Automated Rationale Generation: A Technique for Explainable AI and its Effects on Human Perceptions

AI explanations can be automatically generated and tailored to different styles: The research demonstrates that AI systems can be trained to produce natural language explanations for their decisions in multiple formats, moving beyond simple outputs to more human-like communication that can be customized based on context.

Detailed explanations help people trust and understand AI better: Study participants preferred more detailed rationales over simpler ones when trying to form a mental model of how an AI agent works, suggesting that transparency through comprehensive explanation supports more informed human-AI interaction.

Generated explanations successfully communicate both successes and failures: The research found that automatically generated rationales were effective at helping users understand not just why an AI made correct decisions, but also why it failed or behaved unexpectedly—an important capability for responsible AI deployment.

User perceptions align with explanation design choices: When researchers intentionally varied features of the generated explanations, users perceived these differences as intended across dimensions like confidence, humanlikeness, and understandability, indicating that explanation design directly influences how people interpret AI behavior.

Explainability is a learnable skill for AI systems: Rather than relying on post-hoc explanations added after decisions are made, this approach shows AI can be trained from human explanation examples to generate its own rationales in real-time, potentially making explainability more scalable and integrated into AI systems.