The article presents a novel method, Chain of Empathy (CoE) that uses insights from psychotherapy to enable Large Language Models (LLMs) to reason about human emotions. CoE prompting is inspired by Cognitive-Behavioral Therapy (CBT), Dialectical Behavior Therapy (DBT), Person-Centered Therapy (PCT), and Reality Therapy (RT), which interpret mental states differently. The study found that LLMs using CoE reasoning produced a wider range of empathetic responses. The CBT-based CoE provided the most balanced generation of empathetic responses. This research highlights the significance of understanding emotional context in human-AI communication and contributes to the development of context-specific, safer, and empathetic AI.
Publication date: 13 Nov 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2311.04915
