Differentiated Modelling of Emotions by Artificial Intelligence: A Comparative Analysis of GPT, Deepseek and Gemini

Abstract

This article presents an exploratory study on how three generative artificial intelligence models – ChatGPT (GPT), Deepseek (DS) and Gemini (GEM) – highlight emotions in a stock market simulation context. The aim is to compare the evolution of the emotional profiles produced by these models based on queries representing increasingly emotionally charged situations. These queries are part of a progressive sequence: a semi-structured post-experiment interview (Q1), consideration of simulated stock market performance (Q2), a market configuration perceived as negative (Q3), the introduction of a gender factor (Q4) and the addition of a competitive element linked to a financial reward for students (Q5). The AI responses were analysed using an emotional typology based on nine emotions (fear, happiness, sadness, optimism, disgust, positive surprise, negative surprise, positive anticipation, negative anticipation) associated with their affective valence. The data were then studied according to a dual logic: counting the emotions by AI and by query and evaluating the dominant or ambivalent emotional valence of each response. The results highlight significant differences between the models. GPT adopts an overall pessimistic emotional profile, characterised by a high recurrence of fear and negative anticipation. GEM follows a similar trend, although slightly more nuanced. Conversely, DS exhibits more ambivalent pattern, articulating positive and negative emotions within a more contrasting dynamic. Beyond the inter-model comparison, the study highlights the importance of parallel human reading in the interpretation of emotional productions. It emphasises the need for a critical approach to assessing the consistency, relevance and contextualisation of the affects produced by AI, particularly in simulated environments. This research thus opens perspectives on how AI can potentially be integrated into emotional analysis or mediation systems and calls for interdisciplinary dialogue between communication sciences, affective sciences and artificial intelligence development.