Cognitive architectures like SOAR, ACT-R and Clarion are often described in the AI literature corpus. The elements of such system are oriented on human information processing. Different modules for storing short term memory, long term memory and procedural memory are realized in the system. The cognitive architecture tries to mimicry a psychological model. But how exactly should this concept be useful for creating an AI system which controls a robot?
My hypothesis is, that a cognitive architecture was never proposed as a control system but as a plan recognition system. Which means, that SOAR isn't controlling a robot but it's monitoring a robot. Or better it observes what a human is doing. This results into a memory hierarchy for analyzing the human. If one system tries to understand a second system it will emulates it. Or to explain it shorter. SOAR isn't a robot, but it's a camera who understands what a human is doing.
The problem with my hypothesis is, that i didn't found literature which supports the claim. Because the topic of behavior recognition isn't discussed very often in the AI literature. It's something which is very new. Even the ICAPS conference had introduced the topic only in a small workshop outside the normal curriculum (GAPRec workshop). Which means that plan recognition has a low priority within the academic community.
My question is: Can SOAR be used as an event recognition system? Does a system A which monitors system B will need the same inner structure to understand it fully?