IA explicable (XAI)
Definición
La IA explicable busca hacer comprensible el comportamiento del modelo: qué entradas o características motivaron una decisión, o quéhe model "thinks" in human terms. This supports trust, debugging, and compliance.
Soporta seguridad de IA (auditoría) y sesgo en IA (comprensión de resultados injustos). Required or recommended in regulated domains (por ej. credit, hiring, healthcare). Trade-off: post-hoc explanations (SHAP, LIME) are flexible but can be approximate; inherently interpretable models are limited in expressiveness.
Cómo funciona
Feature attribution (por ej. SHAP, LIME) asigna puntuaciones de importancia a las características de entrada para una predicción dada. Attention visualization shows which tokens or regions the model attended to. Natural language explanations (por ej. from an LLM or a dedicated module) describe the decisión in words. Inherently interpretable models (por ej. linear models, decisión trees, rule lists) are interpretable by diseño. Choice depends on the model type and use case: post-hoc methods work with black boxes but may not reflect the true mechanism; interpretable models are more faithful but less flexible. Evaluate explanations for fidelity (do they igualar the model?) and usefulness (do they help users or auditors?). Integrate with evaluation and bias audits where needed.
Casos de uso
Explainability matters when users or regulators need to understand why a model made a given decisión (compliance, trust, debugging).
- Explaining credit, hiring, or medical decisións for compliance and users
- Debugging and improving model behavior via attributions
- Building trust and transparency in high-stakes applications
Documentación externa
- Interpretable Machine Learning (Molnar) — Free online book
- SHAP documentation