Training Compute Thresholds: Features and Functions in AI Governance

Key Points

Compute thresholds serve as a practical trigger for evaluating AI models, shaping governance actions.
Key advantages include quantifiability and correlation with AI model risks and capabilities, highlighting their importance.
Implementation involves using compute thresholds alongside sector-specific regulations to improve overall AI governance frameworks and risk assessment processes. They act as initial filters for identifying high-risk models and guide regulatory decisions, indicating a notable innovation in AI management. They help ensure models are evaluated before deployment, potentially increasing safety.

Abstract

This paper examines the use of training compute thresholds as a tool for governing artificial intelligence (AI) systems. We argue that compute thresholds serve as a valuable trigger for further evaluation of AI models, rather than being the sole determinant of the regulation. Key advantages of compute thresholds include their correlation with model capabilities and risks, quantifiability, ease of measurement, robustness to circumvention, knowability before model development and deployment, potential for external verification, and targeted scope. Compute thresholds provide a practical starting point for identifying potentially high-risk models and can be used as an initial filter in AI governance frameworks alongside other sector-specific regulations and broader governance measures.

Read Full Paperexternally

Mark Helpful

Bookmark

Relay

View Full Paper