Narration-of-Thought: Inference-Time Scaffolding for Defeasible Ethical Reasoning in Large Language Models

2026-06-24

AI Summary

This work addresses the prevalent issues of stakeholder collapse and uncertainty suppression in moral reasoning by large language models through the introduction of “Narrative of Thought” (NoT)—a training-free, parameter-agnostic, structured inference-time scaffolding method. NoT decomposes ethical reasoning into five stages: protagonist identification, stakeholder analysis, two-step consequence forecasting, explicit uncertainty modeling, and final commitment generation, augmented with a multi-agent debate mechanism to reach consensus. Experimental results on the DailyDilemmas dataset demonstrate that NoT reduces stakeholder collapse to below 1% and mitigates uncertainty suppression to 1–24%. Furthermore, iterative multi-agent debates elevate consensus rates from 6% to over 95%, substantially enhancing the transparency, completeness, and interpretability of ethical reasoning.

Abstract

Standard chain-of-thought on moral dilemmas exhibits two failure modes: stakeholder collapse (the trace names at most one party with a stake in the outcome) and uncertainty suppression (no explicit unknowns or hedges before committing to an action). We introduce narration-of-thought (NoT), a system prompt that structures chain-of-thought into five sections: protagonist, stakeholders, two-step consequences, uncertainty, then commitment. NoT adds no training, parameters, or fine-tuning. On 100 DailyDilemmas scenarios across four generators from three vendors, NoT cuts stakeholder collapse from up to 31% to under 1% and uncertainty suppression from up to 72% to 1-24% on every model. A matched-budget verbose-CoT control rules out token spend as the active ingredient; NoT retains Cliff's delta advantages of +0.79 to +0.90 on stakeholder count and +0.65 to +0.93 on uncertainty score for three of four generators, and a section ablation attributes each shift to its specific sub-instruction. Textual-gradient descent initialised at NoT improves the scaffold further; a cross-family training judge (different vendor from the generator) dominates an in-family one on every measured axis. Extended to a five-round multi-stakeholder debate protocol, the scaffold converts a 6% standoff into 95% full consensus on a calibration set and 100% combined convergence on a DailyDilemmas replication. The resulting traces externalise the stakeholders, consequences, and uncertainty grounding each commitment, providing an auditable substrate for dependable agentic deployment.

Problem

• stakeholder collapse
• uncertainty suppression
• ethical reasoning
• moral dilemmas
• large language models

Innovation

• Narration-of-Thought
• defeasible ethical reasoning
• inference-time scaffolding
• stakeholder representation
• uncertainty modeling