Chatting with an LLM-based AI elicits affective and cognitive processes in education for sustainable development

Abstract

Abstract Personalized interactions have been discussed as beneficial for learning for decades. Now, with the rise of generative artificial intelligence (GenAI), personalized artificial human-like conversations may impact the quality of learning. Manipulating system prompts to design personalities has the potential to enhance the quality of conversation with Large Language Model (LLM)-based AI. However, it is still uncertain exactly to what extent the emotional tone of a generative AI chatbot is relevant for learning. Hence, the current study evaluates the impact of a chat-based conversation with an LLM-based AI on relevant affective (empathy, compassion, distress) and cognitive (perspective-taking, reflection, knowledge) processes in education for sustainable development. Here, the focus is on both the general impact and the particular impact of two different system prompts that assign the AI’s specific personality traits (empathic vs. compassionate). Comparing these two groups and one control group reading a text ( N  = 122) indicates that chatting with an empathic AI can elicit stronger emotions (e.g., empathy, compassion, distress) compared to chatting with a compassionate AI, and compared to the control. Although all groups gained knowledge, we found no group differences. Further research is necessary to ensure reliable and contextually appropriate conversations in the context of education.