ChatGPT and Spontaneous Metacognition: Analyzing AI's Classification of Subjective Cognitive Experiences
Gull Zareen  1@  , Coline Grégoire  2@  , Chris Moulin  3@  , Akira O'connor  4@  
1 : Laboratoire de Psychologie et NeuroCognition
University of Grenoble Alpes
2 : Centre de Recherches sur la Cognition et l'Apprentissage  (CeRCA)
CNRS : UMR7295
3 : Laboratoire de Psychologie et NeuroCognition
Université Savoie Mont Blanc, Centre National de la Recherche Scientifique, Université Grenoble Alpes
4 : University of St. Andrews

Involuntary cognition includes subjective phenomena like déjà vu, tip-of-the-tongue (TOT) state and jamais vu. Given their spontaneity and subjectivity, it remains challenging to understand how these phenomena are perceived and classified. In this paper, we take an existing dataset from a reasonably large sample (n=860) across three countries (Poland, France and Pakistan) and evaluate ChatGPT's ability to classify subjective descriptions of spontaneous experiences. We compared ChatGPT's responses (versions 3.5 and 4.0) with human participants to assess its accuracy in classification. While ChatGPT showed high accuracy for déjà vu and TOTs, it struggled with jamais vu, indicating limitations in classifying less well-defined spontaneous cognition.


Personnes connectées : 3 Vie privée
Chargement...