Adaptation and Evolution in Dynamic Persistent Environments

Optimization (adaptation) of agents interacting with dynamic persistent environments (DPEs) poses a separate class of problems from those of static optimization. Such environments must be incorporated into models of interactive computation.By the No Free Lunch Theorem (NFLT), no general-purpose function-optimization algorithm can exist that is superior to random search. But interactive adaptation in environments with persistent state falls outside the scope of the NFLT, and useful general-purpose interactive optimization protocols for DPEs exist, as we show.Persistence of state supports indirect interaction. Based on the observation that mutual causation is inherent to interactive computation, and on the key role of persistent state in multiagent systems, we establish that indirect interaction is essential to multiagent systems (MASs).This work will be useful to researchers in coordination, evolutionary computation, and design of multiagent and adaptive systems