The Apathetic User Problem
During the spring of 2017, I became interested in solving an issue that people were complaining a lot about but really taking no real action towards. The problem was the concern of fake news and vulnerable users being unable to discern legitimate new stories from those that were more manipulative, deceiving, and plain untrue. So, naturally as a CS major with too much time on his hands, I tried to think of a clever algorithmic way to solve this issue.
The first problem with going with a full blown hard core machine learning algorithm was that such methods need prior information to serve as a basis to learn how to discriminate between good news from bad news. Unfortunately, we as human beings already have a tough time of doing this. Which is not good news for the algorithm.
Training a machine to learn apples from oranges is easy because we as human beings can make this distinction with little problem. We simply gather a sample set of images of apples and oranges and feed this labeled information to the machine to learn.
The bigger issue is whether or not you can reliably separate the good news from the bad news beforehand and build an effective model on it. Due to political biases, this separation might be non-representative of the truth, and even if we do build a model that is accurate on the test data, there are too many ways to deceive using language that such a model lacks the nuance to be of any real use. There will always be too many false positives and false negatives.
So, I opted to design an algorithm designed by taking the input of human beings and using the nonlinearities inherent in each individual to build a better overall picture of the truth. While not necessarily a true picture, it would be one where the most people were sure of it being a good source. With that algorithm I designed a prototype website that allowed the ranking of news.
Then over that summer, blockchain drew me in and I began to regularly blog on Steem. By the fall, I had began to establish myself as a writer and understood approximately how the surface politics and mechanics worked. I found a similar problem to the fake news problem: one of distinguishing quality content.
So, I adjusted the algorithm I designed to sort the news and used it to sort Steem posts. I played with it for awhile and set up a system to incentivize other users to play with it, but it turned out not to be very popular and I didn't really feel like spending more time marketing or spending money on the project.
You see, all human-based algorithms also suffer from a problem. Algorithms can be broken if they don't have enough data or don't have accurate data to work with. Human-based algorithms also fail when they don't have accurate data to work with. The problem is one of trusting users to behave correctly when there is no incentive to do so. This is the basis of the apathetic user problem.
This means that if an incentive doesn't line up with the behavior one should not expect the overall system to work in the desired way. The issue with Newsiphy was that there was zero incentive to rate the news correctly. The issue with qfilter was that there was a small incentive that couldn't compare with the rewards already available with the platform and the algorithm relied too heavily on voting power.
The issue with Steem is that there is no incentive to curate good content when voting for garbage content is more profitable. Since users tend towards apathy, we should expect them to misbehave or simply not care. That sounds a lot like Steemit today.
That is not to say that all users are apathetic. The problem is too many users are apathetic. We see the same thing in elections. Since there is little real incentive to vote, lots of people decide not to vote. The same thing with witness votes. The issue here is that because the behavior is wrong, the algorithm becomes faulty. Yes, I just called Democracy a faulty system.
Now one could imagine that they could remedy the issue by simply offering the users magically generated internet money. The problem is that users are apathetic. They don't care about what you intend for them to do. They will take a path of little resistance in order to obtain the incentive.
Steem ideally wants users to generate quality content and for the applications to display that content. Users on the other hand don't care. They will take a path of little resistance. This means they post low-quality content quickly in order to maximize voting opportunities and thus maximizes rewards. The incentive is misplaced and thus due to apathetic users, the content created is mostly garbage.
Ideally, the human-based algorithm is to act as a community-based verification method of information. For Newsiphy, this was fake news. For Steem, it was engaging social media. Both fail due to apathetic users that misbehaved because there was to reason to behave.
This is also a reason why DPoS is a less trustworthy system than PoW. Proof of Work completely avoids the apathetic user problem, because the apathetic user is not mining. In Delegated Proof of Stake, the witnesses are completely reliant on voting and thus are made to be less secure simply because they have exposure to apathetic users.
If only a small portion of the community votes for the witnesses then how can the system be truly representative of its underlying user base? If a smaller portion actively monitors the positions of their witnesses, then the problem is made even worse. The fact that dead accounts have a permanent voice is also an issue. Given that there is little realizable incentive to the apathetic user, they remain apathetic and vote (or don't vote) and are misrepresented.
That is not to say there isn't a incentive to keep track of witnesses. Such behavior is vital to the blockchain. But unfortunately, most users and people lack the long-term vision to truly feel the sway of this incentive. Long-terms incentives rarely impact the apathetic user unless there are potential devastating consequences.
It seems that we might never be able to leverage the wisdom of the crowd if such incentives are never aligned with the system designer's intent. Although we can build elegant solutions, they quickly become ugly when we assume the best from human beings when we should be expecting nothing at all. Human beings default to the lazy way of doing things. It's part of our nature. Only activists and non-apathetic users put in the extra work, but often times they are outnumbered and can rarely be depended on as a solution.
So, there's the problem in a nutshell. You think you can solve it? Give it a crack, it is a very interesting and rich problem as well as being a very difficult one. While apathetic users may not kill Steemit, a platform that addresses the problem and has a better alignment of incentives will surely outperform the Steem platform given a base level of support. It's something to think about. So think about it.
I don't think this setup is all that different from the real world. You will need activists to move people, just like before. I find it rather promising that there are organizations forming with specific purposes that can potentially change the behavior of others on this platform. Steem cleaners, for one.
Maybe you are assuming too much about what a successful platform looks like. For example, from a purely advertising point of view, it matters more that there are enough eyeballs on the platform, no matter what trash is there. If steem continues to attract and retain users, this could be what it looks like. Contests run by advertisers and trash flying around, and some thoughtful gems here and there. I'm actually quite surprised in its current state the variety of content that is thrown out there, and even creative ways of leveraging this platform.
That said, it's not perfect, but the good part about this kind of system is that it can evolve. Whatever types of tweaks made to incentives, we at least got a lot of minds scratching our heads about what the possibilities are. Maybe the right solution is throwing it all out and redesigning too, certainly a possibility... For now, this has proved to be quite an interesting experiment to watch.
I'm sure if Steem gets no competition, and continues with its roadmap (whatever that may be) it will be successful. I like looking more at ideal solutions or robust solutions rather than ones which are fragile to certain elements of human nature. Granted every solution will be fragile to some element of human nature, but those that are more robust tend to survive and thrive.
I'm sure Steem that will evolve, but will it evolve too slowly? But I guess that's part of the experiment. The Steem developers do have interesting ideas, but they are kind of slow at implementing them (ie Hivemind, Communities)
I have to agree with @eonwarped. It's not an unusual issue (I wouldn't even call it a problem) that people are just into cheap glitter and simple diversion. Cat and dog videos plus travel and cooking stuff are nr1 on youtube and so they are here.
What must never be underestimated when it comes to "lazy" users is that they might have a very different mindset and idea of what to do and what to think about. As in real life you have to create an institutionalized habit first if you want to reach something and the tool set for that is called marketing. If you apply that and don't make too many mistakes, you will end up with a stable and satisfied niche that is doing things "actively". Overall, a normal value for the share of users (or employees or people in a club who are system relevant) is 10%, the rest is "just there". At the end, you have to reach these 10% and get them into your niche.
My approach to positive select quality would be recursive reputation. At the end it is always the integrity of a single user that decides if something is quite likely quality or true, which would be the reputation part. And the more other users refer to output of that one user, the better the content must be, which would be the recursive element.
Practically it would mean that if you write a post, you have to attach it to another post. That can be your own or somebody elses, but the latter is incentivized. The user who wrote the referred content gets a part of the money that goes to the post plus it gets highlighted a bit more, because there is one more post attached to it. On the other hand, the attached post gets highlighted a bit more too and this depending on the size and quality of the post it is attached to. Of course, you can change the post your content is attached to if you see that as necessary (but not in a spammy way)
That would result in content cascades in which the single content creators control each other and give an indication about whether other content can be trusted or not.
I don't know if it would work and it surely has plenty of pitfalls that needed to be worked out. But in my mind that creates the very environment needed to separate the BS from the rest.
That's an interesting idea and by itself given a clean slate I think it may work pretty well. On the current blockchain I feel that many new users would simply attach their posts to popular users regardless of the quality of their content. The popular users could simply form a ring (point to each other) and thus that small group is elevated by everyone else.
That is not to say that this is a bad idea. This may also work better in smaller communities and sub-communities as a means of organizing information and providing structure within the group. Rather than sort the information via a list, you essentially have a reputation tree / network that could be used to find related content more effectively than the disaster of a system that the current applications use.
Yes, you got it as I had thought it. The issue of attaching content to the most popular could be fixed with reward and visibility for more than one level of the tree: The more steps to the first post there are the more a post is highlighted and in return the more attachments your own post has (no matter on what subsequent level) the more you get out. Ideally,that would lead to an equilibrium in which first-posters try to get users to reply to them or reply to their repliers and repliers try to find the ideal position in the top while still getting enough highlighted.
yeah, this is really something that needs to be fixed here on Steemit as well. The search bar doesn't work at all for me. I have go via google site: if I want to find something here. The lack of love they have put into steemit is astonishing.
PS: your q-filter works. I found 2 interesting posts with it:)
True democracy doesn't really work. It develops into chaos very fast. Churchill is credited with saying "The biggest argument against democracy is a five minute discussion with the average voter" and I have to agree :D
People not voting isn't necessarily bad since if they don't care enough to vote then they probably wouldn't make a good choice anyway. I'm perfectly fine with spammers and apathetic users not being represented. (this applies to Steem and US government ;)
That said, I don't know how many of the quality users are involved in voting for witnesses. I personally haven't done much research on witness and haven't voted on many because of that.
Don't know if you've considered it, but one possible improvement to q-filter is to give extra points to blog posts where someone upvotes and comments. This normally shows they really appreciate it. Just a thought. I really like the idea of q-filter and have been trying to spread the word about it.
Apathetic users not having a say is fine, but the bigger issue is that they have a say forever unless they explicitly un-vote that witness. There is no expiration. The witness could do lots of bad stuff and would still retain a lot of delegated power from users that may not use the platform anymore.
For q-filter, the only issues with comments is that it is relatively easily to cheat the system by simply adding a cheap comment. But I am working on changes to qfilter which should benefit the regular users (the non-apathetic ones) with more regular votes and more weight in terms of the sort.
Essentially, I am planning on switching the upvote bot from a stake-weighted score to a participant-based score. I'm also adjusting some of the parameters and make the participant-based score favor those who participate frequently over those that don't. This makes it a little more resistant to apathetic users and also fixes that whales voting problem that was effecting you earlier.
While i know steemit is too new and this is a real genuine problem. I also know that the platform is still growing and few witnesses cant take care of everything on thr own.
While quality content has chances of good returns. I can easily post a regular blog and get it upvoted by the bots by paying a little portion. Lemme try explaining it by using my accounts as example.
i use thenomadictales account for important posts from my travel. There are times when i witness something and want to share it with steem community but then i feel that my current followers won't like it and i made another account for NOTSoRandom stuff.
I so feel closer to not so random account. I see more people liking my NomadicTales account and making it a success. To each his own.
Making content other people like versus making content you like. That's the creators dilemma I guess.
And the platform might be too new. We'll see if Steemit or someone else will learn from these mistakes.
Slowly and steadily wins the race. All the best @steemit
Good points, a very important issue that rears its' ugly head when it comes to all online activity. The bot presence on steemit of dead accounts is definitely there (I actually wrote a few brief articles on the platform, if you'd like feel free to check them out). I saw 11% of accounts started last year posted 30 or more times, making that barely more than twice a month. 1/9 of the platform brings value to the rest, funny how that turns out, huh? I think there should be more selectivity, more of a cost to join because giving these accounts away is encouraging the exact same thing. I would like to see an "Untalking Dead" initiative on here where accounts are placed on a 90 day chopping block one time as a warning if they are not used at least once a every ten days (on the average) at any given time to post original content, or they can void this by delegating all steem in their account. Sounds very harsh, but the vast majority of dead weight on here could easily be cleaned up if the community wished to.
That, or we could apply a bleeding of the stake (or halting interest payments on SP) for inactive users. That would achieve the same thing over time and encourage users to be active. Unfortunately, I don't see such a change ever occurring but it's an interesting idea.
Steemit could make things harder by using different verification schemes prior to being able to receive payout or they could stop giving money away to accounts. But as the same time, we want Steemit to give away their money in order to decentralize the power of the platform. Things to consider.
Yeah I am likely pretty hardcore on the thought, glad to hear your response, keep posting the great thought provoking material many of us think but surely can't articulate in a practical, coherent sense like you @greer184!