Safety Cases: How to Justify the Safety of Advanced AI Systems

TL;DR

This paper explores how to construct structured safety cases for advanced AI systems, proposing a framework and argument categories to justify their safe deployment amidst increasing capabilities.

Contribution

It introduces a comprehensive framework for AI safety cases and categorizes key argument types to support safety justification for advanced AI systems.

Findings

Four categories of safety arguments identified

Concrete examples of safety arguments analyzed

Framework for combining arguments to justify safety

Abstract

As AI systems become more advanced, companies and regulators will make difficult decisions about whether it is safe to train and deploy them. To prepare for these decisions, we investigate how developers could make a 'safety case,' which is a structured rationale that AI systems are unlikely to cause a catastrophe. We propose a framework for organizing a safety case and discuss four categories of arguments to justify safety: total inability to cause a catastrophe, sufficiently strong control measures, trustworthiness despite capability to cause harm, and -- if AI systems become much more powerful -- deference to credible AI advisors. We evaluate concrete examples of arguments in each category and outline how arguments could be combined to justify that AI systems are safe to deploy.