Virtual Soar-Agent Implementations: Examples, Issues, and Speculations

Caveat: Virtual Agents Might Not Even Require Cognitive Architectures

Design motivations behind whether or not to deploy a cognitive architecture often varies according to whether the virtual agent is meant to academically evaluate embodied and/or extended cognition or instead, to implement a virtual character (NPC) for social and/or ludic entertainment purposes. With these contextual contingencies in mind, most contemporary virtual agents are not required to be cognizant of the visual inputs beyond the raw numerical data underlying each object and path. Basically, a virtual agent can successfully interact and navigate its environment without being directly aware of it in any introspective way. An agent can react to numbers alone. For example, a bit-mapped image or any other virtual object can be represented simply as a stream of numeric representations (e.g. binary digits). These numbers provide the agent with coordinates, proximity-relations, and other manually coded identification data that allow the agent to respond in a completely reactive way. In many video-games, for example, an NPC can appear to be intelligent simply by mindlessly reacting to the changes in numerical inputs afforded by the dynamic changes occurring via game-play interaction. However, cognitive architectural enhancement might be useful for any agent-based cognitive task involving reasoning and mnemonic assistance. Otherwise, a cognitive architecture may only be useful in this numerically-reduced case if we wish to imbue the agent with the ability to contemplate its environment as an “[...] idealized version of the underlying reality” (Best et al., 2006, p. 186). This underlying numerical reality might be modeled to enable embodied, extended, situated, and distributed (EESD) cognition so that virtual agents can actively “[...] structure or manipulate their environments” to “simplify, sequence, or otherwise, support cognitive processes” (Smart & Sycara, 2015, p. 3837).