[syntheweave]

In the broad picture it demonstrates a common phenomenon of what I dub "flip-floppy" behavior in digital systems: the optimal design at one scale inverts at another, which creates the layered approach we so often see: for example sometimes graphics are best described as "raster", other times "vector". But then you go to implement a useful graphics renderer for some task, and you do both even if you output only one. Sometimes you need geometry and sometimes you need samples and you don't end up with a pure system of one or the other.

So the same could be said of what we need in consensus: sometimes the right move is to turn to an authority to get an answer, other times we coordinate one out of a network. Twitter is doing some of both to be a useful service.

I have a creeping suspicion that ML and distributed ledgers go together in that one can be the flip-floppy complement of the other - ML as agent and oracle, ledgers as a mechanism to build shared knowledge among agents.

(I am also of the opinion that we are hopelessly clueless about how to apply these technologies well right now. But even if that means it is foolish to try when it results in so many failures, I think joining the fools is a noble idea.)