The same way we predict how the trillions of trillions of air molecules behave when we predict the weather: we build models of aggregate behavior, try to understand relationships, run simulations and test our predictions. We're not 100% right all of the time, but we now have the ability to prepare for disasters days in advance and re-route airplanes around storms before they hit. Weather prediction and the activity around it was once an intractable problem but now is a huge, huge benefit to society.

The things you bring up are not even the hard part… The difficulty is that the system is dominated by feedback loops. I still think it's within the realm of 'engineering can improve this'.