The convergence and deviation of ChatGPT and human individual emotional reactions:A comparative study based on social evaluation
With the widespread application of the generative language model ChatGPT in the social domain,it is worth investigating whether there is a convergence or divergence in emotional responses between human individuals and ChatGPT in social evaluation.92 participants and 92 trials of ChatGPT rated the pleasure level of 180 evaluation sentences(praising or criticizing appearance or personality)on a scale of 1～7.The results showed that the overall pleasure rating of ChatGPT was higher than that of human participants when receiving social evaluations.Further analysis revealed that although ChatGPT scored higher in praise(whether it was appearance or personality)than human participants,it scored lower in criticism,especially in criticism of personality.In summary,ChatGPT exhibited stronger emotional responses to both positive and negative evaluations compared to human participants.These findings suggest that ChatGPT's approach to simulating human emotions tends to be overly simplistic,and its emotional responses currently fall short of aligning with the complex and nuanced emotional mechanisms of humans.
artificial intelligence(AI)large language modelChatGPTsocial evaluationemotional responselanguage
万方数据
