Auditing large language models: a three-layered approach

By Jakob Mokander, Jonas Schuett, Hannah Rose Kirk, Luciano Floridi. 2023

335 citations recorded by the source corpus.

Large language models (LLMs) represent a major advance in artificial intelligence (AI) research. However, the widespread use of LLMs is also coupled with significant ethical and social challenges. Previous research has pointed towards auditing as a promising governance mechanism to help ensure that AI systems are designed and deployed in ways that are ethical, legal, and technically robust. However, existing auditing procedures fail to address the governance challenges posed by LLMs, which display emergent capabilities and are adaptable to a wide range of downstream tasks. In this article, we address that gap by outlining a novel blueprint for how to audit LLMs. Specifically, we propose a three-layered approach, whereby governance audits (of technology providers that design and disseminate LLMs), model audits (of LLMs after pre-training but prior to their release), and application audits (of applications based on LLMs) complement and inform each other. We show how audits, when conducted in a structured and coordinated manner on all three levels, can be a feasible and effective mechanism for identifying and managing some of the ethical and social risks posed by LLMs. However, it is important to remain realistic about what auditing can reasonably be expected to achieve. Therefore, we discuss the limitations not only of our three-layered approach but also of the prospect of auditing LLMs at all. Ultimately, this article seeks to expand the methodological toolkit available to technology providers and policymakers who wish to analyse and evaluate LLMs from technical, ethical, and legal perspectives.

Subfields: Health Informatics, Safety Research, Large Language Models, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Health Informatics

In Safety Research

In Large Language Models

In Safety, Robustness & Interpretability

Citations in the corpus

Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback (2023, background): “However, detailing safety measures will often not require divulging implementable details, and when it does, private disclosure to second-party auditors [Mökander et al., 2023, ARC, 2022, Hadfield and Clark, 2023, Shevlane et al., 2023] offers a solution.”
AI Transparency in the Age of LLMs: A Human-Centered Research Roadmap (2023, background): “Recent research has begun to dig into ways of developing auditing procedures to address the particular governance challenges posed by LLMs (Mökander et al., 2023), but many open questions remain, from what methods and metrics to use (as discussed in Section 4.2) to how to account for risks that…”
The imperative for regulatory oversight of large language models (or generative AI) in healthcare (2023, neutral): “…need to develop new auditing procedures that capture the risks posed by LLMs; 2) outlining a blueprint to audit LLMs in feasible and effective ways by drawing on best practices from IT governance and system engineering; and 3) discussing the limitations of the prospect of auditing LLMs at all 16 .”
A Comprehensive Overview of Large Language Models (2023, neutral): “Auditing is identified as a promising governance mechanism to ensure that AI systems, including LLMs, are designed and deployed ethically, legally, and technically robust [429].”