22 AI Experts Warn: Years of Technological Progress Could Be Compressed Into Months

Twenty-two AI experts have warned of an 'intelligence explosion.' They believe that automating AI development could compress years of technological progress into mere months, urging a policy response. The researchers' diagnosis is that such compression could become a reality if AI-driven automation of AI development takes hold.

The paper, titled 'What If Automation of AI Research and Development Triggers an Intelligence Explosion?' was published through the University of Cambridge's AI Science and Policy Programme (CASP). It bears the names of Geoffrey Hinton, professor emeritus at the University of Toronto, and Yoshua Bengio, professor at the University of Montreal. Active industry researchers also joined, including OpenAI chief scientist Jakub Pachocki, Anthropic co-founder Jack Clark, Microsoft Chief Scientific Officer Eric Horvitz, and Song Dawn, Vice President of AI Research at Meta.

The researchers noted that the pace of automation—delegating AI model development tasks to AI itself—is rapidly accelerating within major tech companies. They cited Anthropic as evidence. The share of developer-approved code written by AI surged from single digits in January last year to more than 80 percent by May this year. The proportion of research and development (R&D) tasks completed autonomously by AI, with humans overseeing only, also rose from 1 percent in March to 26 percent last month.

The researchers predicted that if this trend continues, entire months-long complex AI R&D projects could be fully automated within two years. Furthermore, they concluded that the moment the loop of AI's recursive self-improvement (RSI) is completed, explosive technological acceleration could begin. They explained that an intelligence explosion could dramatically accelerate technological innovation and have positive effects, such as in drug discovery.