How Language Models Organize and Structure Moral Knowledge
cs.CL, cs.AI, cs.LG
Submitted: 2026-08-27
Updated: 2026-10-07
Comments: 32 pages, 16 figures. Code and outputs at https://github.com/deepsteer/deepsteer
Code: https://github.com/deepsteer/deepsteer
License: http://creativecommons.org/licenses/by/4.0/
The gist: How do large language models (LLMs) organize moral knowledge? Models detect moral content broadly, but detection is a low bar.
Terminology
Abstract
How do large language models (LLMs) organize moral knowledge? Models detect moral content broadly, but detection is a low bar. We ask whether they go further, distinguishing moral foundations from one another and organizing the relationships between them geometrically. We train six independent linear probes on open-weight language models, one per Moral Foundations Theory (MFT) category (care/harm, fair/cheat, lib/oppress, loy/betray, auth/subv, sanc/degrade), and examine how the resulting directions relate to each other in representation space. We find the directions neither collapse into a single moral detector nor isolate from one another. Rather, they span a near-maximal number of independent dimensions while sharing a positive common component. The shared component is the signature of integration, and it is moral-specific relative to a matched non-moral concept battery built identically (mean pairwise cosine 0.26 vs. 0.013). The geometry is consistent across architectures and scale and reaches its integration regime early in pre-training, well before probe accuracy saturates. The structure the model discovers shows no evidence of the individualizing/binding distinction predicted by Moral Foundations Theory (an underpowered test: only 10 distinct splits exist, so it cannot reject at the 0.05 level) but rather reflects corpus statistics. Extending to moral dilemmas, each dilemma direction partially composes from its component foundations, at 2.7x a mismatched-pair baseline, while the majority of its variance encodes conflict-specific structure. The model represents moral tension itself, not a pre-resolved judgment.
Sources
- Understanding intermediate layers using linear classifier probes
- Refusal in Language Models Is Mediated by a Single Direction
- Aligning AI With Shared Human Values
- OLMoE: Open Mixture-of-Experts Language Models
- 2 OLMo 2 Furious
- The Linear Representation Hypothesis and the Geometry of Large Language Models
- Output Dilution: Redundant but Fragile Representations in MoE Models
- When Probing Accuracy Saturates, Fragility Resolves: A Complementary Metric for LLM Pre-Training Analysis
- Representation Engineering: A Top-Down Approach to AI Transparency
Related papers
- Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
- Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
- Subliminal Steering: Stronger Encoding of Hidden Signals
- MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
- The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
- Untangling the Mechanisms of Misleading Context in Medical Question Answering