Abstract:

The rapid proliferation of generative AI has exposed a critical gap in human-AI collaboration frameworks. While Prompt Engineering has matured as a discipline for optimizing individual prompts, it addresses only the tactical and executional dimensions of dialogue. The strategic dimension—how humans should structure the sequence and quality of their thinking across an entire AI-assisted inquiry—remains untheorized. This paper introduces the Mon-Gaku Cycle and the Three-Layer Prompt Architecture to fill that gap.

We first propose the Three-Layer Prompt Architecture, distinguishing:
(1) the Prompt Strategy Layer, which governs how a thinker sequences and evolves questions across a dialogue;
(2) the Prompt Tactics Layer, corresponding to existing Prompt Engineering techniques such as Chain-of-Thought; and
(3) the Prompt Execution Layer, referring to individual prompt strings.

Prior research has concentrated almost exclusively on layers two and three. The Mon-Gaku Cycle constitutes the first explicit model for the Strategy Layer, enabling humans to break through the structural ceiling of prompt-level optimization.

The Mon-Gaku Cycle—named from the Japanese for questioning (Mon) and learning (Gaku)—comprises four human-driven activities that recur in a spiral:
(1) Judge, evaluating AI output and deciding to accept, revise, or reject;
(2) Dissent, returning counterexamples, alternative perspectives, or critical concerns;
(3) Reflect, reviewing the trajectory of reasoning and examining cognitive biases; and
(4) Reframe, reconstructing the problem framing and re-entering the cycle with a sharper question.

The AI occupies the center of the cycle not as an answer generator but as a catalyst for human intellectual deepening.

Central to the cycle is propositional thinking: the disposition to approach AI dialogue with a formed proposition ("Is it not the case that X?") rather than an open question ("What is X?"). This shift transforms the human from passive recipient to critical evaluator, driving all four activities and producing a Shift-Left effect—by sharpening questions upstream, the thinker eliminates downstream rework and accelerates overall inquiry. Our guiding principle encapsulates this: "The hasty thinker should first judge."

Sustained practice cultivates four transferable capabilities—Judgment, Dissent, Reflection, and Refinement—that function beyond AI contexts in debate, research, and organizational decision-making. To support development, we introduce the Mon-Gaku Cycle Maturity Model (MCMM), a five-level framework from AI-Dependent (Level 1) to Embodying (Level 5), enabling learners to objectively verify their growth—a feature absent from existing AI literacy frameworks.

The novelty of this work is threefold: the Three-Layer Prompt Architecture as a theoretical framework that first delineates strategic from tactical AI engagement; the Mon-Gaku Cycle as the first cyclic model embedding critical dissent as a mandatory structural activity, addressing sycophantic AI alignment; and the MCMM as an instrument that renders AI mastery objectively measurable. Together, these contributions reframe human-AI collaboration not as a question of prompt technique, but as a question of intellectual character. Mastery of generative AI is inseparable from the cultivation of propositional thinking—the capacity to hold a proposition, subject it to rigorous dialogue, and emerge with a question of higher quality than the one with which one began.

Keywords:

Mon-Gaku Cycle, Human-AI Collaboration, Prompt Architecture, Propositional Thinking, Maturity Model, AI Literacy, Critical Thinking, Generative A.