lead people to form erroneous beliefs ... whether they promote consensus or belief accuracy ... more polarized and less accurate beliefs ... help users find more mutual understanding on these platforms ... content that receives approval across different groups is promoted ... bridge divides and encourage positive interactions across diverse audiences ... bridging-based ranking can promote consensus ... elicit belief updates that benefit collective accuracy ... improve both collective accuracy and average individual accuracy ... aligning them more closely with socially desirable goals
I don't know what's creepier: a polarized society but with a more diverse view on topics, or a cohesive society with an eerily uniform view instead. Polarization is problematic, but so is thought-policing. I would argue feeds shouldn't be algorithmically determined, but composed of content by channels the user has explicitly chosen to subscribe to; and a separate explore feature for interest-based suggestions, without creepy "this is what most people agree with, and so should you" course-correction measures.