LuminidsAI: A Shared Event-Substrate Architecture for Interpretable Developmental Artificial Life in Games

This working paper presents LuminidsAI, a developmental artificial life architecture for real-time games. The system uses a shared embodied event substrate to drive personality formation, awareness progression, memory, social inference, culture emergence, and player-facing explanation in Luminids, a cozy voxel fantasy game. Its central proposition is that artificial lifeforms should not merely act alive, but should remember, change, reveal why they changed, and become visibly more themselves over time. The paper introduces a shared event-substrate architecture in which lived events are classified once and then consumed by multiple developmental systems, including memory, personality, self-beliefs, theory-of-mind beliefs, intrinsic goals, colony culture, and explanation. A central contribution is the causal becoming ledger, an interpretability layer that records not only what changed in a Luminid, but why it changed. The work is positioned as a game-native contribution to artificial life, believable agents, developmental agent architectures, and interpretable human-agent interaction. It does not claim to solve machine consciousness; instead, it treats consciousness language as a symbolic developmental grammar for fictional lifeforms, with evaluation focused on believability, interpretability, player readability, developmental richness, culture stability, and runtime feasibility.