Enhancing AI Communication: How BIT Research Alliance Improves Professionalism and Readability in Chronic Disease Management with the RaR Multi-Agent Method


Introduction: The “Last Mile” of AI Communication – Clarity, Professionalism, and Comprehensibility

As Artificial Intelligence (AI) becomes increasingly integrated into the healthcare domain, we not only expect AI to possess powerful analytical and decision-making capabilities but also place higher demands on its “communication skills.” Whether it’s explaining complex medical conditions to patients, providing health guidance, or collaborating with healthcare professionals, AI-generated content needs to be professional, legible/readable, and accurate. However, content directly generated by Large Language Models (LLMs) can sometimes be overly technical, not conversational enough, or may not fully align with communication goals in specific contexts.

To address this “last mile” challenge, BIT Research Alliance has developed and applied a multi-agent method called “Two-step Rephrase and Respond (RaR)”. This method aims to significantly enhance the expressive effectiveness of AI in key communication scenarios like chronic disease management, ensuring that information is conveyed in a manner that is both professional and easy to understand.

The Core Mechanism of the RaR Multi-Agent Method

The core of the RaR method lies in introducing a “Rephrasing” stage and combining multiple AI agents with different responsibilities to jointly optimize the final output content. Its operational flow can be summarized as:

  1. Original Question/Input:
    • The system receives a question from the user (e.g., a patient) or a scenario requiring an AI response.
    • For example: “Take the last letters of the words in ‘Edgar Bob’ and concatenate them.”
  2. Step One: Rephrasing Agent (LLM):
    • Task: The primary task of this agent is to understand the core intent of the original question and “rephrase” or “expand” it into a clearer format that is easier for subsequent AI agents to understand and respond to.
    • Goal: To ensure all important information from the original question is preserved while eliminating ambiguity and making the question more structured.
    • Example Output (Rephrased): “Can you identify and extract the final letters in both the words that form ‘Edgar Bob’, and then join them together in the order they appear?”
    • As seen, the rephrased version is more detailed and explicitly outlines the operational steps.
  3. Step Two: Responding Agent (LLM):
    • Task: This agent receives the “rephrased question” and generates the final answer to the original question based on this clearer version.
    • Goal: To ensure the accuracy and relevance of the answer.
    • Example Output (Response): “The last letters in the words “Edgar Bob” are “r” and “b”. Concatenating them in the order they appear would be “rb”.”

Application and Benefits of RaR in Chronic Disease Management Communication

In chronic disease management, patients often need to understand complex medical concepts and follow detailed health guidelines. If AI-generated content is too rigid or filled with jargon, it can easily confuse patients, thereby affecting their adherence. The RaR method enhances communication effectiveness in the following ways:

  • Enhancement of Professionalism:
    • Semantic Clarification in Rephrasing Stage: Even if the original question is colloquial or imprecise, the rephrasing agent can transform it into an expression more aligned with medical communication norms, laying the foundation for generating professional answers subsequently.
    • Knowledge Calibration in Responding Stage: When responding, it can be integrated with knowledge graphs or other professional knowledge bases (like the aforementioned KG-RAG) to ensure the medical accuracy of the answer.
    • Results: In communication scenarios for Diabetes and Chronic Kidney Disease (CKD), the application of RaR optimization led to an average 14.29% improvement in professionalism (from 77 to 88).
  • Improvement in Readability:
    • User Perspective Shift in Rephrasing Stage: The rephrasing agent can convert technical questions or statements into language that is more accessible to patients.
    • Simplification and Explanation in Responding Stage: When generating answers, the responding agent can be guided to use more common vocabulary and provide brief explanations for necessary technical terms.
    • Results: In the same scenarios, after RaR optimization, readability improved by an average of 5.81% (from 86 to 91).
  • Highlights of Chronic Disease Management Communication:
    • Clearer Health Guidance: For example, when AI explains dietary control principles or drug side effects, RaR can ensure this information is both accurate and easy for patients to understand and remember.
    • More Empathetic Interactions: The rephrasing process can incorporate more empathetic and encouraging tones, making AI responses feel more human-centric.
    • Reduced Misunderstandings and Communication Barriers: Clear and understandable communication effectively reduces health risks caused by patients misunderstanding medical advice.

BIT Research Alliance’s Practice and Outlook

BIT Research Alliance views the RaR multi-agent method as a key component in optimizing human-computer interaction experiences and enhancing the quality of AI services. We are not only applying it to patient-facing communication but also exploring its potential in assisting healthcare professionals with report writing, academic exchange, and other scenarios.

In the future, we will further research:

  • More Granular Agent Role Definitions: Designing more targeted rephrasing and responding strategies based on different communication goals and audiences.
  • RaR Application in Multi-turn Dialogues: How to dynamically adjust RaR strategies in continuous multi-turn conversations to maintain communication coherence and effectiveness.
  • Integration with Affective Computing: Enabling rephrasing and responding agents to better understand and respond to users’ emotional states, providing interactions with more emotional warmth.

Conclusion: RaR — Making Human-AI Communication Smoother and More Effective

Clear, professional, and comprehensible communication is the cornerstone of successful AI application in the medical field. The “Two-step Rephrase and Respond” (RaR) multi-agent method proposed by BIT Research Alliance offers an effective solution for optimizing AI’s communication capabilities. By achieving a better balance between professionalism and readability, RaR is helping us build more trustworthy and widely accepted medical AI systems, ultimately benefiting every individual in need of health support.


發佈留言

發佈留言必須填寫的電子郵件地址不會公開。 必填欄位標示為 *

返回頂端