[1N3-GS-9] Human interface:

[1N3-GS-9-03] Approaching Empathy Enhancement in Conversational AI

〇Yuuka Kasai¹, Rika Shiraishi¹, Mitsuhiro Hayase¹ (1. Sugiyama Jogakuen University)

Keywords:Large-scale Language Model, Empathy, Conversational AI, User Experience

This study proposes a learning model aimed at enhancing the empathy within interactive AI systems. Specifically, it utilizes "A Japanese Empathetic Dialogue Speech Corpus", training it with LoRA to develop a model that allows AI to understand and mimic human emotions and social interactions profoundly. This approach is anticipated to facilitate the creation of dialogue systems capable of delivering responses that are both natural and empathetic, tailored to the user's emotional state and situation, thereby rendering AI-human interactions more lifelike. To assess the model's performance, experiments were conducted to evaluate the impressions of the model's outputs before and after the learning process. The findings revealed significant difference in perceived friendliness, vitality, trustworthiness, and obedience, indicating that the responses generated by the AI have become of higher quality. Such improvements suggest a novel direction for the progression of interactive AI.