AutoCBT: An Autonomous Multi-agent Framework for Cognitive Behavioral Therapy in Psychological Counseling

Paper · arXiv 2501.09426 · Published January 16, 2025
Therapy Practice and AI

online automated counseling offers a potential solution for those hesitant to seek help due to feelings of shame. Cognitive Behavioral Therapy (CBT) is an essential and widely used approach in psychological counseling. The advent of large language models (LLMs) and agent technology enables automatic CBT diagnosis and treatment. However, current LLM-based CBT systems use agents with a fixed structure, limiting their self-optimization capabilities, or providing hollow, unhelpful suggestions due to redundant response patterns. In this work, we utilize Quora-like1 and YiXinLi2 single-round consultation models to build a general agent framework that generates high-quality responses for single-turn psychological consultation scenarios. We use a bilingual dataset to evaluate the quality of single-response consultations generated by each framework. Then, we incorporate dynamic routing and supervisory mechanisms inspired by real psychological counseling to construct a CBT-oriented autonomous multi-agent framework, demonstrating its general applicability.

posits that individuals experiencing emotional or psychological problems often exhibit logical errors, such as catastrophizing, labeling, or minimizing positive affirmations. These logical errors, also referred to as thinking errors, can distort an individual’s perception of their situation, experiences, or personality, hinder daily functioning, and reduce their quality of life (Hofmann, 2011).

AutoCBT incorporates a routing mechanism and memory mechanism to enhance the autonomous ability of each agent. Each message generated by AutoCBT undergoes a structured process of reasoning and editing to meet specific requirements. From the perspective of CBT, AutoCBT allows for more flexible topology adaptation as therapeutic techniques evolve, and we also demonstrate its efficacy in other purely prompt-based psychological counseling framework.

Upon receiving the original user question, it will forward the message based on professional judgment and seek extra information from different supervisor agents, until the counsellor agent has sufficient confidence to answer the user’s question.

Generally speaking, there will be five different

types of strategies which are defined as following:

  1. [LOOPBACK] Loop back, continue with the

statement.

  1. [UNICAST] Unicast, send to a communicable agent.

  2. [MULTICAST] Multicast, send to several

communicable agents.

  1. [BROADCAST] Broadcast, send to all communicable

agents.

  1. [ENDCAST] Terminated casting, end communication

with the specified agent.

Synthesis notes from this paper's topics 8

This paper has no direct semantic matches yet — here are notes from the topic(s) it's filed under.

Lines of inquiry this paper opens 24

Research framings built by reading the notes related to this paper — the questions it feeds into.

Why do LLMs excel at generating but assessing content poorly? Does AI ideation collapse diversity despite appearing novel? Can human psychological models explain how LLMs actually reason? Can peer review policies effectively prevent undisclosed LLM use? How can we ensure benchmarks measure genuine progress rather than artifacts? Do relational factors like alliance and empathy determine AI therapy outcomes? How should agents learn and use user preferences during conversation? How do open and closed AI models compare in capability and value? Do language models learn genuine linguistic structure or surface-level patterns? What distinguishes evidence-based reasoning from persuasive framing in discourse? How do emotions function as information in reasoning and understanding?