Recursive self-improvement algorithms reportedly date back to 1987

A post traces concrete algorithms for recursive self-improvement (RSI)—AI improving its own methods—back to 1987. It lists neural RSI using gradient descent from 1992, reinforcement learning with self-modifying policies from 1994, and artificial curiosity and intrinsic motivation in 1997. The timeline also includes curriculum-learning approaches from 2002 and describes the Gödel Machine from 2003 as a route to “mathematically optimal” RSI. It links to newer work, including an RSI survey.