Should XAI Nudge Human Decisions with Explanation Biasing?

Nudge-XAI attempts to guide users toward better AI-assisted decisions through strategically biased explanations, rather than forcing compliance. The approach aims to enhance human autonomy by helping users make more informed choices that align with AI recommendations, while still preserving their ability to override the system.

User responses to explanation nudges are highly diverse, according to cluster analysis of experimental results. This diversity suggests the approach can work across different user types, but also reveals a critical challenge: some users distrust AI systems and deliberately make decisions contrary to AI suggestions, even when presented with biased explanations.

The core tension is between helpfulness and manipulation: While nudging explanations could help users avoid AI failures and reduce risks from unexamined AI decisions, the paper acknowledges the ethical concern that automatically biasing explanations—even with good intentions—raises questions about transparency and informed consent in human-AI interaction.

Personalization is essential for responsible implementation. The paper identifies that a one-size-fits-all nudge strength doesn't work effectively. Different users require different levels of nudging based on their trust in AI and decision-making patterns, suggesting that any deployment would need to adapt to individual user profiles.

This approach highlights a fundamental challenge in responsible AI: balancing user autonomy with decision quality. The paper demonstrates that simply providing better explanations isn't enough—systems must account for human psychology and trust dynamics to genuinely support better human decision-making.