Adaptive Mutual-Human Diagnosis: A Unified Framework for Multi-LLM Interaction, Human Judgment, and Dynamic Diagnostic Switching

Contemporary artificial intelligence systems increasingly rely on diagnostic processes to identify errors, detect inconsistencies, assess operational conditions, investigate failures, evaluate decisions, and determine appropriate corrective actions. Large language models (LLMs) have demonstrated the ability to examine their own outputs, while multi-agent architectures enable multiple models to exchange information, challenge conclusions, and collaboratively identify potential errors. Human experts, meanwhile, contribute contextual understanding, professional expertise, independent judgment, and accountability to diagnostic processes. However, these capabilities are frequently implemented through separate or relatively fixed diagnostic arrangements. A fundamental research opportunity is to establish a unified architecture in which multiple LLMs can diagnose one another, humans can independently perform diagnosis, and the system can adaptively transition between these mechanisms according to diagnostic requirements. This paper introduces the Adaptive Mutual-Human Diagnosis Framework (AMHDF), a diagnosis-oriented theoretical framework developed as a functional instantiation and extension of the Universal Mutual-Human Model (UMHM). AMHDF distinguishes three principal diagnostic configurations: Mutual Diagnosis, Human Diagnosis, and Mutual–Human Diagnosis. Mutual Diagnosis involves two or more distinct artificial intelligence models independently examining, cross-questioning, evaluating, and diagnosing outputs, processes, decisions, or conditions. Human Diagnosis involves human individuals or groups independently performing diagnostic activities. Mutual–Human Diagnosis integrates multi-model interaction with human participation. The central contribution is an adaptive diagnostic architecture that does not permanently privilege any single diagnostic pathway. Instead, it introduces Adaptive Diagnostic Mechanism Selection, Adaptive Diagnostic Switching, Adaptive Diagnostic Combination, and Adaptive Diagnostic Reconfiguration. The architecture considers diagnostic uncertainty, model disagreement, evidence sufficiency, error severity, task complexity, participant capabilities, resource constraints, and human expertise when determining the appropriate configuration. A formal representation, diagnostic lifecycle, decision model, transition mechanism, theoretical propositions, application scenarios, governance principles, and empirical evaluation methodology are developed. The framework further distinguishes diagnosis from verification, correction, and final decision-making, allowing diagnostic findings to inform subsequent operations without assuming that identifying a potential problem automatically establishes its cause or determines the appropriate remedy. To the best of our knowledge, this paper is the first to explicitly formulate the particular UMHM-derived integration of multi-LLM Mutual Diagnosis and independent Human Diagnosis as a general-purpose architecture with adaptive selection, switching, combination, and structural reconfiguration. This priority claim concerns the specific theoretical integration and architecture, not the broader invention of multi-agent diagnosis, model critique, human diagnosis, or human–AI collaboration. The proposed framework provides a theoretical foundation for designing diagnostic systems that can change their diagnostic participants and mechanisms according to the requirements of individual tasks and evolving operational conditions.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-19
DOI
https://doi.org/10.5281/zenodo.24361761
Primary Topic
Human-Automation Interaction and Safety
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Adaptive Mutual-Human Diagnosis: A Unified Framework for Multi-LLM Interaction, Human Judgment, and Dynamic Diagnostic Switching

Jincheng Zhang
Zenodo (CERN European Organization for Nuclear Research)
Human-Automation Interaction and Safety
preprint

Adaptive Mutual-Human Diagnosis: A Unified Framework for Multi-LLM Interaction, Human Judgment, and Dynamic Diagnostic Switching

Jincheng Zhang
preprint en

Abstract

Contemporary artificial intelligence systems increasingly rely on diagnostic processes to identify errors, detect inconsistencies, assess operational conditions, investigate failures, evaluate decisions, and determine appropriate corrective actions. Large language models (LLMs) have demonstrated the ability to examine their own outputs, while multi-agent architectures enable multiple models to exchange information, challenge conclusions, and collaboratively identify potential errors. Human experts, meanwhile, contribute contextual understanding, professional expertise, independent judgment, and accountability to diagnostic processes. However, these capabilities are frequently implemented through separate or relatively fixed diagnostic arrangements. A fundamental research opportunity is to establish a unified architecture in which multiple LLMs can diagnose one another, humans can independently perform diagnosis, and the system can adaptively transition between these mechanisms according to diagnostic requirements. This paper introduces the Adaptive Mutual-Human Diagnosis Framework (AMHDF), a diagnosis-oriented theoretical framework developed as a functional instantiation and extension of the Universal Mutual-Human Model (UMHM). AMHDF distinguishes three principal diagnostic configurations: Mutual Diagnosis, Human Diagnosis, and Mutual–Human Diagnosis. Mutual Diagnosis involves two or more distinct artificial intelligence models independently examining, cross-questioning, evaluating, and diagnosing outputs, processes, decisions, or conditions. Human Diagnosis involves human individuals or groups independently performing diagnostic activities. Mutual–Human Diagnosis integrates multi-model interaction with human participation. The central contribution is an adaptive diagnostic architecture that does not permanently privilege any single diagnostic pathway. Instead, it introduces Adaptive Diagnostic Mechanism Selection, Adaptive Diagnostic Switching, Adaptive Diagnostic Combination, and Adaptive Diagnostic Reconfiguration. The architecture considers diagnostic uncertainty, model disagreement, evidence sufficiency, error severity, task complexity, participant capabilities, resource constraints, and human expertise when determining the appropriate configuration. A formal representation, diagnostic lifecycle, decision model, transition mechanism, theoretical propositions, application scenarios, governance principles, and empirical evaluation methodology are developed. The framework further distinguishes diagnosis from verification, correction, and final decision-making, allowing diagnostic findings to inform subsequent operations without assuming that identifying a potential problem automatically establishes its cause or determines the appropriate remedy. To the best of our knowledge, this paper is the first to explicitly formulate the particular UMHM-derived integration of multi-LLM Mutual Diagnosis and independent Human Diagnosis as a general-purpose architecture with adaptive selection, switching, combination, and structural reconfiguration. This priority claim concerns the specific theoretical integration and architecture, not the broader invention of multi-agent diagnosis, model critique, human diagnosis, or human–AI collaboration. The proposed framework provides a theoretical foundation for designing diagnostic systems that can change their diagnostic participants and mechanisms according to the requirements of individual tasks and evolving operational conditions.

Zenodo (CERN European Organization for Nuclear Research)
Human-Automation Interaction and Safety
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon