>How many cores of my Raspi3 would be occupied by that
The Broadcom chips inside RPis actually have a pretty decent GPU for the power it consumes (low-power). Someone would have to create a system that could use that in a practical way for processing AI loads. So if someone solves that, then your GPU would be maxed pretty much (you'd want a heat sink kit for it), but the cores load could be kept to just one of the 4 available cores.
>Alone our energy costs would be higher than theirs.
I kind of think it's the other way around, but that aside the whole point of donating compute is that each of us bear a small(ish) part of the hardware and electricity costs.
Together if there are enough of us doing that can represent a total system far larger than any of us individually is likely to afford.
Yes, we'd need to have both a well-written description of the goal (each project for folding@home
has something like this for example), and then some mechanism for accountability in the beginning would be essential as well, IMO.
Once a robowaifu@home
system was up and running, and good AI results are coming from it for everyone, then I think that would speak pretty clearly all on it's own.