Two-year update on my personal AI timelines

Abstract

This article updates a previous analysis of the timeline for transformative AI, arguing that the arrival of such AI is now likely to occur sooner than previously expected. The author’s analysis, based on bio-anchors for forecasting, now suggests a 15% probability of transformative AI by 2030, a 35% probability by 2036, and a 60% probability by 2050. The updates are mainly driven by a lower bar for transformative AI, the potential for short-horizon training of AI systems, and the surprisingly fast progress observed in the field of AI development. The article also discusses factors that push toward longer timelines, including potential regulations on AI development, the need for massive investments, and the difficulty of translating AI capabilities into economic impact. – AI-generated abstract