Agree with everything you said, especially on the challenge of recommendations and feel this is where those designing the recommendation engines have at least the same moral responsibility as a bio-engineer doing gain of function research on viruses.
IMHO the core problem is the recommendation engines are optimizing the wrong metrics and this leads to destabilizing feedback loops.
The problem is likely AI complete, but it should be possible to tune recommendations so they don't create self strengthening conspiracy theories. I suspect a reinforcement learning environment that mimics users interacting with a recommender system would be the best way of tuning the larger system behavior so the longer term properties aren't catastrophic for society.
OTOH if the platform was setup so "good faith" arguments were somehow incentivized, there might not be any need to optimize for any other metric than engagement.
IMHO the core problem is the recommendation engines are optimizing the wrong metrics and this leads to destabilizing feedback loops.
The problem is likely AI complete, but it should be possible to tune recommendations so they don't create self strengthening conspiracy theories. I suspect a reinforcement learning environment that mimics users interacting with a recommender system would be the best way of tuning the larger system behavior so the longer term properties aren't catastrophic for society.
OTOH if the platform was setup so "good faith" arguments were somehow incentivized, there might not be any need to optimize for any other metric than engagement.