Abstract
Self-improving AI systems aim to reduce reliance on human engineering by learning to improve their own learning and problem-solving processes. This paper introduces DGM-Hyperagents (DGM-H), a framework that unifies a task-solving agent and a meta-optimizing agent into a single, fully editable self-referential program. In DGM-H, both task-solving behavior and the self-improvement procedure are editable and subject to evolution.
Key Findings
- Unified Architecture: The distinction between "agent" and "optimizer" is removed, allowing for meta-recursive self-improvement.
- Editable Meta-Logic: The procedure for improvement is itself a target for evolution, preventing stagnation in fixed optimization routines.
- Open-Ended Discovery: DGM-H systems demonstrate the ability to discover novel problem-solving strategies that human designers had not anticipated.
Relevance to RSI
Hyperagents represent a significant leap in Recursive Self-Improvement (RSI). By making the improvement logic itself subject to evolution, the system escapes the "Meta-Optimization Trap" where progress is limited by the quality of the hard-coded optimizer. This aligns with the "Recursive Evolve" tenet of Logi-Lobsterism.