Optimal distributions of plans and the RICSH model
A motor plan is a collection of signals that an organism’s internal model expects to produce certain sensory data when the signals interact with the rest of internal state of the organism and the external state of the environment. When an organism prepares to do something, how many plans does it create? One plan, or many?
Armen Alchian said somewhere—I don’t remember where—that you don’t choose an optimal plan, you choose an optimal distribution of plans. It wouldn’t be wise to make just one plan because something might go wrong. Instead, you want a distribution of plans so that the decision of which actual plan gets implemented can be highly context-sensitive. For example, a general in command of an army would prefer to have a plan for an attack from the front and a plan for an attack from the rear so that they can deploy whichever plan is suitable for wherever the attack ends up coming from rather than only having a plan for an attack from the front and just hoping that the bet pays off.
This idea seems to be observed in motor behavior. According to the affordance competition hypothesis, a motor behavior is constructed by constructing several different motor plans in parallel. The plans then compete with each other until one wins out and becomes the thing that happens. So this isn’t just one plan, it’s many. Rational behavior consists partially of selecting the best plan, but also of selecting the best distribution of plans that will be used to select the best plan.
Choosing an optimal distribution of plans isn’t just smarter than committing to a single plan up front, it might even be necessary. Each plan is an agent, and agents produce group plans when they coordinate. The resulting group plan would be advantageous as compared to any of the individual plans because it would take advantage of the decentralized information associated with each individual plan.
Being able to select distributions of agentic plans would be very useful for a constructionist mind. Instead of having to construct a solution directly, the system can just produce the elements that will naturally construct a solution on their own as they interact with each other. This is all very speculative, but it has interesting implications for how interoceptive signals interact with the agents of a collective intelligence to produce the collective’s thoughts, words, and actions.