[evilduck]

I think we can speculate in the entirely opposite direction where the same action leads to positive outcomes.

Lots of legitimate human companions are abusive. People have a wide range of qualities and many of them are bad. AI may be a poor blanket replacement for all human companionship but it could easily be less bad than someone's immediately available alternatives and be used therapeutically to help someone model healthier behaviors to establish better actual relationships. Or in lieu of normal relationships being possible like long term isolation during space exploration or for life sentence prisoners or just neurodivergent or disabled people who have challenges the average person does not.

Going back to the food analogy, if given the choice between fast food and starving, or fast food and something poisonous suddenly everyone will overwhelmingly choose fast food because for many people "home cooked meal" was never an option.

[gls2ro]

I am not sure this is a good inference.

first, what does it mean lots? is it majority? because the AI and AI minded products are targeting everybody.

second, imagine the argument being about Facebook: some life interactions are sometimes not good, but connecting with people online will make it better. fast forward 10 years and we have studies how social media is making most of the people using it depressed and badly influencing our democratic choices. not sure we really solve anything here.