What is recursive self-improvement?

Recursive self-improvement is a theoretical process by which an AI agent could enter a feedback loop that causes it to improve itself very quickly. For example, an AI capable of human-level AI research might make adjustments to its own software or hardware that improves its performance, enabling it to then make more adjustments that lead to even further improved performance. This would create a cycle of increasingly advanced AI capabilities and even a slow takeoff could result in AI agents that are far more capable than humans.

While humans can also improve our capabilities by studying new topics, practicing new skills, etc., we don't have much ability to alter more fundamental facts about ourselves, like our biology.¹ AI agents, on the other hand, might be able to recursively self-improve much faster than humans can because they can be copied, directly edited, and run on new hardware.

Not everybody agrees that recursive self-improvement by AIs is possible. Few AI systems to date have been designed to pursue recursive self-improvement as an explicit goal. However, if recursive self-improvement is possible, it seems plausible that this would be an instrumentally convergent strategy for most AI agents that are capable of it.