Theory of Mind and Self-Attributions of Mentality are Dissociable in LLMs

2026-03-30

AI Summary

This study investigates whether safety fine-tuning impairs large language models’ theory of mind (ToM) by suppressing their self-attribution of mental states. Through ablation experiments on safety fine-tuning, representational similarity analysis, and ToM evaluation tasks, the work provides the first evidence that self-attribution of mental states and ToM are behaviorally and mechanistically dissociable in these models. The findings indicate that safety fine-tuning does not diminish overall ToM capabilities; however, it significantly reduces the models’ attribution of mental states to non-human animals and decreases expressions of spiritual beliefs. These results suggest that current alignment practices may inadvertently constrain models’ capacity to represent diverse forms of mindedness beyond human-centric perspectives.

Abstract

Safety fine-tuning in Large Language Models (LLMs) seeks to suppress potentially harmful forms of mind-attribution such as models asserting their own consciousness or claiming to experience emotions. We investigate whether suppressing mind-attribution tendencies degrades intimately related socio-cognitive abilities such as Theory of Mind (ToM). Through safety ablation and mechanistic analyses of representational similarity, we demonstrate that LLM attributions of mind to themselves and to technological artefacts are behaviorally and mechanistically dissociable from ToM capabilities. Nevertheless, safety fine-tuned models under-attribute mind to non-human animals relative to human baselines and are less likely to exhibit spiritual belief, suppressing widely shared perspectives regarding the distribution and nature of non-human minds.

Problem

• Theory of Mind
• Self-Attribution
• Mind Attribution
• Safety Fine-tuning
• Large Language Models

Innovation

• Theory of Mind
• mind-attribution
• safety fine-tuning
• representational similarity
• mechanistic dissociation