Symbolic AI

concept · updated Jul 23, 2026

person concept tool org talk claim — click a node to jump to its page; hover an arrow for the relation

A paradigm of artificial intelligence based on explicit rule-based systems and formal knowledge representation, as contrasted with modern neural and statistical approaches such as LLM|large language models.

Characterization and Scope

Frank Coyle (UC Berkeley) places rule-based systems within the broader category of symbolic AI, distinguishing it from the connectionist/neural approaches that dominate contemporary AI development. 1:07:19

Historical Context and Revival

Coyle draws an explicit parallel between today's agentic AI architectures and the symbolic AI tradition of the 1980s, arguing that current agentic AI loops represent a return to and extension of expert systems from that era: "we're going back to the world of expert systems." 3:55:30 This framing suggests that the structured, rule-governed reasoning embedded in modern agent pipelines recapitulates ideas that predated the deep learning era.

Relationship to Knowledge Graphs

Coyle also connects symbolic AI's legacy to contemporary Knowledge Graph|knowledge graphs and graph databases, noting that knowledge representation work in that tradition "relates to what people are doing now with graph databases and knowledge representation." 52:12 In this view, Knowledge Graph|knowledge graphs are not simply an alternative to symbolic AI but a modern descendant and practical instantiation of it.

Position in the AI Landscape

Within the talk's framing, symbolic AI and LLMs occupy complementary or competing roles: symbolic approaches offer deterministic, auditable rule application, while LLMs provide flexible natural-language reasoning. The argument for ontologies and knowledge graphs in agentic systems implicitly rehabilitates symbolic AI methods as necessary complements to purely statistical models.