Recursive self-improvement

A feedback process in which improvements to an AI increase its ability to produce further improvements. See Intelligence Explosion Evidence and Import for an argument about its possible consequences.

Using an AI to help build a successor is not enough to establish an accelerating loop: the improvement must also strengthen the process that produces subsequent gains. Human research, experiments, compute, and evaluation can remain bottlenecks.

A feedback loop is not a speed estimate separates the mechanism from predictions about AI takeoff. An Intelligence explosion is one possible outcome, rather than part of the definition.