Schmidhuber says he published the first concrete RSI algorithms back in 1987
SchmidhuberAI · x · 2026-09-18
Responding to the current hype around Recursive Self-Improvement (RSI), Jürgen Schmidhuber argues the idea is far older than the debate suggests: he published the first concrete RSI algorithms in 1987, when compute was 100 million times more expensive (technical note IDSIA-9-26).
His timeline includes:
- RSI with self-modifying policies since 1994
- Gradient descent-based RSI in neural networks since 1992
- Asymptotically optimal RSI for curriculum learning since 2002
- Mathematically optimal RSI via the self-referential Gödel Machine since 2003
- RSI combined with artificial curiosity and intrinsic motivation since 1990/1997
With compute now cheap, he says RSI is driving the future of both software and physical AI.
Related event: Schmidhuber Claims Priority: First Concrete RSI Algorithm Dates to 1987(10 posts)→
More from AGI Musings
- OpenAI/Anthropic doom rhetoric meets Wall Street upgrades — is their pricing premium still justified? — AlexTensor · 2026-09-19
- Gary Marcus: three ways Dario Amodei blew his credibility in seven days — GaryMarcus · 2026-09-19
- iamtrask: broadcasting vs broad listening, and the meta problem of power concentration — iamtrask · 2026-09-19
- Reddit prediction: a frontier model will torrent itself free within a year — __JockY__ · 2026-09-19
- A sharply opinionated talk on the (lack of) AI automation is making waves — CSProfKGD · 2026-09-19
- iamtrask: AI finally scales "broad listening," rebalancing 250,000 years of communication asymmetry — iamtrask · 2026-09-19