There's some slippery feedback loops involved, even if the models were very good, the reflexive nature of doing something like this would be very hard to get around