Cognitive architectures

Starting since the eighties, a number of cognitive architectures have arisen as computer models of psychological and experimental data, the most prominent of which include Soar, ACT-R, 4CAPS and EPIC. These systems are fundamentally frameworks or languages for constructing production systems (that is, they are rule-based systems). Despite the vast differences between the machinery of human thought and production systems, these cognitive architectures are designed to share an architectural correspondence with conceptual models of the human mind. They incorporate structures and capabilities that resemble the abstract functions of the human mind: working memory, long term memory (including declarative and procedural knowledge), perceptions, reinforcement learners, visual imagery and even cycles of action-selection that mimic human reaction times.

The psychological inspiration of these architectures mean that they serve as both architectures for constructing intelligent systems and experimental models for understanding human intelligence. In particular, the ACT-R architecture has been applied to numerous psychological experiments in an attempt to provide a computational account for human behaviors.

Thus, these systems are architectures for constructing rule-based systems within a paradigm inspired by human cognition. In this sense, they share many of the difficulties of other rule-based and semi-formal methods. While applying a system like Soar, or any other cognitive architecture, will simplify the development of systems that use knowledge in an intelligent and 'human-like’ manner, successful systems development requires careful development of production rules that will bring about appropriate behaviors in the particular domain.