Recursive self-improvement (RSI)

The idea that an AI system can take over part of the work of building the next, better AI, instead of only helping a human researcher. The research loop is propose an idea, build it, run the experiment, check the result, learn from it, pick what to try next. RSI is when machines start running stages of that loop and the output of one generation helps create a stronger next one, with less human involvement each time. It is worth being precise, because the phrase "AI that builds AI" oversells where things actually are. Today RSI mostly means automated engineering and automated research on things that are easy to measure, like making model training faster or cheaper, not AI inventing a brand new model on its own. It is also different from a self-improving agent, which improves its own prompts, tools, and memory but does not rebuild the model. For people accountable for AI in healthcare the relevant shift is about oversight: as more of the loop moves from human hands to machine hands, the human job moves toward setting goals, validating results, and governing the process, which is exactly where the hard questions sit.

Terms like this come up in real clinical scenarios across the HelloAI courses: bite-sized modules with verifiable certificates. An account takes one minute, no password needed.