Lessons Learned from Curation Rewards Discussion

I would like to thank everyone who has participated in the curation rewards discussion! Many of you have provided incredibly useful insights. I would like to take an opportunity to summarize those insights for everyone who doesn't have time to read through the entire discussion as well as add new thoughts of my own.

1. Curation Rewards Attract Users

One thing is abundantly clear, curation rewards have had a major impact in attracting users to Steem. Removing curation rewards all together would have a negative impact on a meaningful part of our community. These rewards are one thing that sets Steem apart from the competition.

2. People and Bots will Play the Game

With the advent of advanced AI algorithms it is clear that computer models can provide high quality estimates of the value of a post and do so faster than any human could. This means that bots will always have an advantage in this game and they cannot be prevented.

Assuming we have the right incentives, the participation of bots is a good thing. It rewards those who invest time and money improving how new content is discovered. If we don't have the right incentives, then the same power will be applied to destroy us. In other words, blockchain incentives are a powerful weapon that needs to be pointed in the right direction to avoid shooting ourselves in the foot.

3. Curation Rewards Discourage Voting

People have repeatedly expressed that they withhold from voting to "save their power" for future use. What this implies is that "good content" is being ignored because people are saving for "great content". This implies that there is less total information (links between users and content). In some ways this is good, it means the votes that are cast have a higher quality.

4. Curation Rewards Impact Culture

When people are paid to curate they start to adapt strategies. Votes are cast for personal gain rather than for the benefit of the person they are voting for. This makes voting a selfish act rather than a generous act. The spirit of the site changes from one of rewarding content to one of "playing the game".

5. Rewarding the Little Guy

Someone with little Steem Power is just as human as someone with a lot of Steem Power. Their opinion carries a certain amount of weight simply because they are a person. Under Steem everything is weighted by Steem Power to prevent Sybil attacks. This means that people with just $3 of free Steem Power are currently unable to earn curation rewards because their weight is so small compared to larger players.

Any attempt to give advantage to smaller accounts will result in large users dividing their balance into many accounts. This in and of itself is not a reason to stop looking for more democratic solutions.

6. Designing a Better System

Some quality suggestions have been made on how to defeat the auto-upvote bots that have come out. One of the best ideas is discounting curation rewards for authors that have a history of earning high payouts. This means that the reward someone would receive for upvoting my posts will be less than the reward you would receive for upvoting a new user's post that is of equal quality.

The generalization of this algorithm is to implement our own on-chain AI for predicting the value of a post on a scale between 0 and 1. A high prediction will mean low curation rewards, a low prediction will mean high curation rewards. Looking at an author's history is just a simplistic form of such an AI. A more advanced form would consider the existing voters, the time of day, the tags, links, images, and even the content.

The goal of the curation rewards would be to discover unexpected results. Those who upvote things that our algorithm wouldn't expect to get upvoted should be rewarded. It is the unexpected results that add the most value and are the hardest to automate.

7. An Arms Race

Any algorithm we implement to estimate votes can be improved upon. The curation rewards are kind of like the Netflix Prize for blogging. Those who can write bots that are more effective than our own on-blockchain prediction will make money. The blockchain will have to constantly evolve its algorithms to incorporate the best known 'bot algorithms' and reduce the rewards for those who use them. Through iterative releases of the post-prediction algorithms the quality and speed of the curation on Steem will advance. User's who identify quality posts that are not recognized by Steem's algorithms or the bots will earn the most. Those who follow predictable patterns will earn the least.

8. Constant Tweaking may be Necessary

What I have concluded from this is that we must be prepared to tweak the curation rewards algorithm in response to advances in automated curation. Through this tweaking we can continuously re-bias the curation rewards toward human curators. We should welcome bots that figure out how to game the system, they only make us stronger. Perhaps there is an algorithm that can automatically "learn" and "adapt" to advancements in upvote bots.

Proposed Solution

In an earlier post I suggested rewarding those who who accurately predict the final payout. This kind of prediction market is harder for most people to use, but ultimately the hardest to abuse. I believe that if bot's were the ones to play the prediction market game, then the outcome of the prediction market could impact the weight of the existing curation rewards.

If the prediction market were to estimate a post to earn $1000, then the voting rewards would be small for that post. If the prediction market predicts $1.00 then voting rewards could be much higher. This creates a tension between making high predictions and rewarding the votes that make the prediction come true. The higher the prediction, the less financial incentive there is for voters to vote and therefore, the less likely for it to become a self fulfilling prophesy. The opposite would also be true: a low prediction would financially incentivise voters to prove it wrong.

The majority of the rewards should be allocated toward accurate predictions in order to advance research into bots that curate content. A token amount should be rewarded to those who vote. To maximize the perceived value of voting, the rewards may have to be in the form of raffle tickets. Most voters get nothing, but all voters get a chance to win $1000!

I believe with the right incentives we can motivate the community to produce the best curated content on the internet.