Understanding Artificial Theory of Mind: Perturbed Tasks and Reasoning in Large Language Models

TL;DR

This paper evaluates the robustness of large language models' Theory of Mind abilities using perturbed false-belief tasks, introduces a new annotated dataset, and assesses how Chain-of-Thought prompting affects reasoning and accuracy.

Contribution

It provides a new richly annotated ToM dataset with perturbations and reasoning chains, and analyzes the impact of Chain-of-Thought prompting on LLM ToM performance.

Findings

LLMs' ToM abilities drop significantly under task perturbation

Chain-of-Thought prompting generally improves ToM reasoning faithfulness

CoT prompting can decrease accuracy on certain perturbation classes

Abstract

Theory of Mind (ToM) refers to an agent's ability to model the internal states of others. Contributing to the debate whether large language models (LLMs) exhibit genuine ToM capabilities, our study investigates their ToM robustness using perturbations on false-belief tasks and examines the potential of Chain-of-Thought prompting (CoT) to enhance performance and explain the LLM's decision. We introduce a handcrafted, richly annotated ToM dataset, including classic and perturbed false belief tasks, the corresponding spaces of valid reasoning chains for correct task completion, subsequent reasoning faithfulness, task solutions, and propose metrics to evaluate reasoning chain correctness and to what extent final answers are faithful to reasoning traces of the generated CoT. We show a steep drop in ToM capabilities under task perturbation for all evaluated LLMs, questioning the notion of any robust form of ToM being present. While CoT prompting improves the ToM performance overall in a faithful manner, it surprisingly degrades accuracy for some perturbation classes, indicating that selective application is necessary.