Overview
What if you could get a social feed that was representative of a city, state, or nation? If you read it, it was like mind-melding with that full population at once, because it had a representative weighting of every point of view. The opposite of a filter bubble.
This post is me thinking out loud about how to build it. It’s early.
Baseline Biases
Bubbles must represent the baseline e.g. actual ratios of real interior beliefs of everyone it covers. This is hard, and this section is a list of the biases I see which must be handled.
Loudness
Only certain kinds of people speak publicly fill out surveys, or respond to pollsters. Getting the ground truth of how many people believe is hard.
Preference Falsification
Some things are not perceived to be safe to say out loud. Anonymity is thus important, and yet that makes the whole project far more complicated to accomplish. A soulbound identity system would make this easier, but for lack of that a reputable organization making strong legal guarantees, and using cryptographic systems whenever possible, may have to suffice.
Drift
Opinion shifts, and if we only measure ground truth at a population level at long intervals, the algorithm constructing representative bubbles will be behind that curve. Mitigation could include faster population polling, or primarily tracking beliefs that are more fundamental and change less over time. Without a solution here, fast swings in mood will not be tracked by a bubble. You might thing you can use the information in the bubble to update the baseline, but that makes the system vulnerable to astroturfing and other meddling (although - a soulbound system also can help with this). Perhaps some kind of “Nielsen Family” system can be employed, where every so often when the baseline is generated, a set of vocal people with verified social feeds are weighted and followed by an algorithm which plans to update the baseline based on their comments until a larger snapshot can be regenerated.
Bubble Building
Once you have a baseline, how do you surface a subset of information which accurately conveys it? None of these ideas are mutually exclusive.
Twitter-like
One way to do this is to build a feed of posts. The posts let in should each be considered in isolation based on if they are tilting the recent collection of posts closer or further from baseline representation. Furthermore posts should also be filtered to remove mean, crude, gross, or otherwise offensive content - that may reduce accuracy but if the point of this exercise is a society better at listening to itself who then acts more wisely, I believe maintaining pleasant civility is also an important goal.
Substack-like
It should also be possible surface opinion articles and posts that represent well-articulated instances of POVs in the baseline. There is an open question of how many perspectives of less literate and less educated citizens would be left out via such a process, one possible fix there would be to surface video and podcast content transcribed and edited down. And on the topic of editing, if the size of the feed is too verbose, condensed versions curated by LLMs might streamline things.
Dashboard-like
Finally, one might create a clustered visualization of what groups of people are saying. The system could take in everything muttered online recently, weight attribution to real people according to the baseline, and generate infographics of how many people think what, which opinions have big trends up or down, new beliefs coming onto the scene, etc.
Anyway, just a sketch of an idea.