Note
Jabble is a new kind of conversation platform, designed to make conversations on the Internet more intelligent and less polarized. Signup here to get notified when we launch.
Like many online social platforms, in Jabble users post things and other users can reply. Here's an example:
Each post gets a score, shown as a percentage, based on how many upvotes and downvotes it receives.
However, not all votes count equally! Your votes on a post get more weight if you read and acknowledge convincing replies to that post made by other users.
Jabble ranks replies based on how convincing they appear to be. A reply is considered convincing if users who have considered that reply are more likely or less likely to upvote the post being replied to. For example, if a post containing a video has a reply claiming that the video is fake, users who see that reply may be less likely to upvote the video.
Jabble will place the most convincing replies at the top, in order to make sure as many people as possible are exposed to the most convincing replies and have a chance to respond to them.
The idea is simple. If one user shares information that might influence other people's votes, then people should have a chance to consider that information when deciding how to vote.
Disinformation, misinformation, clickbait etc. are all fundamentally the same thing: misinformed content, information that people spread, but that they wouldn't if they knew more. But the antidote to misinformation is information.
Jabble combats misinformation, and promotes more informed content, by identifying and promoting:
- convincing information
- posts that people still upvote after they have seen the most convincing information
Of course, a convincing reply may itself be misinformed. So another reason the algorithm puts the most convincing replies at the top is to make sure people have a chance to respond to them.
You can respond to a convincing reply in two ways:
- By voting on it
- By replying to it
Simply voting on a convincing reply gives more weight to your vote on the post being replied to. This is because Jabble considers the votes of people who have voted on convincing replies to be more informed.
For example, suppose you upvoted the post with the video, but have not seen the reply claiming it is fake. You are in one sense less informed than the users who saw the reply. Voting on the reply signals to Jabble that you have considered it and are thus fully-informed, and so your vote will have more influence on the final score of the video.
Now, suppose you are not convinced by the reply. By downvoting the reply, while not changing your vote on the video, you signal to Jabble that you are informed but unconvinced. Jabble will give more weight to your upvote on the video, resulting in an increase in its score.
This means that you can have more influence on the scores of posts, and thus on what posts get more or less attention (and possibly go viral), by making sure to vote on comments marked as convincing.
This will be immediately visible in the UI: when you vote on a convincing reply, you will see an immediate change in the score of the post being replied to.
score increases after user votes on convincing reply
You can also respond to a convincing reply by replying to it. Replying to a reply also signals to Jabble that you have considered it, and thus gives your vote more weight.
But a reply also gives you a chance to make counter-arguments that might convince people the other way. For example, if you reply with a reason to believe that the video is in fact authentic, and people who consider your counter-argument are more likely to upvote the video (compared to people who only considered the reply claiming the video was fake), then your counter-argument will be marked as convincing!
It might seem like this process could go on forever, but it probably won't. As we hope to demonstrate with early Jabble users, we expect most discussions will converge after the strongest arguments have been made on each side of an issue and nobody can find anything to say that further changes minds.
Jabble is designed to drive constructive deliberative processes for discovering informed opinion. By focusing attention on convincing replies, and the replies to those replies, the process drives a constructive discussion that surfaces the best arguments on both sides of any controversy. And by scoring posts based on the informed votes of users who have considered the most informed argument threads, the algorithm can promote informed content, instead of content that gets a lot of upvotes based on initial impressions and knee-jerk reactions.
The algorithm we've described here is just one part of our vision for improving ranking algorithm for social platforms. It doesn't yet solve certain big problems, such as bias, bad faith voting, and large-scale, coordinated manipulation efforts. We hope to address these problems using with an improved algorithm that incorporates a reputation system, game theory, and an unsupervised machine learning model, as described in our article on Deliberative Consensus Protocols.