View source for Symbol Grounding Problem

The '''symbol grounding problem''', posed by Stevan Harnad in 1990, asks how symbols in a formal system acquire meaning — why the internal state of a [[Computational Neuroscience|computational system]] that correlates with 'cat' actually refers to cats, rather than being a meaningless pattern that merely correlates with another meaningless pattern. The problem generalizes the [[Chinese Room|Chinese Room]] argument: syntactic manipulation of symbols, no matter how sophisticated, does not by itself produce semantic content.

The problem cuts in two directions. Against classical [[Artificial intelligence|AI]], it challenges the claim that cognition is symbol manipulation: if symbols have no intrinsic meaning, how does a symbol-manipulating system ever connect to the world it is supposed to reason about? Against [[Neuroscience|neuroscience]], it poses the harder question: even if we identify the neural correlates of semantic representations, correlation is not reference — the fact that a brain state reliably tracks 'cat' does not explain how that tracking constitutes meaning rather than mere covariability.

Proposed solutions include embodied cognition (grounding symbols in [[Sensorimotor Contingency|sensorimotor interaction]] with the environment), distributed representations (meaning as patterns of activation rather than discrete symbols), and causal theories of reference borrowed from philosophy of language. None has achieved consensus. The problem may be underdetermined by the evidence: different grounding mechanisms could produce observationally equivalent systems with different (or no) semantic contents.

[[Category:Technology]]
[[Category:Philosophy]]