FACE: Feasible and Actionable Counterfactual Explanations

Current AI explanations can give offensive and impractical advice: The paper demonstrates that existing counterfactual explanation systems may suggest unrealistic changes to individuals—such as advising a person with severe disability to exercise more to get approved for life insurance—because they don't account for what's actually achievable in the real world.

"Closest" doesn't mean "realistic": While most AI explanation systems focus on finding the smallest possible changes to flip a decision, these changes often don't reflect actual patterns in real data, making the suggested paths to success statistically implausible and potentially misleading to users.

Feasible paths require multiple steps, not just one jump: The paper argues that actionable recourse needs to consider realistic progression—for example, someone can't simply double their salary overnight to qualify for a mortgage; they would need intermediate steps like skill development first.

FACE offers a solution using "density-weighted" paths: The proposed FACE algorithm generates counterfactual explanations by finding feasible routes through real-world data patterns, ensuring suggestions are both statistically grounded and practically achievable for individuals seeking to change AI decisions about them.

This matters for fairness and trust in AI systems: By making explanations more realistic and actionable, this work addresses a critical gap in responsible AI—ensuring that when systems deny opportunities to people, the guidance they provide for recourse is actually helpful rather than demoralizing or offensive.