City Research Online

Abstract

Large Language Models (LLMs) such as GPT‑5 are increasingly consulted for advice across a wide range of domains, yet little is known about how their probability judgments compare to those of humans. This study examined GPT‑5’s adherence to classical probability rules, focusing on conjunction fallacies, disjunction fallacies, and violations of binary complementarity. Using a large dataset on human probabilistic judgments, in which participants displayed multiple types of fallacies, we tested GPT‑5 on the same task and with matched participant profiles. GPT‑5 produced only single conjunction or disjunction fallacies and showed near‑perfect compliance with binary complementarity constraints. Its overall response pattern aligned with predictions of early quantum‑probabilistic models rather than more recent variants incorporating noise. These findings suggest that GPT‑5 implements a more coherent and internally consistent form of probabilistic reasoning compared to naïve human participants.

Publication Type:	Article
Additional Information:	© 2026 Imannezhad, Pothos and Wills. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
Publisher Keywords:	AI participants (AI subjects), complementarity, conjunction fallacy, disjunction fallacy, GPt-5, human vs. AI cognition, large language models (LLMs), probabilistic reasoning
Subjects:	B Philosophy. Psychology. Religion > BF Psychology Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Departments:	School of Health & Medical Sciences School of Health & Medical Sciences > Department of Psychology & Neuroscience
SWORD Depositor:	Symplectic Administrator

Export

Downloads