Symbolic artificial intelligence

An approach to artificial intelligence that represents knowledge as explicit symbols and manipulates those symbols through rules, logic, or search.

Knowledge representation: Methods for encoding information about the world so computational systems can use it. Symbolic AI begins by encoding facts, relations, and rules in machine-readable forms.
First-order logic: A formal logic with predicates, variables, functions, and quantifiers over objects. It expresses relationships and general rules more precisely than propositional logic.
Logic Theorist: A 1956 program by Allen Newell, Herbert A. Simon, and Cliff Shaw that proved theorems in propositional logic. It demonstrated that computers could perform tasks resembling formal reasoning.
Expert system: A program that applies encoded specialist knowledge and rules to make decisions in a domain. Expert systems became the best-known commercial application of symbolic AI.
Connectionism: An approach to cognition and computation based on networks of simple interconnected units. Connectionist systems learn distributed representations rather than relying primarily on hand-written symbols and rules.
Formal logic: The study of formal systems for expressing propositions and deriving conclusions. Logical inference lets symbolic systems derive conclusions from represented premises.
Semantic network: A graph representation in which nodes denote concepts and edges denote relations. Semantic networks encode connected facts in a form used by early symbolic AI.
General Problem Solver: A 1950s program by Newell and Simon designed to solve problems through means–ends analysis. It exemplified the search-based, general reasoning ambitions of early symbolic AI.
MYCIN: A 1970s expert system that recommended treatments for bacterial infections using rules and certainty factors. It showed how explicit medical expertise could support diagnosis and treatment recommendations.
Machine learning: Methods that enable computational systems to improve performance by learning patterns from data. Data-driven learning contrasts with symbolic systems that depend on explicitly encoded knowledge.

Linked from 17 pages