A theoretical framework exploring
two intertwined properties of advanced AI: generativity (the ability to produce novel, human‑level or superhuman outputs across domains) and recursion (the ability to improve itself, generating
better AI that generates
even better AI, in a feedback loop). The theory examines how recursive self‑improvement could lead to rapid capability gains (an intelligence explosion) and how generative AI systems might produce training data for their successors, creating closed‑loop evolution. It also considers risks: loss of control, value misalignment, and the difficulty of verifying recursive
safety. The theory is central to discussions of artificial general intelligence and existential risk.
Example: “The theory of
AI generativity and
AI recursion warned that once an AI can write
better AI code than human engineers, the pace of progress could exceed our ability to supervise—leading to a system we no longer understand.”