Alignment Protocols

"Collective intelligences are networks of smaller intelligences, acting as a single more-or-less coherent one. To make this work, each smaller agent needs to know what it should be doing, based on its location in the network. In theory this could be decided on an action-by-action basis, based on the entire history of everything the agent has ever observed. In practice, such an approach is monumentally unworkable because the number of such contexts is so high that it's likely this particular one has never been seen before, let alone enough times to learn the optimal policy. The solution to this is the purpose protocol: agents choose from a tree of shared finite set of possible purposes ("roles"), based mostly on the roles selected by nearby network neighbors."