SGM: A Statistical Godel Machine for Risk-Controlled Recursive Self-Modification
This addresses the safety issue for AutoML and adaptive optimization systems by providing a risk-controlled framework, though it is incremental as it builds on Godel machines with statistical adaptations.
The paper tackles the problem of ensuring safe recursive self-modification in machine learning systems by introducing the Statistical Godel Machine (SGM), which uses statistical confidence tests to certify improvements before allowing modifications, and demonstrates its effectiveness in experiments on datasets like CIFAR-100 and ImageNet-100.
Recursive self-modification is increasingly central in AutoML, neural architecture search, and adaptive optimization, yet no existing framework ensures that such changes are made safely. Godel machines offer a principled safeguard by requiring formal proofs of improvement before rewriting code; however, such proofs are unattainable in stochastic, high-dimensional settings. We introduce the Statistical Godel Machine (SGM), the first statistical safety layer for recursive edits. SGM replaces proof-based requirements with statistical confidence tests (e-values, Hoeffding bounds), admitting a modification only when superiority is certified at a chosen confidence level, while allocating a global error budget to bound cumulative risk across rounds.We also propose Confirm-Triggered Harmonic Spending (CTHS), which indexes spending by confirmation events rather than rounds, concentrating the error budget on promising edits while preserving familywise validity.Experiments across supervised learning, reinforcement learning, and black-box optimization validate this role: SGM certifies genuine gains on CIFAR-100, rejects spurious improvement on ImageNet-100, and demonstrates robustness on RL and optimization benchmarks.Together, these results position SGM as foundational infrastructure for continual, risk-aware self-modification in learning systems.Code is available at: https://github.com/gravitywavelet/sgm-anon.