The Vector Grounding Problem

TL;DR

This paper explores whether large language models can achieve referential grounding, connecting their internal states to real-world entities, by satisfying causal and functional conditions derived from teleosemantic theories.

Contribution

It clarifies the concept of referential grounding in LLMs and argues they can attain it through causal relations and evolutionary history, without multimodal input.

Findings

LLMs can satisfy conditions for referential grounding.

Referential grounding is distinct from other forms of grounding.

LLMs do not require multimodality or embodiment for grounding.

Abstract

Large language models (LLMs) produce seemingly meaningful outputs, yet they are trained on text alone without direct interaction with the world. This leads to a modern variant of the classical symbol grounding problem in AI: can LLMs' internal states and outputs be about extra-linguistic reality, independently of the meaning human interpreters project onto them? We argue that they can. We first distinguish referential grounding -- the connection between a representation and its worldly referent -- from other forms of grounding and argue it is the only kind essential to solving the problem. We contend that referential grounding is achieved when a system's internal states satisfy two conditions derived from teleosemantic theories of representation: (1) they stand in appropriate causal-informational relations to the world, and (2) they have a history of selection that has endowed them with the function of carrying this information. We argue that LLMs can meet both conditions, even without multimodality or embodiment.