Defining Knowledge: Bridging Epistemology and Large Language Models

TL;DR

This paper examines the concept of 'knowledge' in large language models by reviewing epistemological theories, analyzing current NLP practices, and proposing evaluation methods aligned with philosophical standards.

Contribution

It bridges epistemology and NLP by formalizing knowledge interpretations for LLMs, identifying conceptual gaps, and proposing standardized evaluation protocols.

Findings

Inconsistencies in current NLP knowledge assessments

Philosophers and computer scientists differ on LLMs' knowledge status

Proposed evaluation protocols align with epistemological standards

Abstract

Knowledge claims are abundant in the literature on large language models (LLMs); but can we say that GPT-4 truly "knows" the Earth is round? To address this question, we review standard definitions of knowledge in epistemology and we formalize interpretations applicable to LLMs. In doing so, we identify inconsistencies and gaps in how current NLP research conceptualizes knowledge with respect to epistemological frameworks. Additionally, we conduct a survey of 100 professional philosophers and computer scientists to compare their preferences in knowledge definitions and their views on whether LLMs can really be said to know. Finally, we suggest evaluation protocols for testing knowledge in accordance to the most relevant definitions.