CLARION (cognitive architecture)
Overview
The CLARION (Connectionist Learning with Adaptive Rule Induction On-line) is a comprehensive, multi-level, hybrid cognitive architecture. It is designed to simulate and explain a wide range of cognitive phenomena, including both lower-level and higher-level cognition, with a focus on the integration of implicit and explicit processes.
Design Principles
CLARION is built on a set of design principles that guide its development and application. These principles include:
- The distinction between implicit and explicit processes: CLARION makes a clear distinction between implicit (non-conscious) and explicit (conscious) cognitive processes. This distinction is reflected in the dual-process theories of cognition, which posits that cognition is a product of two distinct systems or processes – one that operates automatically and without conscious awareness, and another that operates consciously and with deliberate control.
- The integration of reactive and deliberative processes: CLARION integrates reactive (automatic, non-conscious) and deliberative (conscious, controlled) processes in its architecture. This integration allows for the simulation of a wide range of cognitive phenomena, from simple associative learning to complex problem-solving and decision-making.
- The use of a hybrid approach: CLARION uses a hybrid approach that combines connectionist (neural network) and symbolic (rule-based) models. This approach allows for the representation and processing of both distributed and localist information, and for the integration of bottom-up and top-down processes.
- The emphasis on learning and adaptation: CLARION emphasizes the role of learning and adaptation in cognition. It incorporates various learning algorithms, including both supervised and unsupervised learning, and reinforcement learning.
Architecture
The architecture of CLARION consists of two main levels: the top level (the Action-Centered Subsystem, or ACS) and the bottom level (the Non-Action-Centered Subsystem, or NACS). Each level is further divided into two parts: the implicit part and the explicit part.
- The Action-Centered Subsystem (ACS): The ACS is responsible for action decision-making. It includes both an implicit part (the bottom level of the ACS), which learns and makes decisions based on reinforcement learning, and an explicit part (the top level of the ACS), which learns and makes decisions based on rule-based reasoning.
- The Non-Action-Centered Subsystem (NACS): The NACS is responsible for non-action cognitive functions, such as concept learning and semantic memory. It includes both an implicit part (the bottom level of the NACS), which learns and represents knowledge in a distributed, connectionist manner, and an explicit part (the top level of the NACS), which learns and represents knowledge in a localist, symbolic manner.
In addition to the ACS and NACS, CLARION also includes a Motivational Subsystem (MS) and a Metacognitive Subsystem (MCS). The MS is responsible for motivation and emotion, while the MCS is responsible for metacognition (thinking about thinking).
Applications
CLARION has been used to simulate and explain a wide range of cognitive phenomena. Some of the applications of CLARION include:
- Learning and memory: CLARION has been used to simulate various aspects of learning and memory, including associative learning, skill learning, and episodic memory.
- Problem-solving and decision-making: CLARION has been used to simulate problem-solving and decision-making processes, including means-ends analysis, decision tree search, and multi-attribute decision-making.
- Social cognition: CLARION has been used to simulate social cognition, including social decision-making, social influence, and cultural evolution.
- Cognitive development: CLARION has been used to simulate cognitive development, including the development of object permanence, the development of theory of mind, and the development of self-concept.