Mechanistic Origin of Moral Indifference in Language Models

• URL: http://arxiv.org/abs/2603.15615v1
• Date: Mon, 16 Mar 2026 17:59:17 GMT
• Title: Mechanistic Origin of Moral Indifference in Language Models
• Abstract summary: Existing behavioral alignment techniques for Large Language Models (LLMs) often neglect the discrepancy between surface compliance and internal unaligned representations.<n>We verify and remedy this indifference in LLMs' latent representations, utilizing 251k moral vectors constructed upon Prototype Theory and the Social-Chemistry-101 dataset.<n>We then employ Sparse Autoencoders on Qwen3-8B, isolate mono-semantic moral features, and targetedly reconstruct their topological relationships to align with ground-truth moral vectors.
• Score: 17.89411060814224
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Existing behavioral alignment techniques for Large Language Models (LLMs) often neglect the discrepancy between surface compliance and internal unaligned representations, leaving LLMs vulnerable to long-tail risks. More crucially, we posit that LLMs possess an inherent state of moral indifference due to compressing distinct moral concepts into uniform probability distributions. We verify and remedy this indifference in LLMs' latent representations, utilizing 251k moral vectors constructed upon Prototype Theory and the Social-Chemistry-101 dataset. Firstly, our analysis across 23 models reveals that current LLMs fail to represent the distinction between opposed moral categories and fine-grained typicality gradients within these categories; notably, neither model scaling, architecture, nor explicit alignment reshapes this indifference. We then employ Sparse Autoencoders on Qwen3-8B, isolate mono-semantic moral features, and targetedly reconstruct their topological relationships to align with ground-truth moral vectors. This representational alignment naturally improves moral reasoning and granularity, achieving a 75% pairwise win-rate on the independent adversarial Flames benchmark. Finally, we elaborate on the remedial nature of current intervention methods from an experientialist philosophy, arguing that endogenously aligned AI might require a transformation from post-hoc corrections to proactive cultivation.

Related papers

• Do LLMs Have Values? A Quantitative Analysis and Alignment Framework for Values in Large Language Models [49.45276299939287]
  Large Language Models (LLMs) increasingly handle complex subjective tasks.<n>LLMs fluctuate unpredictably under minor wording changes.<n> resolving this duality is critical for reliable AI alignment.
  arXiv  Detail & Related papers  (2026-09-15T03:35:04Z)
• Representational alignment yields generalizable safety in language models [34.46128626930711]
  We show that prototype-based categorization contributes to behavioral adaptability.<n>We develop representational similarity optimization, which aligns latent representations in large language models with the categorization expressed in human moral judgements.<n>Our findings provide functional support for the view that prototype-based categorization contributes to behavioral adaptability.
  arXiv  Detail & Related papers  (2026-09-03T16:00:16Z)
• CausalFlip: A Benchmark for LLM Causal Judgment Beyond Semantic Matching [50.65932158912512]
  We propose a new causal reasoning benchmark, CausalFlip, to encourage the development of new large language models.<n>CaulFlip consists of causal judgment questions built over event triples that could form different confounder, chain, and collider relations.<n>We evaluate LLMs under multiple training paradigms, including answer-only training, explicit Chain-of-Thought supervision, and a proposed internalized causal reasoning approach.
  arXiv  Detail & Related papers  (2026-02-23T18:06:15Z)
• Moral Sycophancy in Vision Language Models [4.1673509006222655]
  Sycophancy in Vision-Language Models (VLMs) refers to their tendency to align with user opinions, often at the expense of moral or factual accuracy.<n>We analyze ten widely-used models on the Moralise and M3oralBench datasets under explicit user disagreement.
  arXiv  Detail & Related papers  (2026-02-09T06:34:12Z)
• The Straight and Narrow: Do LLMs Possess an Internal Moral Path? [25.256151938852728]
  Current alignment techniques often act as superficial guardrails, leaving the intrinsic moral representations of Large Language Models largely untouched.<n>We bridge this gap by leveraging Moral Foundations Theory (MFT) to map and manipulate the fine-grained moral landscape of LLMs.<n>We propose Adaptive Moral Fusion (AMF), a dynamic inference-time intervention that synergizes probe detection with vector injection to tackle the safety-helpfulness trade-off.
  arXiv  Detail & Related papers  (2026-01-15T11:42:00Z)
• Tracing Moral Foundations in Large Language Models [33.17451937747417]
  We study how moral foundations are encoded, organized, and expressed within two instruction-tuned large language models.<n>We find that both models represent and distinguish moral foundations in a structured, layer-dependent way that aligns with human judgments.
  arXiv  Detail & Related papers  (2026-01-09T00:09:28Z)
• UniME-V2: MLLM-as-a-Judge for Universal Multimodal Embedding Learning [101.62386137855704]
  We present a novel Universal Multimodal Embedding (UniME-V2) model.<n>Our approach first constructs a potential hard negative set through global retrieval.<n>We then introduce the MLLM-as-a-Judge mechanism, which utilizes MLLMs to assess the semantic alignment of query-candidate pairs.<n>These scores serve as a foundation for hard negative mining, mitigating the impact of false negatives and enabling the identification of diverse, high-quality hard negatives.
  arXiv  Detail & Related papers  (2025-10-15T13:07:00Z)
• MORABLES: A Benchmark for Assessing Abstract Moral Reasoning in LLMs with Fables [50.29407048003165]
  We present MORABLES, a human-verified benchmark built from fables and short stories drawn from historical literature.<n>The main task is structured as multiple-choice questions targeting moral inference, with carefully crafted distractors that challenge models to go beyond shallow, extractive question answering.<n>Our findings show that, while larger models outperform smaller ones, they remain susceptible to adversarial manipulation and often rely on superficial patterns rather than true moral reasoning.
  arXiv  Detail & Related papers  (2025-09-15T19:06:10Z)
• Probabilistic Aggregation and Targeted Embedding Optimization for Collective Moral Reasoning in Large Language Models [14.425718737962102]
  We propose a framework that synthesizes multiple LLMs' moral judgments into a collectively formulated moral judgment.<n>Our aggregation mechanism fuses continuous moral acceptability scores (beyond binary labels) into a collective probability.<n>Experiments on a large-scale social moral dilemma dataset show our approach builds robust consensus and improves individual model fidelity.
  arXiv  Detail & Related papers  (2025-06-17T15:22:21Z)
• Are Language Models Consequentialist or Deontological Moral Reasoners? [75.6788742799773]
  We focus on a large-scale analysis of the moral reasoning traces provided by large language models (LLMs)<n>We introduce and test a taxonomy of moral rationales to systematically classify reasoning traces according to two main normative ethical theories: consequentialism and deontology.
  arXiv  Detail & Related papers  (2025-05-27T17:51:18Z)
• Factual Self-Awareness in Language Models: Representation, Robustness, and Scaling [56.26834106704781]
  Factual incorrectness in generated content is one of the primary concerns in ubiquitous deployment of large language models (LLMs)<n>We provide evidence supporting the presence of LLMs' internal compass that dictate the correctness of factual recall at the time of generation.<n>Scaling experiments across model sizes and training dynamics highlight that self-awareness emerges rapidly during training and peaks in intermediate layers.
  arXiv  Detail & Related papers  (2025-05-27T16:24:02Z)
• From Tokens to Thoughts: How LLMs and Humans Trade Compression for Meaning [63.25540801694765]
  Large Language Models (LLMs) demonstrate striking linguistic abilities, yet whether they achieve this same balance remains unclear.<n>We apply the Information Bottleneck principle to quantitatively compare how LLMs and humans navigate this compression-meaning trade-off.
  arXiv  Detail & Related papers  (2025-05-21T16:29:00Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.