Introducing User Feedback-based Counterfactual Explanations (UFCE)

Users can now shape AI explanations to match real-world constraints: UFCE allows people to provide feedback and constraints when asking an AI system to explain its decisions, ensuring the suggested changes are actually practical and achievable rather than theoretically optimal but unrealistic.

AI explanations are becoming more actionable and minimal: The method focuses on identifying the smallest number of meaningful changes needed to change an AI decision, rather than modifying many features across the entire system—making recommendations easier for users to understand and act upon.

Feasibility of AI recommendations has been significantly improved: Testing across five datasets showed UFCE outperformed existing explanation methods in "feasibility," meaning the changes suggested by the AI are more likely to work in practice, which builds appropriate trust in AI systems.

Feature relationships now matter in explanations: The approach considers how different input features depend on and relate to each other, preventing the AI from suggesting contradictory or impossible combinations of changes—a critical step toward more realistic AI guidance.

User involvement strengthens AI trustworthiness: By incorporating user feedback into the explanation process, UFCE demonstrates that collaborative approaches between humans and AI systems can produce more reliable and contextually appropriate explanations, supporting more responsible AI deployment.