The Influence of Explainable Artificial Intelligence: Nudging Behaviour or Boosting Capability?

Different types of AI explanations have different effects on how people behave: The paper distinguishes between "nudge-like" explanations (local and concept-based) that subtly influence decisions, and "boost-like" explanations (global and counterfactual) that help people develop better understanding and decision-making skills. This matters because the same AI system could either manipulate or educate users depending on how its decisions are explained.

Understanding XAI's influence is essential for ethical AI deployment: The authors argue that measuring how explanations actually affect human behavior and cognition should be a standard part of responsible AI development. Without this measurement, organizations cannot know whether their explanations are genuinely helping people make better decisions or simply steering them toward predetermined conclusions.

Explanations designed as "boosts" may be preferable for human autonomy: The paper suggests that explanations aimed at building human capability (like showing global patterns or alternative scenarios) better support ethical collaboration than those designed primarily to influence behavior. This distinction is important for maintaining meaningful human agency in AI-assisted decision-making.

Current XAI practice may lack intentionality about behavioral effects: The framework implies that many organizations deploying XAI may not have deliberately considered whether their explanations nudge or boost users—suggesting a gap between current practice and responsible AI principles.