Taking A Deeper Look At "Proof-of-Brain"

The following is a post examining current structure of Steem's distribution mechanisms which when put together act as a "proof-of-brain." Some issues are talked about and some ideas and considerations are put forward. This post is made in the name of improvement and addresses potential concerns I have with the current ecosystem versus the ecosystem imagined in the Steem whitepaper and bluepapers.

I think that it is important to define and get a clear idea of what we are talking about before trying to analyze any idea or concept. In order to make sure everyone is on the same page, we will be making multiple references to both the Steem whitepaper and Steem bluepaper. Proof-of-brain is described in the Steem bluepaper as having the following two properties:

a pool of tokens dedicated to incentivizing content creation and curation (called the "rewards pool")

a voting system that leverages the wisdom of the crowd to assess the value of content and distribute tokens to it

These two properties come together to allow the Steem blockchain to distribute a large portion of the block rewards using human participation as the mining mechanism rather than using electricity and finding hashes which traditional proof-of-work blockchains utilize. The human work is what is valued here rather than the hash rate of your army of computers.

While some may debate the terminology and argue that such properties do not encourage actual brain power, we'll ignore semantic arguments and try to get to whether or not the whole proposal idea actually works as intended.

First off, we need to understand what the goal of Proof-of-Brain tokens are. The bluepaper helps to clarify this goal a little bit:

The users who produce content are adding value to the network by creating material that will drive new users to the platform, as well as keep the existing users engaged and entertained. This aids in distributing the currency to a wider set of users and increases the network effect.

So the goal of the system is to attract users by rewarding them for their contributions. This overarching goal should be no surprise to anyone using the platform. But let's dig a little deeper. The goal is to keep users engaged in the platform and to distribute rewards to a wider set of users. So, we want compelling content (not necessarily quality) and a fair distribution amongst the users. These two goals ideally lead to desirable network effects and the platform gaining mainstream traction.

In terms of distribution of rewards, however, we should be a little more specific than simply distributing to a larger set of users. Given the nature of Steemit, this goal is achieved simply due to content creation being more accessible to the public than mining. We get a clearer goal of what an ideal distribution achieves in the Whitepaper:

the funds must be as decentralized and distributed as possible

However, the whitepaper and I will both concede that ideal distributions are impossible. The whitepaper ultimately settles on the view that the most popular content will receive the majority of the rewards while the other contributors to the ecosystem will be rewarded with the small remainder. This is similar to how Steem works now. Whether this achieves the goals of distributing funds in a decentralized manner effectively we'll go into later.

One last thing before we judge the Steem model, is the concept that the whole proof-of-brain model is designed around. Note that this is what the Steem whitepaper says and emphasizes with italics:

Steem is designed around a relatively simple concept: ​everyone’s meaningful contribution to the community should be recognized for the value it adds.

While this is very much a subjective notion and proof-of-brain model is a subjective proof-of-work, there is something to note about this lofty goal. Contributions should be recognized for the value that they add. If we use the information earlier, value should be determined by how well the content engages the community when in turn should lead to desirable network effects.

Well, does proof-of-brain hit the target? Yes and no. It does a fair job of distributing rewards to popular content creators and rewards them for creating content. These powers users do drive new users to the site and many of them have initiated efforts to help new users and promote a greater sense of community. However, the goal of keeping the older users engaged through compelling content is not achieved to the extent one may imagine it would to achieve those desirable network effects.

The major issue is not the content itself, but the misalignment between the goals of the curators and the goal of the platform and the proof-of-brain system. In the current system, it is not compelling content that is rewarded, but profitable content. While at times compelling content can be profitable, this is not always the case.

And if we extend that compelling content is meaningful content, and that profitable content is not always meaningful content, then we see that Steem fails in it ultimate goal to reward meaningful contribution.

Proof-of-brain thus dissolves into people concocting schemes on how to make a quick buck and generate profit rather than use their brains to create meaningful content. While one could consider such activity as a legitimate proof that one has a brain, one would look down at such behavior as harmful to the ecosystem.

So, lastly I'll address some more specific problems and address some potential ideas with regards to those issues. While these ideas are not meant to be specific changes I would desire, they are meant to spur discussion and encourage brainstorming potential solutions.

First off, I argue that the main problem with the distribution of rewards is due to curation not working as desired. This is due to profit chasing rather than looking for and engaging with meaningful content. The main issue with this profit chasing is that it centralizes where the rewards go which is an attribute of a non-ideal distribution.

The most obvious aspect of profit seeking that users seek to reward their own content rather than seek meaningful content. This goal is achieved through self-voting and vote collusion. The first can be done directly, through biding bots, or through multiple accounts and the second is performed by a group of people conspiring to upvote each other's content without regard to the content itself.

The most obvious solution would be to ban the self upvote. But this only prevents one of the four avenues of rewarding oneself. That is not to say that such a change would not be desirable. Making the life of a profit-seeker harder does discourage the behavior and could encourage users to vote for other content given that they have to take additional steps in order to use the other methods. An idea, but not a perfect one.

Another solution would be to decrease the weight of repetitive votes on particular accounts in a manner similar to how voting power decays and replenishes over time. This mitigates the effects of all four of the methods as all methods require the same account or group of accounts to dish out the rewards. While it does not solve the problem explicitly, it prevents the centralization of rewards by a single user via a friendly group of individuals or accounts one happens to also own. Thus, only the truly popular are able to attract heavy rewards.

So, that goes over several basic ideas that have been discussed by several users in the community and the potential effects such implementations would have.

But other than rewarding one self, one can vote on profitable content and maximize their curation rewards. This is often done through frontrunning or following powerful voters and voting on the content they vote on (trailing). The issue with both of these means of rewarding users is that they centralize rewards. Rather than find meaningful content, profit is used to maximize one's own self and in the meanwhile and creates a situation where we see a non-ideal distribution.

If we go back to the decaying weight idea from earlier, we can argue that the rewards generated for upvoting the same content creator should decrease over time decrease one form of frontrunning, but you could simply frontrun a variety of popular voters.

Another idea involves simply decreasing the curation rewards for popular users using a variable split via some weighting mechanism determined by prior rewards or having users themselves determine the variable split. The issue with allowing users to determine the variable split is that they may keep curation rewards high in order to attract a large amount of users.

What we want to do is to attract the discovery of undervalued users and increase their popularity and because of this, a variable split that is performed automatically based on prior rewards might make more sense. In either case, we want the percentage of curation rewards to be lower for popular content and higher for unpopular content in order to encourage the discovery of meaningful content. Note that these ideas do not encourage upvoting bad content, but rather discourage the centralization of rewards among the same group of popular users and encourages users to find content meaningful to the community by undiscovered users.

That does it for this look at the proof-of-brain mechanism and what it has yet to achieve. While these solutions are vague and open, I feel that any discussion about these mechanisms and how we as a community want them to work are only helpful in achieving the goals that Steem has ultimately set out to achieve.