The downvote cannot be fixed

remlaps (74)in #blog • 7 years ago (edited)

Introduction

About 10 months ago, there was a lot of contention on the platform over the issue of flagging. At the time, there was a bot running around haphazardly issuing downvotes to pretty-much everyone at some time or another. As a result, I wrote Down-votes: Steemit's Achilles' Heel?. Since that time, flag-wars have come and gone and come again any number of times, and now they're back. Here was my conclusion:

[Image Source: Pixabay.com, License: CC0, Public Domain]

Conclusion
If we use our imaginations, I'm sure we could come up with numerous other scenarios where a well-funded adversary could use the downvote to harm or destroy the platform. Are any of the current steem whales prepared to launch a counter-voting bot against a multi-billion dollar corporation or a state actor? Will they be ready to do it when the time is at hand? Or maybe we can just count on the financial Goliaths around the world to play nice when faced with a new competitive threat?

One mitigating factor is that a well-funded actor would drive up by the price of steem by purchasing their war chest, which would also strengthen the existing whales. I have doubts about whether that's enough of an immune system, though.

I have been persuaded by discussions here that there are theoretical and rarely encountered situations where down-votes truly are helpful, such as the crab bucket scenario in the whitepaper, but does the up side of down-voting really outweigh the down side or does the down-vote give a well-funded adversary a convenient lever to implement a denial of service attack? I'm not so sure.

Another thing that I hadn't even thought of at the time is this. The time is coming, sooner or later, when someone puts up an application that encrypts its data before storing it on the Steem block chain and internally controls who can see what data.

How do you even begin to know what to flag if you can't read it?

Here's the thing, everyone just assumes that the best way to guard the platform from abuse is through the downvote. So the prevailing belief, apparently in the complete absence of data, is that the benefits to the platform automatically outweigh all the potential ways that the downvote itself can be abused.

I don't expect this post to go far, because the down-vote mindset is deeply entrenched, but I want to point out that the use of the down-vote is always a matter of trade-offs. Maybe some tinkering can improve the ratio of benefit / harm, but it fundamentally cannot be fixed.

Maybe it protects the reward pool, but it also inspires fear in minnows and drives them away.
Maybe it buries some low quality content, but it also leads to "flag wars" that make the platform look like an episode of The Running Man.

Although I know it's likely to make little headway in the short term, I'd like to point out that sometimes it is useful to question assumptions. There are other ways to achieve consensus and evaluate quality, and many of them do not involve negative voting. I don't claim to have a solution, but I am going to suggest that maybe we need to think differently about the problem. Here are a couple of other approaches that might be taken:

Quorum sensing

In quorum sensing, bacteria hang around some area signaling other bacteria to their presence and monitoring their environment for signals from others. When they eventually sense signals in their environment in excess of some threshold, they suddenly begin performing some action - like turning virulent or luminescing.

Could Steem use a method like that for weeding out the spam and curbing abuse?

Karate Tournament Scoring

When @cmp2020 was younger, he competed in karate tournaments. Competitors would take turns executing their kattas (or forms), and five judges would announce their scores. The head judge would throw out the highest score and the lowest score, and take the average of the scores that remained.

An extreme version of this was actually suggested by @demotruk around the same time as my post above, in Up and Down Votes Are A Big Part of the Problem on Steemit. That article proposed using the median value as the reward threshold.

So in this model, some number of "outlier" votes would be ignored, and all posts eligible for rewards would be ranked based entirely on the remaining positive votes.

If a single whale wanted to capture some portion of the reward pool, they would be forced to split their stake, which would limit the harm they could do. If you throw out one vote at the top and bottom, they'd lose half their influence, unless their votes were in proportion with other voters. If you throw out more votes, they'd lose even more of their influence.

Second price auctions

GameTheory.Net says this about second price auctions:

An auction in which the bidder who submitted the highest bid is awarded the object being sold and pays a price equal to the second highest amount bid. Alternately, in a procurement auction, the winner is the bidder who submits the lowest bid, and is paid an amount equal to the next lowest submitted bid. In practice, second-price auctions are either sealed-bid, in which bidders submit bids simultaneously, or English auctions, in which bidders continue to raise each other's bids until only one bidder remains. The theoretical nicety of second price auctions, first pointed out by William Vickrey, is that bidding one's true value is a dominant strategy. Alternately, first price auctions also award the object to the highest bidder, but the payment is equal to the amount bid.

As of a few years ago, this was the method that Google was using for Ad-word pricing. It likely still is, but I don't know.

Take special note of this excerpt: The theoretical nicety of second price auctions, first pointed out by William Vickrey, is that bidding one's true value is a dominant strategy.

As with the Karate Tournament Model, if this model were implemented and a whale wanted to determine the award size, they would have to divide their stake in two in order to claim the top-2 votes. That division would limit the potential for abuse from any one player.

Both of the last two models could probably be combined with a quorum sensing model whereby self-vote abuse would be limited because every post would require two or more votes in order to get any pay out at all.

Conclusion

Are any of these the right answer? I doubt it. I'm simply planting a seed to suggest that there may be other ways to think about the problem.

As with down-votes, I don't expect that a perfect solution can be found. The question is one of optimizing the ratio of benefits / harms. I understand that most everyone who reads this article has already decided that down-votes are a necessary part of curbing abuse. So I'm not going to ask anyone to change your mind. However, will you at least spend some time thinking about the foundations of your beliefs about down votes?

How do you know that the benefits from the down-vote outweigh the harm? How do you know that no other way is possible? Will you consider the possibility that Steem could exist and even thrive without flag-wars?