Three types of intelligence explosion

Abstract

Once AI systems can design and build more capable AI systems, we could see an intelligence explosion, where AI capabilities rapidly increase. Besides the classic scenario of AI improving AI software, AI could also improve other inputs to AI development. This paper analyzes three feedback loops: software, chip technology, and chip production. These could drive three types of intelligence explosion: a software intelligence explosion driven by software improvements alone; an AI-technology intelligence explosion driven by software and chip technology improvements; and a full-stack intelligence explosion driven by all three feedback loops. Even if a software intelligence explosion never happens or quickly plateaus, AI-technology and full-stack explosions remain possible. While these might start more gradually, they could accelerate to very fast rates of development. Each feedback loop could potentially increase effective compute by 20-30 orders of magnitude before reaching physical limits, enabling dramatic improvements in AI capabilities. The type of intelligence explosion also has implications for power distribution: a software explosion would likely concentrate power within one country or company, while a full-stack explosion would be more spread out across many countries and industries.