Understanding Moral Reasoning Trajectories in Large Language Models: Toward Probing-Based Explainability

2026-03-16

AI Summary

This work addresses the lack of interpretability in how large language models invoke ethical frameworks during morally sensitive decision-making. The authors introduce the concept of “moral reasoning trajectories” and propose the MRC (Moral Reasoning Consistency) metric to systematically analyze dynamic ethical framework switching across multi-step reasoning. Using linear probing for localization, lightweight activation steering, KL divergence evaluation, and human annotation validation, they find that 55.4–57.7% of reasoning steps involve framework shifts, with unstable trajectories exhibiting greater vulnerability to adversarial attacks. Probing-based interventions significantly reduce KL divergence, and MRC scores show strong correlation with human judgments (r = 0.715), revealing an intrinsic link among ethical representations, behavioral stability, and human moral assessment.

Technology Category

• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Accountability, Transparency, and Ethics for personalization
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Large language models (LLMs) increasingly participate in morally sensitive decision-making, yet how they organize ethical frameworks across reasoning steps remains underexplored. We introduce \textit{moral reasoning trajectories}, sequences of ethical framework invocations across intermediate reasoning steps, and analyze their dynamics across six models and three benchmarks. We find that moral reasoning involves systematic multi-framework deliberation: 55.4--57.7\% of consecutive steps involve framework switches, and only 16.4--17.8\% of trajectories remain framework-consistent. Unstable trajectories remain 1.29$\times$ more susceptible to persuasive attacks ($p=0.015$). At the representation level, linear probes localize framework-specific encoding to model-specific layers (layer 63/81 for Llama-3.3-70B; layer 17/81 for Qwen2.5-72B), achieving 13.8--22.6\% lower KL divergence than the training-set prior baseline. Lightweight activation steering modulates framework integration patterns (6.7--8.9\% drift reduction) and amplifies the stability--accuracy relationship. We further propose a Moral Representation Consistency (MRC) metric that correlates strongly ($r=0.715$, $p<0.0001$) with LLM coherence ratings, whose underlying framework attributions are validated by human annotators (mean cosine similarity $= 0.859$).

Problem

• moral reasoning
• large language models
• ethical frameworks
• reasoning trajectories
• explainability

Innovation

• moral reasoning trajectories
• linear probing
• activation steering
• ethical framework
• Moral Representation Consistency