AI Improving AI Isn't Always RSI: Five Levels of Autonomy Explain What Makes the Loop Truly Recursive
TheTuringPost · x · 2026-09-18
Turing Post argues that not every case of AI improving AI counts as true Recursive Self-Improvement: if a model edits code and humans test the result, the loop repeats but stays the same.
Recursion emerges as more parts of the loop become editable by AI:
- Improving models/algorithms: editing code, kernels, or training methods (AI4AI-Bench gives agents hours to modify real training algorithms, then retrains from scratch)
- Improving search: choosing strategies, experiments, and compute allocation
- Creating useful experience: deciding what to learn and feeding it into the next round
- Adapting the research environment: changing tools and workflows
- Improving the improvement process itself: altering how search, experimentation, and evaluation work
These levels come from the paper "The Last AI Built by Humans," while "Recursive Criticality" adds that even fast, highly automated AI development isn't RSI if parts of the loop remain fixed. The point: recursion lies in the loop itself becoming something AI can change.
More from AGI Musings
- Gary Marcus: fixation on AI extinction scenarios distracts from practical misuse defenses — GaryMarcus · 2026-09-18
- Matt Parlmer calls Yudkowsky's evidence-proof AI doom-mongering a rationalist cautionary tale — repligate · 2026-09-18
- Most rationalists have rightly slashed P(doom by 2030), but some haven't — repligate · 2026-09-18
- Commenter argues OpenAI and Anthropic are slowing everyone down to keep their Coke-and-Pepsi duopoly — haider1 · 2026-09-18
- MIT Engineering Dean Huttenlocher: AI accountability stays with people, not systems — AlexTensor · 2026-09-18
- Anthropic publishes three metrics tracking AI-driven R&D as Stanford lab builds the audit dashboard — erikbryn · 2026-09-18