Understanding Understanding: A Pragmatic Framework Motivated by Large Language Models

TL;DR

This paper introduces a pragmatic framework for testing whether agents, including AI models like LLMs, truly understand a subject by evaluating their performance on a carefully scoped set of questions, using probabilistic methods to ensure confidence.

Contribution

The paper proposes a novel, performance-based framework for assessing understanding in agents, incorporating probabilistic sampling and explanation-based improvements, applicable to AI and humans.

Findings

Current LLMs do not fully understand nontrivial domains.

The framework allows high-confidence testing of understanding through random sampling.

Explanation of answers reduces the sample complexity needed for confidence.

Abstract

Motivated by the rapid ascent of Large Language Models (LLMs) and debates about the extent to which they possess human-level qualities, we propose a framework for testing whether any agent (be it a machine or a human) understands a subject matter. In Turing-test fashion, the framework is based solely on the agent's performance, and specifically on how well it answers questions. Elements of the framework include circumscribing the set of questions (the "scope of understanding"), requiring general competence ("passing grade"), avoiding "ridiculous answers", but still allowing wrong and "I don't know" answers to some questions. Reaching certainty about these conditions requires exhaustive testing of the questions which is impossible for nontrivial scopes, but we show how high confidence can be achieved via random sampling and the application of probabilistic confidence bounds. We also show that accompanying answers with explanations can improve the sample complexity required to achieve acceptable bounds, because an explanation of an answer implies the ability to answer many similar questions. According to our framework, current LLMs cannot be said to understand nontrivial domains, but as the framework provides a practical recipe for testing understanding, it thus also constitutes a tool for building AI agents that do understand.