Yes I think one main principle of technology is to be more energy efficient.

Scientifc method as word is already kind of limited for the object it tries to describe. Probably we need new words. Neural network is already blackbox so it's causality is not comprehended by human mind in case network is very complex. Network with few parameters can be causally comprehended with mind. So even if it is black box we can control and guide it.

I think AGI might be big black box with steering wheel and pedals. You see some stuff pass by from windows.

Science is just one way to collect knowledge. There is also traditional knowledge and then there is the actual experience. If it is scientific does not make it magically better than anything else. Example quality of research can be bad as it can be good. Research can claim all sort of things for all sort of reasons. With research there are also tabu subjects that are hard to make progress with. Let say extended water fast like 30 days is really tabu subject to research. Then by looking traditional knowledge, long fast has been more common at past. Then from subjective experience people have had good and bad experiences.

I think AI will need get these 3 sources knowledge on same line, not just scientific. Then level up and let the new knowledge emergence from this new level.