A game theory requirement for steem downvotes has not been shown

Image Source: pixabay.com, License: CC0, Public Domain

Introduction

A while back, @bitcoindoom, and @sigmajin wrote some articles which made a number of compelling arguments for the steem downvote:

• TIL the best strategy for reducing rewards disparity (in defense of the flag, part I) (@sigmajin)
• The Dwin fallacy(In defense of the flag part II) (@sigmajin)
• Why Down Votes and Flags are an Unavoidable consequence of Game Theory (@bitcoindoom)

With recent discussions of flagging, I thought I would revisit these arguments. @sigmajin's first article had to do with the N² voting curve. It demonstrated that the downote is a more efficient way than the upvote for equalizing the distribution of author rewards. The second article from @sigmajin showed that if the downvote exists, the rules for using it must be individualized and subjective. The article from @bitcoindoom invoked the economic concept of externalities and looked at voting from the perspective of the multi-player prisoner's dilemma. In this article, @bitcoindoom argued that the inability for one player to "punish" defection would necessarily lead to defection by other players.

In this article, I will briefly respond to each of those articles, and attempt to make the argument that they have not shown a requirement for downvotes in the steem article ranking system. The basis of my counterclaim is that these articles did not consider three aspects of the steem rewards auction. Those three aspects are: (i) competition; (ii) vote removal; and (iii) alternative auction systems.

Before proceeding, however, I want to point out a recent comment from @vandeberg:

The rewards pool is a differential equation that is an input to a different equation whose solution is an equation that is dependent upon user behavior.

What this means is that the rewards pool is complex, and all of the metaphors that we use to explain it to ourselves are just abstractions. The abstraction that @bitcoindoom used was the multiplayer prisoner's dilemma. The abstraction that I'll be using in this article is a two tier auction, where the steem witnesses are auctioning off the steem rewards pool to authors in exchange for bids of articles, and to voters in exchange for bids of voting power. Framing the rewards pool under this abstraction immediately raises the question, if the downvote is necessary, then why is an analogous type of "negative bid" so uncommon (non-existent?) in real-world auctions?

Image Source: pixabay.com, License: CC0, Public Domain

Responding to @sigmajin and @bitcoindoom

TIL the best strategy for reducing rewards disparity (in defense of the flag, part I) (@sigmajin)

Since this analysis focused on the N² influence curve, the changes announced with hard fork 19 are anticipated to render it obsolete. Additionally, it only compared the efficiency of the upvote and the downvote within the existing auction scheme. It also did not consider the possibility of implementing different bidding protocols.

The Dwin fallacy(In defense of the flag part II) (@sigmajin)

This analysis took it as given (from @bitcoindoom) that the downvote is a necessity, and concluded that if the downvote exists, each voter must be able to use it as (s)he chooses. It also made the useful observation that steemit is a unique platform in that, voting determines the acceptable use rules for the platform. That observation is only partially true, though, because the platform determines the rules for voting.

Why Down Votes and Flags are an Unavoidable consequence of Game Theory (@bitcoindoom)

This is the analysis that most directly contradicts my claim today. I would argue, however, that it is not clear that a multiplayer prisoner's dilemma is the right framing for the game, and that there are no externalities in play. To understand this, it is necessary to understand the concept of an externality. In economics, an externality happens when one entity is forced to bear the cost (or benefit) of another entity's actions. A common example is waste from a factory that pollutes the properties of downstream neighbors. @bitcoindoom argued that spam polluting the steem blockchain is analogous to factories that pollute an ecosystem, and the downvote is necessary to punish the polluters. However, in the steem ecosystem, low value content doesn't directly reduce the value of anyone's property. All that is needed to limit the volume of low value content is to refrain from rewarding it.

Filtering irrelevant content is the role of the user interface. The role of voting is merely to provide a stake-weighted ranked list of articles in order to determine the auction winners. This analysis, however, attempts to move filtering from the UI to the voters. So in my opinion, despite the title of the article, it does not demonstrate that, under all possible auction schemes, a stake-weighted ranked list will be higher quality with a downvote than without it, or even that the best possible ranking system will include a downvote.

I highly recommend reading the articles from @sigmajin and @bitcoindoom, but I don't think they quite made the case for the necessity of downvotes. The following sections will discuss some of the factors that I think they omitted.

Image Source: pixabay.com, License: CC0, Public Domain

Competition

Before hard fork 19, competition for upvotes is an admittedly weak factor. It is simply impossible for almost any number of small stakeholders to bid enough to counter large stakeholders. After the influence curve is adjusted, however, that should change. Once this happens, authors will need support from more voters. If everyone votes for their own posts, and only their own posts, the incentive will be high for even a small stakeholders to vote for something else. Also, for every author - sybil or not - voting for their own posts, there will be tons of voter-only or mostly-voter accounts voting for many other posts, and authors will need to compete for those votes.

In the real world, if I think an item is overpriced at an auction, I simply decline to bid, and spend my money elsewhere. That competition is generally sufficient for pricing things fairly. I have not seen anything to persuade me that there's something fundamentally different about the rewards pool auction.

Vote Removal

In addition to self-voting, another concern is plagiarism. What do you do if there is a high value article that turns out to be plagiarized? The naive answer is that voter2 casts a vote to negate an older vote from voter1. However, the better answer would be for voter1 to remove their own vote.

To police against plagiarism, protocols could be developed whereby accounts (like @cheetah) comment or vote on an article to signal plagiarism, and voters could use this information to go back and remove their own votes. These protocols could even by automated with competing @steemvoter style of services so that voters could effortlessly guard against voting for plagiarism. Competition among these services would encourage them to "get it right."

What if users don't subscribe to these services or remove their own upvotes? I guess it means that the community supports plagiarism. But it's the same voters. If we can't trust them to wield their upvotes responsibly, how can we trust them with downvotes?

Image Source: pixabay.com, License: CC0, Public Domain

Different Auction Types

A number of years ago, I remember reading an article about how Google does research on auction theory in order to determine how to price their advertising. I can't find the article again now, but one of the concepts I remember is an auction type where the first two bids matter. The highest bid wins, but the 2nd place bid sets the price. Wikipedia describes one such auction as a Vickery Auction, and they have a whole page on different auction types. Now, I'm not economist, but with all that variety, I would be hugely surprised if one of those different techniques doesn't solve many of the problems that downvotes are intended to address.

For example, what impact would it have on self-voting if the payouts were calculated based on all votes but the highest and lowest? What would it do to overvalued articles if they were based on the median vote, as suggested by @demotruk?

Conclusion

Before closing, I would like to repost a comment that I posted the other day, on a post by @lukestokes:

In my 9 months here, I have seen nothing to indicate that the value (to the platform) of the flag/downvote outweighs the harm that it does. What problem does it solve that cannot be solved in another way?

• Spam? Unfollow / mute / bandwidth limiting.
• Disagreement over value? Vote for something else (or lots of other things).
• Illegal / abusive content? Unfollow / mute. Also, flagged or not, it's still on the blockchain.
• Fraud / plagiarism / trolling? This is probably the strongest argument, but is it strong enough to outweigh the damage that flag-wars do to the platform? Can be addressed through comments and automated removal of up-votes after plagiarism is detected.
• Self-voting? (i) it still happens, anyway; (ii) stakeholders should be able to vote for content subject to whatever criteria they choose. Competition will penalize the ones who don't vote wisely.
• Reputation? See @neoxian's proposal

If flagging didn't exist, as far as I can tell, the above problems could all still be addressed in other ways. For example, plagiarism might be combatted by creating services (like @steemvoter) to monitor comments and automate removal of one's own upvotes if plagiarism is indicated in a comment (or even by an upvote from a particular account).

At this point, I remain unconvinced that flagging solves any problems that cannot be solved by plain old-fashioned competition for rewards/up-votes. Flagging would only be, perhaps, necessary if up-votes could not be removed.

When discussing downvotes, it's important to keep two things in mind. First, is the perpetual question from Thomas Sowell, "Compared to what?" The question has to be is the system better with downvotes than it is without them? Yes, downvotes address real problems, but in doing so they create new ones. The analysis has to consider all the trade-offs. Saying that they're needed for a few particular reasons is an incomplete analysis.

Second, people like to talk about "banning downvotes" or someone's "right to a downvote." Downvotes are arbitrary rules of the game, we don't have a right to a downvote any more than we have a right to a capital letter in our steemit account. As long as we're all playing under the same rules, the game is fair. Which rules to use should be driven by optimizing the value of steem, not individual wishes to wield retributive justice.

So I am not saying that downvotes should be eliminated. I simply don't know. It's certainly possible that I'm overestimating the negative impact of today's "flag wars" on the price of steem, and underestimating the good that it does for the platform. My argument here is just that the case that downvotes are required has not been made in the best arguments I've seen, so I hope that the developers are considering this question with open minds.

@remlaps is an IT professional with three decades of professional experience in data communications and information systems. He holds a bachelor's degree in mathematics, a master's degree in computer science, and a master's degree in information systems and technology management. He has also been awarded 3 US patents.