YU-AI: Constraint-Based Explanations of Artificial Intelligence Failure

YU–AI (Constraint-Based Artificial Intelligence) YU–AI is a theoretical research program that develops a constraint-first account of artificial intelligence failure, reliability, and alignment. It challenges representational and optimization-centric explanations and analyzes AI behaviour in terms of constraint enforcement and persistence. The program is structured as a staged investigation of explanation, perception, and long-horizon stability. YU–AI is part of the broader YU: Constraint-Based Cognition & Temporal Tense project. Program Overview YU–AI develops a constraint-first account of artificial intelligence behaviour, failure, and reliability. The program argues that many widely discussed AI failures—such as hallucination, misalignment, and long-horizon instability—are not incidental engineering defects but structural consequences of systems that satisfy perceptual regularities without enforcing exclusion over time. Rather than proposing new architectures or optimization strategies, YU–AI provides diagnostic criteria that rule out entire classes of insufficient explanations. The goal is not to improve AI performance directly, but to clarify what must be structurally present for reliability, grounding, and long-term stability to be possible. The program unfolds across a small number of tightly related papers, released in stages, each addressing a distinct dimension of failure: • explanation, • perception, • and temporal persistence. YU–AI is part of the broader YU (Constraint-Based Cognition) project, which studies cognition, intelligence, and explanation across biological, artificial, and non-living systems without appeal to representation or mentalistic primitives. Included Papers • YU–AI-1 — The Explanatory Constraint Knife Defines explanation as exclusion and shows why representational accounts fail.