Słownik AI
Kompletny słownik sztucznej inteligencji
XAI (Explainable AI)
Set of techniques and methods that make artificial intelligence systems understandable to humans, transforming complex decision-making processes into interpretable explanations.
Local interpretability
Ability to explain a model's prediction for a specific instance, identifying the features that influenced this particular decision.
Global interpretability
Overall understanding of an AI model's behavior, explaining how it makes decisions in general across the entire dataset.
Black box
AI system whose internal functioning is opaque or incomprehensible to humans, making it difficult to explain its decisions and reasoning processes.
LIME (Local Interpretable Model-agnostic Explanations)
Interpretation technique that explains individual predictions by creating simple local models that approximate the behavior of the complex model around a specific prediction.
SHAP (SHapley Additive exPlanations)
Game theory-based explanation method that quantifies the impact of each feature on the final prediction by fairly distributing credit among all features.
Decision traceability
Ability to track and document each step of an AI system's decision-making process, from input data to final result, to ensure auditability.
AI auditability
Possibility to systematically examine an AI system to assess its compliance with standards, regulations, and ethical requirements, particularly regarding bias and discrimination.
Model explainability
Ability of an AI system to provide understandable and coherent justifications for its decisions, allowing users to understand the underlying reasoning.
Feature importance
Quantitative measure of the influence of each input variable on the model's predictions, allowing to identify the most determining factors in decision-making.
Decision visualization
Graphical techniques representing the decision process of an AI model, allowing users to intuitively understand how predictions are generated.
Meta-explanations
Second-level explanations that describe why the AI model produces certain explanations rather than others, increasing confidence in the explanatory system itself.
Automated reports
Systems automatically generating detailed reports explaining AI decisions, including the data used, the reasoning followed, and confidence levels.
Human validation
Process by which human experts review and validate the explanations provided by AI systems to ensure their relevance and accuracy.
Model documentation
Structured and complete recording of the characteristics, performance, limitations, and behaviors of an AI model to ensure its transparency and reusability.
Bias diagnosis
Systematic analysis of a model's decisions to identify, quantify, and understand potential discriminations based on protected or sensitive characteristics.
Attention heat maps
Visualizations showing the areas or features on which an AI model particularly focuses to make its decision, facilitating the understanding of its reasoning.