Informative Semi-Factuals for XAI: The Elaborated Explanations that People Prefer

Semi-factual explanations help people understand AI decisions by showing what could change without affecting the outcome. For example, a bank customer learning "even if you asked for double the loan amount, you would still be accepted" gets insight into how robust their approval is, which is different from traditional explanations that show what would need to change to get a different result.

Current AI explanation methods often miss the "why" behind their explanations. The paper identifies that existing semi-factual approaches tell people what can change without affecting a decision, but fail to explain the underlying reasons—such as revealing that a strong credit score is what actually protects an applicant's loan approval even with larger requests.

The new "informative semi-factuals" method adds crucial context by identifying hidden factors influencing AI decisions. By supplementing explanations with information about additional features that matter (like credit score in lending decisions), this approach provides more complete transparency about which factors are actually driving automated decisions.

People prefer more detailed, elaborated explanations over simpler ones. The experimental results indicate that users find value in explanations that go beyond surface-level information to reveal the deeper factors influencing AI outcomes, suggesting that transparency efforts should prioritize comprehensiveness.

This work addresses a gap in responsible AI by making decision-making factors more visible to affected individuals. By revealing hidden features that influence automated decisions, this method helps close information asymmetries between organizations using AI systems and the people subject to those decisions.