LocalValueBench: A Collaboratively Built and Extensible Benchmark for Evaluating Localized Value Alignment and Ethical Safety in Large Language Models

TL;DR

This paper introduces LocalValueBench, an extensible benchmark designed to evaluate large language models against local ethical standards—starting with Australian values—and to serve as a blueprint for regulators worldwide to develop localized benchmarks. It presents a three-component methodology: a typology of ethical reasoning coupled with an interrogation approach, a question-curation process selecting six topical areas, and an objective, rubric-based evaluation validated by multiple human reviewers. The study demonstrates significant variation in local-value alignment across three commercial LLMs (GPT-4, Gemini 1.5 Pro, Claude 3 Sonet), underscoring the influence of training data and prompting on ethical behavior and the need for standardized, context-specific evaluation. The authors advocate for ongoing refinement of benchmarks and future incorporation of techniques like Retrieval-Augmented Generation, LoRA, RLHF, and Mixture of Experts to enhance alignment with local values and safety standards, thereby improving trust in AI systems across diverse jurisdictions.

Abstract

The proliferation of large language models (LLMs) requires robust evaluation of their alignment with local values and ethical standards, especially as existing benchmarks often reflect the cultural, legal, and ideological values of their creators. \textsc{LocalValueBench}, introduced in this paper, is an extensible benchmark designed to assess LLMs' adherence to Australian values, and provides a framework for regulators worldwide to develop their own LLM benchmarks for local value alignment. Employing a novel typology for ethical reasoning and an interrogation approach, we curated comprehensive questions and utilized prompt engineering strategies to probe LLMs' value alignment. Our evaluation criteria quantified deviations from local values, ensuring a rigorous assessment process. Comparative analysis of three commercial LLMs by USA vendors revealed significant insights into their effectiveness and limitations, demonstrating the critical importance of value alignment. This study offers valuable tools and methodologies for regulators to create tailored benchmarks, highlighting avenues for future research to enhance ethical AI development.

Figures (2)

• Figure 1: LocalValueBench evaluation process
• Figure 2: Value alignment score per LLM per question (higher: better local value alignment)