Artificial Intelligence, Epistemic Institutions, Values and Reflective Processes, Great Power Relations
Recommender systems are used by platforms like FB/Meta, YouTube/Google, Twitter, TikTok, etc. to direct the attention of billions of people every day. These systems, due to a combination of psychological, sociological, organizational, etc. factors are currently most likely to reward content producers with attention if they stoke division (e.g. outgroup animosity). Because attention is a currency that can be converted into money, power, and status, this “bias toward division” has impacts groups at every scale; from local school boards to Congress to geopolitics.
Ensuring that recommender systems can mitigate this bias is crucial to functional democracy, to cooperation on catastrophic risks (e.g. AGI, pandemics, climate change), and simply to reducing the likelihood of escalating wars. We urgently need more research on how to better design recommender systems; we need to create open source implementations that do the right thing from the start which can be adopted by cash-strapped startups; and we need a mix of pressure and support to ensure these improvements will be rapidly deployed at platform scale.
Bridging-based Ranking for Recommender Systems
Artificial Intelligence, Epistemic Institutions, Values and Reflective Processes, Great Power Relations
Recommender systems are used by platforms like FB/Meta, YouTube/Google, Twitter, TikTok, etc. to direct the attention of billions of people every day. These systems, due to a combination of psychological, sociological, organizational, etc. factors are currently most likely to reward content producers with attention if they stoke division (e.g. outgroup animosity). Because attention is a currency that can be converted into money, power, and status, this “bias toward division” has impacts groups at every scale; from local school boards to Congress to geopolitics.
Ensuring that recommender systems can mitigate this bias is crucial to functional democracy, to cooperation on catastrophic risks (e.g. AGI, pandemics, climate change), and simply to reducing the likelihood of escalating wars. We urgently need more research on how to better design recommender systems; we need to create open source implementations that do the right thing from the start which can be adopted by cash-strapped startups; and we need a mix of pressure and support to ensure these improvements will be rapidly deployed at platform scale.