Are language models rational? The case of coherence norms and belief revision

TL;DR

This paper examines whether language models adhere to rational coherence norms, introducing a new credence measure based on internal probabilities, and finds that some models do align with these norms while others do not.

Contribution

It introduces the Minimal Assent Connection (MAC) and a novel credence account, providing a framework to evaluate rational coherence in language models.

Findings

Some language models satisfy coherence norms

The new credence measure correlates with model behavior

Implications for AI safety and alignment

Abstract

Do norms of rationality apply to machine learning models, in particular language models? In this paper we investigate this question by focusing on a special subset of rational norms: coherence norms. We consider both logical coherence norms as well as coherence norms tied to the strength of belief. To make sense of the latter, we introduce the Minimal Assent Connection (MAC) and propose a new account of credence, which captures the strength of belief in language models. This proposal uniformly assigns strength of belief simply on the basis of model internal next token probabilities. We argue that rational norms tied to coherence do apply to some language models, but not to others. This issue is significant since rationality is closely tied to predicting and explaining behavior, and thus it is connected to considerations about AI safety and alignment, as well as understanding model behavior more generally.