For Better or Worse: The Impact of Counterfactual Explanations' Directionality on User Behavior in xAI

Upward counterfactual explanations significantly outperform other explanation types: When AI systems showed users how to improve outcomes (upward CFEs), participants performed better at understanding the system and extracting useful knowledge compared to explanations showing worse scenarios or no explanations at all.

The direction of explanations matters for user learning: Users who received upward-directed explanations (showing positive changes) developed statistically higher explicit knowledge of how the system worked, suggesting that explanation directionality directly impacts what people learn from AI systems.

Mixed explanations offer a middle ground: Combining both upward and downward counterfactual explanations improved user performance compared to downward-only explanations or no explanations, indicating that variety in explanation types may have practical benefits.

"Regulatory fit" is key to effective AI explanations: The study suggests that alignment between the type of explanation and the user's task goal—a concept called regulatory fit—may be crucial for making AI systems more understandable and useful, pointing to an important design principle for future explainable AI systems.

Explanation design choices have real behavioral consequences: This research demonstrates that seemingly technical choices about how to frame AI explanations (showing improvements vs. deteriorations) meaningfully affect how users interact with and understand automated systems, highlighting the importance of thoughtful design in responsible AI deployment.