AI intelligence explosion: growth may be faster than exponential, but not infinite

The paper by Toby Ord addresses the dynamics of intelligence explosions, where AI is used to improve AI R&D. This feedback loop could, under certain conditions, lead to rapidly escalating capabilities. Ord's mathematical exploration reveals that singular growth—toward a vertical asymptote—is more difficult to achieve than recent economics-inspired models suggest. Instead, there is an important class of growth rates that are faster than exponential but do not result in a vertical asymptote.

The generation time, or the time to complete one feedback loop, emerges as a critical factor. The analysis shows that without generation time rapidly approaching zero, singular growth cannot occur. This insight challenges the assumption that an intelligence explosion is inevitable if AI capabilities improve, highlighting a potentially overlooked parameter in forecasting AI's future.

For business readers, this research provides a more nuanced view of AI growth scenarios. While the idea of an intelligence explosion is often discussed, Ord's work suggests that the most explosive possibilities may be less likely, but there are still growth rates faster than exponential that could have significant implications. Understanding these dynamics could help in assessing long-term risks and opportunities in AI development.