UN Forms AI Panel Amid Milestones
The United Nations has established a new scientific advisory panel to provide evidence-based guidance on the global impacts of artificial intelligence, drawing comparisons to the IPCC for climate change. The move comes as scholars gather at Dartmouth to mark the 70th anniversary of the term "AI" and chart its future. Meanwhile, the 2025 Stanford Index reveals AI systems are making rapid progress on demanding reasoning and problem-solving benchmarks.
The UN's 40-member "Independent International Scientific Panel on Artificial Intelligence" was formed after a resolution in the General Assembly and is designed to provide authoritative, science-based analysis on AI's rapid evolution. The panel's members, who will serve three-year terms, were selected from a pool of over 2,600 applicants and include experts from academia, civil society, and the private sector across all five UN regions. The advisory body has drawn comparisons to the Intergovernmental Panel on Climate Change (IPCC), intended to synthesize complex research to inform global policymaking and serve as an "early-warning system." However, it faces the challenge of adapting the IPCC's slow, consensus-driven model, which can take 5-7 years between assessments, to the much faster pace of AI development. The creation of the panel was not unanimous. The United States voted against the resolution, arguing that AI governance should not be under UN jurisdiction and raising concerns about the transparency of the panel's formation process. Despite this, 117 countries voted in favor, with many U.S. allies in Europe supporting the move. The Dartmouth conference marks the 70th anniversary of the 1956 "Dartmouth Summer Research Project on Artificial Intelligence," widely considered the event that founded AI as a distinct field of study. It was organized by pioneers like John McCarthy, who coined the term "artificial intelligence," and Marvin Minsky, with the goal of exploring how machines could be made to simulate learning and other features of human intelligence. This year's anniversary events include "The Dartmouth Conference, Revisited," which aims to clarify where AI can assist human creativity and where ethical guardrails are most needed. The year-long commemoration also includes forums on the future of work and entrepreneurship in the age of AI. The 2025 Stanford AI Index provides concrete evidence of AI's accelerating capabilities on difficult new benchmarks designed to test the limits of machine reasoning. On the SWE-bench, a test composed of real-world software engineering problems, AI performance jumped from solving 4.4% of issues in 2023 to 71.7% in 2024. Similarly, on a benchmark of graduate-level questions designed to be difficult to answer via web search (GPQA), AI model scores improved by nearly 49 percentage points within the year. While these models excel at specific tasks, the report notes that complex, multi-step reasoning remains a significant challenge, hindering reliability in high-stakes applications. While
