People want AI of superhuman intelligence capabilities, but don't want AI with superhuman intelligence capabilities to manipulate people into using it.

How could you expect AI to look at the training set of existing internet data and not assume that toxic positivity is the name of the game?