Polarization, radicalization, and the “echo chamber” problem:

  • The bias problem of self-selecting groups

  • Groupthink: people coming from the same background already have many shared ideas

  • Reinforcement: discussion can reinforce apparent agreement on already-prevalent ideas

  • Tribalism and identity politics: when “us” versus “them” dominates other considerations

    • Example: US politics at the moment 

  • Polarization online: do social platforms increase polarization (more than other groups)?

    • Example: Facebook/Twitter newsfeeds: you see mostly what your friends like

      • Likely to be mostly what “your community” already agrees with

      • Platforms have been trying to 

    • How much of a problem actually?  The research is inconclusive so far… 

  • Radicalization online: do algorithms like YouTube help radicalize people?

    • Intuition: 

      • Algorithm’s goal is to keep people watching more

      • Turns out that angry/emotionally-triggered people keep watching

      • Algorithm “learns” to suggest emotionally-triggering videos because it’s empirically effective in keeping people watching more/longer

      • Emotion-based radical rightwing media benefits more than others

    • How much of a problem actually?  The research is still early, inconclusive… 

  • Potential mitigation factors:

    • Labeling of information origin, provenance

      • Helps people take detached/skeptical perspective, evaluate sources

    • Diversity of perspective opinion: make groups “divers” (how?)

      • Random sampling-based groups, e.g., deliberative polls, juries

    • Increase breadth, diversity of information diet

      • Not just from friends but from larger perspectives

        • Traditional role of “national/international media”: e.g., NY Times

        • Risk: “too broad”, disconnected from lives of most readers?

      • Is there a suitable balance between local, global, and in between?

  • Potential technological/algorithmic approaches

    • Basic: diversity through random sampling

    • Advanced: balancing local and global with compact graph summarization

      • Graphs (e.g., information topic graphs with semantic linkage edges)

      • Metric spaces (e.g., high-dimensional topic spaces)

      • Compact summarization schemes: e.g., approximate distance oracles



Post-lecture blackboard snapshot 2019:



Last modified: Thursday, 26 November 2020, 13:08