Midwife: CPU Cluster Load Distribution of Virtual Agent AIs

This describes a CPU cluster load distribution algorithm that may be helpful when simulating crowds of artificially intelligent Virtual Agents. The cluster load distribution algorithm was instantiated in Java code, demonstrated successful execution on a small cluster of inexpensive CPUs, and supported the simulation of Virtual Agents within a readily available, open source, Virtual World client-server toolset. The client-server toolset used was the network-enabled, multi-user virtual environment (s) of Open Simulator and OpenMetaverse operating together. Although these Virtual World tools are well known, this experiment was unique in that the psycho-socially significant small-group oriented crowds of Virtual Agents in the simulation enjoyed interacting, individual, cognitive, affective, and behavioral Agent controllers executing entirely client-side. In this Artificial Life experiment, each CPU in a scalable CPU cluster controlled up to fifteen, individual, C#-based Agent controllers. This argues for and demonstrates how disembodied computational Agencies can be executed in distal computer environments when network-coupled to convenient Host embodiments in a Virtual World. There are implications in general here for a paradigm shift in Virtual Agents/Virtual Worlds modeling and in specific for positive computational load distribution of artificial intelligence in Virtual Agents inhabiting complex, emergent, crowd-scene models just by using more effective tools and modeling methodologies.