Adaptive Self–Human Reasoning: A Universal Framework for Dynamic Switching Between Large Language Model Reasoning and Human Reasoning

Large language models (LLMs) have demonstrated increasingly sophisticated reasoning capabilities, enabling them to perform mathematical problem-solving, logical inference, scientific analysis, planning, and complex knowledge integration. Nevertheless, autonomous model reasoning is not uniformly reliable across tasks, contexts, and levels of complexity. Human reasoning, meanwhile, provides contextual understanding, experiential knowledge, critical judgment, and the capacity to interpret situations that may exceed the operational capabilities of computational models. Existing research has investigated chain-of-thought prompting, self-consistency, tree-based reasoning, self-refinement, human-in-the-loop systems, and human–AI collaborative decision-making. However, these approaches generally emphasize particular reasoning mechanisms, computational strategies, or human–AI interaction arrangements. A general mechanism-independent architecture that treats model reasoning and human reasoning as dynamically selectable, switchable, combinable, and reconfigurable functional configurations throughout the reasoning process remains insufficiently formalized. This paper introduces Adaptive Self–Human Reasoning (ASHR), a reasoning-specific theoretical framework grounded in the Universal Self–Human Model (USHM). ASHR establishes three fundamental operational configurations: Self-Reasoning (S), in which an LLM performs reasoning through its own internal computational mechanisms; Human Reasoning (H), in which a human independently performs or directs the reasoning process; and Self–Human Reasoning (S–H), in which model-based and human reasoning contribute to a shared reasoning task. Unlike approaches that permanently assign reasoning to either an AI system or a human participant, ASHR introduces Adaptive Reasoning Configuration Selection, Adaptive Reasoning Configuration Switching, Adaptive Reasoning Combination, and Adaptive Reasoning Reconfiguration. These mechanisms allow a reasoning process to move between model-directed, human-directed, and jointly organized reasoning according to task complexity, uncertainty, capability, evidence quality, computational resources, human expertise, and risk. A formal architecture is developed to represent reasoning states, configuration selection, transitions, contribution allocation, reasoning continuity, and transition costs. The framework further distinguishes changes in reasoning strategy from changes in the entity responsible for reasoning, establishing an important conceptual separation between internal reasoning adaptation and reasoning-participant switching. Human authority, independent judgment, accountability, cognitive workload, and appropriate boundaries of automation are incorporated into the architecture. To the best of our knowledge, this paper is the first to formally articulate a mechanism-independent, function-specific architecture that integrates autonomous LLM reasoning, independent human reasoning, and combined human–LLM reasoning through adaptive configuration selection, switching, combination, and reconfiguration as a unified reasoning-control principle. This priority claim concerns the specific theoretical integration and architecture proposed here, rather than the broader established fields of automated reasoning, cognitive science, human–AI collaboration, or mixed-initiative systems. The paper establishes a conceptual foundation for intelligent systems that dynamically determine who should reason, how reasoning should be organized, and when reasoning responsibility should transition between artificial and human participants.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-19
DOI
https://doi.org/10.5281/zenodo.24360006
Primary Topic
Logic, Reasoning, and Knowledge
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Adaptive Self–Human Reasoning: A Universal Framework for Dynamic Switching Between Large Language Model Reasoning and Human Reasoning

Jincheng Zhang
Zenodo (CERN European Organization for Nuclear Research)
Logic, Reasoning, and Knowledge
preprint

Adaptive Self–Human Reasoning: A Universal Framework for Dynamic Switching Between Large Language Model Reasoning and Human Reasoning

Jincheng Zhang
preprint en

Abstract

Large language models (LLMs) have demonstrated increasingly sophisticated reasoning capabilities, enabling them to perform mathematical problem-solving, logical inference, scientific analysis, planning, and complex knowledge integration. Nevertheless, autonomous model reasoning is not uniformly reliable across tasks, contexts, and levels of complexity. Human reasoning, meanwhile, provides contextual understanding, experiential knowledge, critical judgment, and the capacity to interpret situations that may exceed the operational capabilities of computational models. Existing research has investigated chain-of-thought prompting, self-consistency, tree-based reasoning, self-refinement, human-in-the-loop systems, and human–AI collaborative decision-making. However, these approaches generally emphasize particular reasoning mechanisms, computational strategies, or human–AI interaction arrangements. A general mechanism-independent architecture that treats model reasoning and human reasoning as dynamically selectable, switchable, combinable, and reconfigurable functional configurations throughout the reasoning process remains insufficiently formalized. This paper introduces Adaptive Self–Human Reasoning (ASHR), a reasoning-specific theoretical framework grounded in the Universal Self–Human Model (USHM). ASHR establishes three fundamental operational configurations: Self-Reasoning (S), in which an LLM performs reasoning through its own internal computational mechanisms; Human Reasoning (H), in which a human independently performs or directs the reasoning process; and Self–Human Reasoning (S–H), in which model-based and human reasoning contribute to a shared reasoning task. Unlike approaches that permanently assign reasoning to either an AI system or a human participant, ASHR introduces Adaptive Reasoning Configuration Selection, Adaptive Reasoning Configuration Switching, Adaptive Reasoning Combination, and Adaptive Reasoning Reconfiguration. These mechanisms allow a reasoning process to move between model-directed, human-directed, and jointly organized reasoning according to task complexity, uncertainty, capability, evidence quality, computational resources, human expertise, and risk. A formal architecture is developed to represent reasoning states, configuration selection, transitions, contribution allocation, reasoning continuity, and transition costs. The framework further distinguishes changes in reasoning strategy from changes in the entity responsible for reasoning, establishing an important conceptual separation between internal reasoning adaptation and reasoning-participant switching. Human authority, independent judgment, accountability, cognitive workload, and appropriate boundaries of automation are incorporated into the architecture. To the best of our knowledge, this paper is the first to formally articulate a mechanism-independent, function-specific architecture that integrates autonomous LLM reasoning, independent human reasoning, and combined human–LLM reasoning through adaptive configuration selection, switching, combination, and reconfiguration as a unified reasoning-control principle. This priority claim concerns the specific theoretical integration and architecture proposed here, rather than the broader established fields of automated reasoning, cognitive science, human–AI collaboration, or mixed-initiative systems. The paper establishes a conceptual foundation for intelligent systems that dynamically determine who should reason, how reasoning should be organized, and when reasoning responsibility should transition between artificial and human participants.

Zenodo (CERN European Organization for Nuclear Research)
Logic, Reasoning, and Knowledge
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