Cognitive Architectures
Cognitive architectures are comprehensive theories of the structure and function of the mind, providing a formal framework for simulating human cognition in Artificial Intelligence systems. They model information processing, memory systems, and control structures to explain how intelligent behavior emerges.
Core Components
- Memory Systems: Subsystems for storing and retrieving information.
- Episodic Memory: Context-specific past experiences.
- Semantic Memory: General factual knowledge.
- Procedural Memory: Skills and action sequences.
- Working Memory: Temporary storage for active manipulation.
- Production Rules: Condition-action pairs that drive behavior.
- Attentional Control: Mechanisms for selecting relevant stimuli or internal states.
- Learning Mechanisms: Processes for modifying architecture parameters based on experience.
Key Models
- ACT-R: Adaptive Control of Thought-Rational; focuses on declarative and procedural memory.
- SOAR: Problem-solving spaces and production systems.
- CLARION: Dual-process model with implicit and explicit representations.
- ACT-R: Emphasizes modular memory structures.
Relevance to AI Agents
Modern large-language-models (LLMs) often lack persistent state, necessitating external cognitive architectures to achieve agentic behaviors. Integrating structured memory types allows AI agents to maintain context, learn from interactions, and plan long-term.
- See: AI Agent Memory Types: CoALA Framework Overview for a breakdown of the four memory types (short-term, factual, learned skills, and planning) proposed by IBM’s CoALA framework, drawing direct parallels to human cognitive structures.