Understanding Artificial Theory of Mind: Perturbed Tasks and Reasoning in Large Language Models
This work addresses the debate on whether LLMs have genuine ToM, with incremental insights into robustness and prompting effects for AI researchers.
The study investigated whether large language models (LLMs) exhibit robust Theory of Mind (ToM) capabilities by testing them on perturbed false-belief tasks and using Chain-of-Thought prompting (CoT) to enhance performance and explain decisions. Results showed a steep drop in ToM capabilities under perturbations for all LLMs, questioning robust ToM presence, and while CoT improved overall performance faithfully, it degraded accuracy for some perturbation classes, indicating selective application is needed.
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.