Towards Directive Explanations: Crafting Explainable AI Systems for Actionable Human-AI Interactions

Most current AI explanations are designed for experts, not everyday users: The paper identifies a critical gap where existing Explainable AI (XAI) methods prioritize technical audiences rather than non-technical consumers, making explanations "overwhelmingly complex" and difficult for average people to understand and act upon.

Explanations should be "directive" and actionable: Rather than simply describing how an AI system works, effective explanations for non-technical users should actively guide them toward achieving desired outcomes, transforming explanations from passive information into tools for meaningful human-AI collaboration.

Current XAI approaches fail to support user goals: The research reveals that typical AI explanations rarely help users understand what actions they can take to change or influence AI predictions, leaving them unable to navigate or improve their interactions with AI systems.

Transparency and trust require user-centered design: The paper emphasizes that building genuine transparency and trust among non-technical users requires fundamentally rethinking how explanations are crafted—moving away from technical accuracy as the primary goal toward designs that prioritize user comprehension and empowerment.

Human-AI collaboration needs further research: The paper identifies open questions about how to better design AI systems that truly collaborate with users, suggesting this remains an important frontier for responsible AI development that balances system performance with meaningful human agency.