Interactive Agents: Simulating Counselor-Client Psychological Counseling via Role-Playing LLM-to-LLM Interactions

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Conversational AI · Depth: Expert, medium

Summary

The "Interactive Agents" framework, presented by Huachuan Qiu and Zhenzhong Lan at *SEM 2026, simulates naturalistic psychological counseling dialogues through controlled LLM-to-LLM interactions. This novel framework addresses challenges like privacy concerns, high costs, and limited scalability associated with collecting real counselor-client conversation data. It features two key innovations: a personalized client agent that maintains consistent psychological characteristics, and a counselor agent that implements a theoretically grounded three-stage therapeutic model (exploration, insight, and action phases). Rigorous evaluation, including automatic metrics and professional-counselor assessments based on the Working Alliance Inventory, demonstrates that the framework generates therapeutically valid dialogues comparable to human-generated sessions. Models fine-tuned on its synthetic dataset, SimPsyDial, achieve state-of-the-art performance in LLM-based counselor evaluations.

Key takeaway

For AI Scientists and Machine Learning Engineers developing mental health support systems, this framework offers a scalable, privacy-preserving solution for data generation. You can overcome the scarcity of high-quality counseling dialogue data by leveraging LLM-to-LLM simulations. Consider adopting the personalized client and three-stage counselor agent designs to ensure therapeutic validity and achieve state-of-the-art performance in your LLM-based counselors.

Key insights

LLM-to-LLM simulation with specialized agents generates high-quality, therapeutically valid counseling dialogue data.

Principles

• Synthetic data can achieve human-comparable therapeutic validity.
• Agent design benefits from consistent psychological characteristics.
• Counselor agents should follow grounded therapeutic models.

Method

The framework uses a personalized client agent with consistent psychological traits and a counselor agent following a three-stage therapeutic model (exploration, insight, action) to simulate dialogues.

In practice

• Generate large-scale counseling datasets.
• Fine-tune LLMs for mental health support.
• Develop specialized therapeutic AI agents.

Topics

• LLM-to-LLM Interaction
• Psychological Counseling
• Synthetic Data Generation
• Mental Health AI
• Dialogue Systems
• Agent-based Simulation

Best for: Research Scientist, AI Engineer, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• An Agentic LLM-Based Framework for Population-Scale Mental Health Screening — An agentic LLM framework enhances mental health screening by ensuring robust, adaptable, and trustworthy AI pipelines.
• Fake users generated by AI can't simulate humans — review of 182 research papers — LLMs fail to simulate human cognition and behavior due to their training on text, not diverse lived experience.
• Psy-Chronicle:A Structured Pipeline for Synthesizing Long-Horizon Campus Psychological Counseling Dialogues — Long-horizon psychological counseling data can be synthesized by modeling student profiles, temporal stress events, and cross-session memory.
• Advancing E-commerce Merchants Telemarketing with Synthetic Data-Driven LLMs — Synthetic data generation can significantly improve LLM performance in complex, strategy-driven telemarketing dialogues.
• Young people's perceptions and recommendations for conversational generative artificial intelligence in youth mental health — Young people prioritize transparency, human connection, and personalized, safe integration of genAI chatbots in mental health.
• SocialWise: LLM-Agentic Conversation Therapy for Individuals with Autism Spectrum Disorder to Enhance Communication Skills — SocialWise offers scalable, LLM-agentic conversation therapy for ASD, providing instant, structured feedback via a browser-based platform.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.