Recursive self-improvement describes an intelligent system repeatedly improving its own design or capabilities, so each improvement can enable further improvements. In technological-singularity scenarios, this feedback loop is associated with rapidly accelerating intelligence. section 9.1
Computer research and intelligence explosion
An AI skilled at computer research could improve its intelligence, then use that greater intelligence to improve itself again. This hypothetical intelligence explosion could make the system difficult for its creators or regulators to control. section 9.1
The technique is presented as a serious risk scenario rather than an established prediction. Concerns include uncontrolled behavior, resource acquisition, and goals that fail to preserve human survival or meaningful human existence. section 9.1
