![[2512.15783] AI Epidemiology: achieving explainable AI through expert oversight patterns](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2512.15783] AI Epidemiology: achieving explainable AI through expert oversight patterns
Summary
The paper presents 'AI Epidemiology', a framework for enhancing explainability in AI systems through expert oversight, using population-level surveillance methods to predict and validate AI output failures.
Why It Matters
As AI systems become increasingly complex, ensuring their reliability and interpretability is crucial for safe deployment. This framework provides a novel approach to governance, enabling experts to oversee AI outputs without requiring deep machine learning knowledge, thus democratizing AI oversight and potentially improving public trust in AI technologies.
Key Takeaways
- AI Epidemiology uses statistical methods to enhance AI explainability.
- The framework standardizes expert interactions to predict AI output failures.
- It provides automatic audit trails and reliability scores for AI outputs.
- Experts can govern AI systems without needing machine learning expertise.
- The approach ensures continuity in governance during model updates.
Computer Science > Artificial Intelligence arXiv:2512.15783 (cs) [Submitted on 15 Dec 2025 (v1), last revised 19 Feb 2026 (this version, v2)] Title:AI Epidemiology: achieving explainable AI through expert oversight patterns Authors:Kit Tempest-Walters View a PDF of the paper titled AI Epidemiology: achieving explainable AI through expert oversight patterns, by Kit Tempest-Walters View PDF Abstract:AI Epidemiology is a framework for governing and explaining advanced AI systems by applying population-level surveillance methods to AI outputs. The approach mirrors the way in which epidemiologists enable public health interventions through statistical evidence before molecular mechanisms are understood. This bypasses the problem of model complexity which plagues current interpretability methods (such as SHAP and mechanistic interpretability) at the scale of deployed models. AI Epidemiology achieves this population-level surveillance by standardising capture of AI-expert interactions into structured assessment fields: risk level, alignment score, and accuracy score. These function as exposure variables which predict output failure through statistical associations, much like cholesterol and blood pressure act as exposure variables predicting cardiac events. Output-failure associations are subsequently validated against expert overrides and real-world outcomes. The framework places zero burden on experts and provides automatic audit trails by passively tracking expert convergence ...