Embedding alignment improves ethical reasoning in large language models
IREA: Intermediate Representation-based Embedding Alignment for Normative RAG
Computation and Language
Summary
Large language models struggle to answer questions involving ethics because ethical rules can vary widely. The authors offer a method that helps these models make better ethical judgments by allowing them to look up ethical knowledge from outside sources. Their method, called IREA, changes how the computer matches detailed story-like questions with general ethical statements by using an intermediate representation that captures key ethical details. This approach helps the system understand ethical questions better and produce fairer, more accurate answers.
What this means in practice
- •For ai ethics teams: Improve AI systems’ ability to reason about ethical issues by integrating aligned normative knowledge retrieval.
- •For legal technology developers: Enhance legal document analysis tools by aligning context-rich queries with normative rules for ethical judgment support.
Authors
Mirae Han, Sihyeong Yeom, Harksoo Kim
Abstract
Large language models (LLMs) have shown strong performance across various tasks, but they still struggle with questions involving ethical judgment. Previous studies have attempted to train LLMs on ethical standards, but the diversity and relativity of ethical norms make them difficult to fully internalize in model parameters. As an alternative, we introduce normative RAG, a retrieval-augmented approach that supports ethical judgment using external normative knowledge. Normative retrieval involves a distinct asymmetry between context rich narrative queries and generalized normative statements. Existing factual retrieval methods rely on query-only expansion into a document-like form, making them insufficient for resolving this asymmetry. Therefore, we propose Intermediate Representation-based Embedding Alignment (IREA), a bidirectional alignment method that maps both text types into a shared situation-behavior representation. This representation captures ethically salient contextual and behavioral information in a normalized form, reducing surface-level discrepancies and improving alignment in the embedding space. Experimental results show that IREA improves normative retrieval and downstream ethical judgment across multiple settings, demonstrating the effectiveness of bidirectional alignment for normative RAG.