Modular AGI: From Reactive Cores to Safe Autonomy — A Survey and Roadmap

This survey presents a mathematically grounded modular architecture for Artificial General Intelligence (AGI), designed to overcome the limitations of monolithic systems. AGI is formalized as a set of independent cognitive cores, each with proven mathematical necessity, ensuring safety through intrinsic design rather than external constraints. We address critical bottlenecks integration core overload, memory saturation, communication limits, irreversibility risks, verification complexity, and emergent behavior uncertainty with robust, practical engineering solutions. Enhancements include distributed consensus, federated memory, secure update mechanisms, and adaptive anomaly detection. The result is a scalable, fault-tolerant, and safe AGI framework that balances mathematical rigor with implementable solutions. Related document: Ethics as Optimal Strategy: Mathematical Analysis of Moral Choice in AI (DOI: 10.5281/zenodo.15858394)