AI Glossary: RSI is the hot acronym of the hour - Recursive Self-Improvement

Image generated by ChatGPT 4Nov2026

Simply put, Recursive Self-Improvement (RSI) is when AI learns over time and improves itself. For genealogists, RSI could mean models could improve tasks genealogists already use AI for: transcription, abstracting, and drafting sourced narratives. But genealogists still need to proof the work as the risk of hallucination is still possible. Remember to use the Coalition of Responsible AI in Genealogy's Accuracy principle: "AI can generate false, biased, or incorrect content. Therefore, members of the genealogical community verify the accuracy of the information with other records and acknowledge credible sources of content generated by AI." The other effect that RSI will have is on output. If RSI improves and increases output,workflows may need adjustment.

There is current debate as to whether RSI has occurred already, so I posed this question to ChatGPT and it replied that it has occurred:

"My assessment: AI RSI has already occurred in a limited, technically meaningful sense. The evidence is strongest for systems that recursively improve their software, research methods, and agent performance. What remains unproven is the much more consequential development: an AI that can independently and repeatedly produce substantially more intelligent successors. That is the distinction between self-improving AI, which already exists, and an autonomous intelligence explosion, which remains hypothetical."

Published in September 2026, researchers developed AIDE², an AI research agent capable of modifying its own code, testing the changes, and retaining those that improved its performance. During an eight-day experiment conducted without human intervention, AIDE² made seven successive improvements to its research capabilities, including more effective search methods and memory management. The researchers reported that these improvements also enhanced the system’s performance on previously unseen evaluation tasks. It will be interesting to see the advancements that come from RSI agents and when they may be made available to the public.