Reasoning or memory? Has ChatGPT developed a human-like theory of mind?
Year: 2025
Type: dissertation
Abstract: Theory of mind (ToM) is a human cognitive capability that allows individuals to effectively communicate and engage in interpersonal relationships, through the ability to attribute desires, emotions, beliefs, intentions and mental states, to oneself and others. Artificial Intelligence (AI) engineers have long endeavoured to replicate these human capabilities with Large Language Models (LLMs), which has critical implications, particularly in job replacement. Studies have been reporting LLMs high performance in standardized ToM tests, but without disentangling possible memory effects due to previous exposure to such tests, which is a research gap. This study addressed this gap by comparing human and LLMs performance in the original and novel ToM tests. Findings show not only LLMs have high levels of performance across original and new ToM tests but also outperform humans in both versions. This discourages the memory-effect hypothesis and supports the contention that LLMs have acquired ToM human-like capabilities already.
Source: Repositório do ISCTE-IUL
Author: Araújo, Maria Clara Vieira
Language: English
Subfield: Developmental and Educational Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7219476312
