From 0-to-1 to 1-to-N: Reproducible Engineering Evidence for MetaAI Recursive Self-Design
2026-06-08 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors describe recursive self-design as a process where AI systems help modify their own design and improvement methods. They propose a framework with four key parts to identify when this is happening: a clear system to modify, a mechanism to make changes, a way to choose improvements based on feedback, and a process that continues recursively. They compare several public AI systems against these parts and highlight Darwin Goedel Machine, which shows measurable improvement over many iterations. Finally, the authors introduce MetaAI-Mini, a reproducible protocol for studying these ideas, but note it does not include completed experiment results yet.
recursive self-designMetaAIDarwin Goedel Machinemeta-level modifierfeedback-directed selectionopen-ended explorationself-improvementSWE-bench VerifiedPolyglotHumanEval
Authors
Dun Li, Jiatao Li, Hongzhi Li
Abstract
Recursive self-design refers to AI-assisted modification of the mechanisms by which an AI system is built, evaluated, and improved. This paper treats MetaAI not as a mature paradigm, but as a working term for a human-seeded, AI-expanded development pattern in which the design space itself becomes a target of modification. We propose an operational evidence framework with four criteria: inspectable target system, meta-level modifier, feedback-directed selection, and recursive continuation. We then map public systems, including Darwin Goedel Machine (DGM), STOP, Goedel Agent, and ShinkaEvolve, against these criteria. DGM provides the most direct currently reported evidence: its published results show improvement from 20% to 50% on SWE-bench Verified and from 14.2% to 30.7% on full Polyglot after 80 iterations, with ablations suggesting that both open-ended exploration and self-improvement contribute. Finally, we provide MetaAI-Mini, a reproducible HumanEval-based protocol and codebase. Because no completed model run is included in this build, MetaAI-Mini is reported as a protocol rather than as an experimental result.