I'd imagine to begin to approach decent data sets to mimic humans and their functions several thousands of humans, singles, couples, & families would have to be monitored 24/7/365 for a couple of years. Needless to say not a lite undertaking but expensive as hell and well, problematic. (No wonder big tech likes to collect if not sell user data with impunity) Rounding up thousands of candidates would be a hard job to begin with, and then monitor the subjects for a couple of years. Remember those early internet days where just a few subjects were up online pretty much all day & night, and I don't think they lasted for years, only so many months.

Maybe start in China where they are in the middle of an experimental social graph on the masses. (I don't think that is going to work out in the fullness of time either.)