Inquiry into emergence of high level AGI, and the computation needed to host the growing network


The emergence of high level AGI and/or ASI might require the independent allocation of tasks, if the allocation of tasks is not done with significant influence over the network by an intelligent AI itself, from the network itself, how will this emerge?

Secondly the network will mostly be hosted by Raspberry Pi devices after algorithms are trained 3rd party, this means although there is a lot of computational power – it is not at all significantly high. How will the network itself be able to train in real time by the data of it’s own use?
because that’s how a brain works.

so if there was like a pyramid of AI, those at the top are given a task or decide a task and then on each subsequent level, the AI’s are trained/train to find the best combination of more narrow subsequent AI then a lot of computational power is needed, even if it does not work like this – where will all the real time computational power come from? just devices? just a pseudo-integration with DBC?