PsychAgent: AI Counselor with Lifelong Learning Skills
Researchers have introduced PsychAgent, an AI psychological counselor that learns from experience like human therapists. Unlike existing models that rely on static dialogue datasets, PsychAgent uses a Memory-Augmented Planning Engine for multi-session interactions, a Skill Evolution Engine to extract new skills from past counseling, and a Reinforced Internalization Engine to integrate these skills via rejection fine-tuning. The system aims to improve performance across diverse scenarios.
Key facts
- PsychAgent is an experience-driven lifelong learning agent for psychological counseling.
- It uses a Memory-Augmented Planning Engine for longitudinal multi-session interactions.
- A Skill Evolution Engine extracts new practice-grounded skills from historical counseling trajectories.
- A Reinforced Internalization Engine integrates evolved skills via rejection fine-tuning.
- The approach contrasts with supervised fine-tuning on static datasets.
- The paper is on arXiv with ID 2604.00931.
- The announcement type is 'replace'.
- The goal is self-evolution and improved performance across diverse scenarios.
Entities
Institutions
- arXiv