Multi-agent norm perception and induction in distributed healthcare.

Area of Science:

• Artificial Intelligence
• Multi-Agent Systems
• Healthcare Informatics

Background:

• Integrating autonomous agents into healthcare requires understanding complex medical norms.
• Medical norms are categorized into descriptive (collective tendencies) and prescriptive (ideal behaviors).
• Existing multi-agent systems face challenges in learning these distinct norm types within dynamic environments.

Purpose of the Study:

• To present a Multi-Agent Norm Perception and Induction Learning Model for autonomous agents in healthcare.
• To enable agents to simultaneously learn both descriptive and prescriptive medical norms.
• To facilitate the dynamic integration of agent systems into distributed healthcare settings.

Main Methods:

• Utilizing parameterized mixed probability density models for norm representation.
• Employing practice-enhanced Markov games for agent interaction and learning.
• Developing a system capable of perceiving descriptive norms and inducing prescriptive norms.

Main Results:

• The model successfully perceived descriptive collective medical norms, reflecting current best clinical practices.
• Experimental results showed the model's learned norms converged with real-world diagnostic trends over five years.
• Agents effectively learned key clinical protocols, avoiding belief in invalid norms, with smooth belief updates.

Conclusions:

• The developed model enables effective perception and induction of medical norms by multi-agent systems.
• This facilitates the integration of autonomous agents into healthcare by aligning with clinical best practices.
• The model demonstrates a robust approach to learning and adapting to dynamic healthcare environments.