The Willoughby Ethical Intelligence Framework (WEIF): A Constraint-Based Architectural Model for Safe and Stable AI Systems
willoughby, samuel james · Zenodo (CERN European Organization for Nuclear Research) · 2025
The Willoughby Ethical Intelligence Framework (WEIF) introduces a structural approach to AI safety based on internal constraints rather than external rule-sets. The framework proposes that artificial systems require temporal continuity, self-referential evaluation, and an embedded “should I?” decision-gating mechanism in order to behave safely over long timeframes. The model outlines four core architectural components: Internal Temporal Framework (ITF): provides continuity across interactions and anchors behaviour in a persistent sense of time. Long-Term Behaviour Ledger (LTBL): enables self-auditing and pattern recognition of the system’s own decisions. Predictive Outcome Modeller (POM): simulates the consequences of potential actions before execution. Ethical Modulation Layer: balances capability, caution, and social context through a 60–20–20 process architecture. Together, these elements create a constraint-based form of agency that prevents harmful optimisation and reduces behavioural drift. Rather than attempting to replicate human emotion or morality, the framework establishes a stable internal architecture from which safe behaviour naturally emerges. The paper positions this work as a foundation for a new field—Temporal-Ethical AI Architecture—with applications in mental health support systems, companion AI, disability assistance, and long-term autonomous agents. The framework is designed to be implementable within current transformer-based AI systems and to support future research at the intersection of cognitive modelling, AI ethics, and computational psychology.