This paper proposes a configurable architecture for training artificial general intelligence (AGI) and artificial superintelligence (ASI) through immersive simulation. Inspired by iconic platforms like SimCity, Second Life, and the evolving Metaverse, we introduce the Learning Stack: a modular, multi-layer system capable of hosting diverse and adaptable virtual environments that emulate the human learning process.
The Learning Stack enables AGI/ASI agents to undergo developmental learning experiences in configurable, context-rich worlds. Each layer of the architecture—from substrate to ethics—can be adjusted to simulate different sociotechnical, emotional, or cognitive scenarios. This makes the framework not only a theory of training but a tool for experimentation, benchmarking, and iterative refinement.
Rather than imposing static logic or datasets, this proposal shifts toward experiential learning—intelligence as a process of growing within, and adapting to, a lived synthetic reality. The Learning Stack offers a roadmap for cultivating judgment, values, memory continuity, and ethical reasoning in AI.