WRRA Artificial Intelligence

Abstract

This paper develops WRRA as a minimal-computation artificial intelligence architecture. WRRA transforms SOURCE through RELATION into STATE, converts that state into observable behavior through a RENDERER, and limits every conclusion by explicit OBSERVABLE and BOUNDARY conditions. Its objective is not maximum performance in a fixed environment, but minimum-sufficient computation for survival across changing environments without falling below a defined survival threshold. The study integrates worm-inspired neural modeling, a ground-up WRRA agent without a biological prior, memoryless and recurrent baselines, cue-memory experiments, separate success-and-failure residues, delayed reward assignment, energy reserves, environmental shifts, extreme shocks, luck, inheritance, and exhaustive circuit ablation. Food seeking is treated as prediction and control, while completed experimental outcomes are treated as interpretation. A reduced Caenorhabditis elegans-anchored model identifies chemosensation, short search residue, and energy reserve as the reference minimum survival circuit. The results show that memory is unnecessary when current sensory information is sufficient, but becomes indispensable when cues disappear or environmental rules change. They also demonstrate that apparently inefficient reserve capacity and selective redundancy can protect survival under extreme change. Luck is not eliminated; instead, survival probability and uncertainty are explicitly measured. WRRA therefore proposes a different criterion for artificial intelligence: not the system that performs best everywhere, but the smallest system that retains enough memory, energy, and conditional redundancy to remain viable across relevant environments. Keywords WRRA, artificial intelligence, minimal computation, survival intelligence, embodied AI, Caenorhabditis elegans, C. elegans neural circuit, worm-inspired AI, ground-up agent, memoryless model, RNN, GRU, present residue, cue memory, success and failure memory, delayed reward, reinforcement learning, food seeking, prediction, environmental change, energy reserve, redundancy, robustness, luck, inheritance, survival boundary, circuit ablation, adaptive behavior