the others, and there is no reason to believe that after all the other votes are counteracted this one wouldn't be too.

You could make a similar argument for non-linear. As votes become more concentrated there is no incentive to vote for anything but the single highest-paid post (even if it is complete and utter garbage/abuse/etc.). Everything else will pay nothing making the vote worthless! It becomes a tyranny of vote-for-the-biggest-or-your-vote-is-worthless.

These extreme cases are not helpful.

They're up for grab for the next least selfish self-voters. The most likely contender for countering selfvoting is independent flagging pool along with burning of flagged reward, which I guess due to technical implementation difficulty doesn't have much traction.

They're up for grab for the next least selfish self-voters.

Indeed. It's a loophole with serious issues as I've stated.

The most likely contender for countering selfvoting is independent flagging pool along with burning of flagged reward

Leaving the rest for grab for the next least selfish self-voters leaving the same loophole open as I understand it.

"next least selfish" = Not selfish.

Not sure how you understand it otherwise.

It doesn't put potential abusers at a disadvantage.

yes it does

Of course it does. They are more likely to be downvoted. Even if it isn't 100% guaranteed it pushes their incentives toward better (less likely to be downvoted) behavior