Emergent hierarchies for knowledge agents

We discuss concepts for constructing self-organizing hierarchies of agents to realize intelligent systems. Several hierarchies are introduced, including goal, model, control, and supervisory. Prior literature developed many essential components, but not general methods for organizing them into complete agents. We propose an architecture for extracting and organizing usable knowledge from environmental experience. The conjecture considers sensory and perception agents, memory, supervisory elements, activation, reaction, evaluation, and selection agents, alternative generators, modelers, and maintainers within structures which synthesize agents hierarchically. A supervisory hierarchy provides contextual awareness and overall operations management. An experimental model with two scenarios is implemented under the C-SIM hierarchical agent-simulator. The first operates in a text-based mathematical equation space. The second involves agent navigation in a three dimensional space with a series of progressive challenges providing mixed agent roles. Both scenarios evaluate the approach's generality.