Understanding the different types of explanations in AI is essential for interpreting how complex models make decisions. You will often encounter three main forms of explanation: local explanations, global explanations, and example-based explanations. Each serves a different purpose and is suited to different audiences and use cases.

Local explanations focus on clarifying a model's decision for a specific instance. For example, if an AI predicts that a loan application should be denied, a local explanation would highlight which features of that particular application—such as income or credit score—had the most influence on the outcome. In contrast, global explanations aim to summarize the overall behavior of a model across all data. This could involve describing which features are generally most important for predictions or summarizing the decision boundaries of the model as a whole.

Feature importance is a key concept that bridges both local and global explanations. It quantifies how much each input feature contributes to the prediction. For global explanations, feature importance can help you understand which features the model relies on most across all decisions. For local explanations, feature importance can show which features were most influential for a single prediction.

Example-based explanations provide insight by referencing specific data points. These explanations may highlight prototypes (typical examples that represent a class) or counterfactuals (examples that show how small changes to input would change the prediction). By relating decisions to real or hypothetical examples, these explanations can be intuitive and practical, especially for non-technical users.

To help you compare and remember these explanation types, consider the following summary table: