Ensemble of Counterfactual Explainers

Multiple explanation approaches are better than one: The paper demonstrates that combining several different counterfactual explainers produces more comprehensive explanations than relying on a single method, addressing limitations where individual explainers only capture some desirable properties (such as minimality or actionability).

Explanations can be made practical and actionable: The ensemble approach prioritizes "actionability," meaning it generates explanations that show users what specific changes to input features would alter an AI system's decision—information that is directly useful for understanding and potentially contesting automated decisions.

The method works across different types of AI systems: Because the approach is "model-agnostic" and "data-agnostic" (using autoencoders as a wrapper), it can be applied to various AI systems and data types without requiring specialized redesign, making responsible AI practices more broadly accessible.

Explanations can balance competing goals: The ensemble strategically combines explanations that emphasize different priorities—such as finding minimal changes versus diverse options versus realistic scenarios—rather than forcing a choice between them, giving stakeholders more complete insight into how decisions are made.

Diversity in explanations matters: The method uses "diversity-driven selection" to ensure users receive varied counterfactual examples rather than similar ones, which helps prevent narrow or biased understanding of what factors influence an AI system's decisions.