Whether we could do better if we (to take the given example) used the scientific method
I don't really see how we could do that. Yes, most predictions are rubbish - but a lot are rubbish because predicting AI is not something we have good ways of doing.
An over-simplification, but an evocative one: