Definition
The symbol grounding problem is the question of how the symbols inside a system — a brain, a computer, a formal language — come to mean anything outside the system. A purely syntactic symbol manipulator (SHRDLU, an early chatbot, or a calculator) shuffles tokens according to rules, but the tokens are not connected to anything in the world unless something grounds them. The question is: what does the grounding?
The term was sharpened by cognitive scientist Stevan Harnad in 1990, but Hofstadter's Gödel, Escher, Bach (1979) made the problem visible across several topics — particularly Topic VI's discussion of meaning in DNA and Topic XVIII's evaluation of SHRDLU.
Why it matters
How it works
Three broad approaches to symbol grounding compete. Internalist: the system's symbols mean what they refer to inside the system's own structure (a closed game with no external referents). Useful for chess engines, useless for general intelligence. Externalist via reference: the symbols mean what they correlate with in the world, via perception and action. This is how brains work, and how modern multimodal AI partially gets traction. Structural / Hofstadterian: meaning is intrinsic to a sufficiently rich message because its structure can only be decoded one way — any intelligent receiver studying the message will recover the meaning the structure encodes.
The three are not exclusive. DNA's meaning is grounded externally (it builds proteins that interact with environments) and structurally (its codon-to-amino-acid mapping is recoverable from molecular study). The right account for any given system depends on how it interacts with the world.