Symbolic Systems
The classical approach to AI based on symbols and rules
Symbolic Systems
Symbolic AI, also known as Good Old-Fashioned AI (GOFAI), is based on the premise that intelligence can be achieved through the manipulation of symbols according to logical rules.
The Physical Symbol System Hypothesis
Newell and Simon (1976) posited that "a physical symbol system has the necessary and sufficient means for general intelligent action."
Key Components
- Symbols: Physical patterns representing entities in the world.
- Expressions: Combinations of symbols.
- Rules: Operations for transforming expressions.
Strengths of Symbolic AI
- Compositionality: Symbols can be combined in infinitely many ways to express complex ideas.
- Explainability: The "reasoning" of a symbolic system is usually a sequence of logical steps that humans can understand.
- Abstraction: Excellent at representing high-level concepts and long-range dependencies.
- Data Efficiency: Rules can be hand-coded or learned from very few examples (given a strong prior).
Historical Milestones
Logic Theorist and GPS
The first programs designed to mimic human problem-solving and prove mathematical theorems.
Expert Systems
Domain-specific systems like MYCIN (medical diagnosis) and R1 (computer configuration) that used thousands of "IF-THEN" rules.
Cyc
The most ambitious symbolic AI project, aimed at encoding millions of pieces of human "common sense" knowledge into a massive logic-based system.
The "AI Winter" and Limitations
- Knowledge Acquisition Bottleneck: Hand-coding every rule of the world is impractically slow.
- Brittleness: Symbolic systems often fail when faced with slightly novel situations or noisy data.
- The Grounding Problem: Symbols need to be connected to the physical world (perception) to have "meaning."
- Inability to handle ambiguity: Classical logic is usually "binary," whereas the real world is probabilistic.
Symbolic AI in Contemporary AGI
Symbolic AI is making a comeback as part of hybrid systems. It is used for:
- Formal Verification: Ensuring AI safety and correctness.
- Automated Reasoning: Proving theorems and checking logic.
- Structure Planning: High-level task decomposition.