[Paper Review] International AI Safety Report

Yoshua Bengio, Sören Mindermann+94 more|ArXiv.org|Jan 29, 2025

Ethics and Social Impacts of AI10 citations

TL;DR

The paper synthesizes the current evidence on capabilities, risks, and safety of advanced AI systems, assembled for the AI Safety Summit's Expert Advisory Panel. It reflects contributions from 100 AI experts across 30 nations and major international bodies.

ABSTRACT

The first International AI Safety Report comprehensively synthesizes the current evidence on the capabilities, risks, and safety of advanced AI systems. The report was mandated by the nations attending the AI Safety Summit in Bletchley, UK. Thirty nations, the UN, the OECD, and the EU each nominated a representative to the report's Expert Advisory Panel. A total of 100 AI experts contributed, representing diverse perspectives and disciplines. Led by the report's Chair, these independent experts collectively had full discretion over the report's content.

Motivation & Objective

• Assess the current state of evidence on the capabilities of advanced AI systems.
• Evaluate associated risks and safety considerations across diverse domains.
• Provide independent expert guidance for international policymakers and stakeholders.

Proposed method

• convened an Expert Advisory Panel with representatives from 30 nations, the UN, OECD, and the EU.
• 100 AI experts contributed, representing diverse disciplines and perspectives.
• Independent experts had full discretion over the report's content.

Experimental results

Research questions

• RQ1 What are the current capabilities of advanced AI systems as evidenced by expert input?
• RQ2 What are the primary risks and safety concerns associated with advanced AI systems?
• RQ3 What policy and safety recommendations arise from an international, multidisciplinary assessment?

Key findings

• The report synthesizes evidence on capabilities, risks, and safety of advanced AI systems.
• It was mandated by nations attending the AI Safety Summit in Bletchley, UK.
• Contributors represented diverse perspectives and disciplines, totaling 100 AI experts.