XLAR—Cognitive architecture for intelligent action

摘要

The Artificial Intelligence (AI) reductionist approach (i.e., considering knowledge representation, reasoning, learning issues separately, and combining them later), especially for solving large real-world problems, proved to be ineffective. As a result, unifying theories and integrated computational architectures are emerging at a fast rate in the AI scene. This article discusses one such investigation—XLAR (Cognitive Architecture for Intelligent Action)—an integrated cognitive architecture unifying knowledge representation (KR), reasoning, and multiple learning methods. XLAR uses a unique unified KR scheme called CGRAF, based on Conceptual Graphs (CGs), rules, and frames, and its explicit knowledge base organization drives the layered (humanlike) inferencing and includes a unique learning mechanism called Unconscious Learning in addition to the other basic learning mechanisms Chunking, Caching, and Explanation-based Learning. This has been implemented on the TI Explorer machine and tested on three domains: one nonformal domain—Agriculture—and two formal domains—Trigonometry and Games and Puzzles.