Exploring the Impact of Explainable AI and Cognitive Capabilities on Users' Decisions

High AI confidence can be misleading: The study found that when AI systems displayed high confidence levels, users relied more heavily on the AI's decisions while experiencing less cognitive strain—but this didn't necessarily improve decision accuracy, suggesting people may be over-trusting confident-sounding AI systems.

Popular explanation methods may not be most effective: Contrary to common practice, feature-based explanations (which highlight which factors the AI considered) did not improve user accuracy compared to other explanation styles, indicating that the most widely-used XAI approach may not be the best for helping people make better decisions.

Counterfactual explanations show promise despite complexity: Although counterfactual explanations (showing "what would need to change for a different outcome") were harder for users to understand, they actually led to better overall decision accuracy and reduced cognitive load when the AI was correct—suggesting this less intuitive approach deserves more attention.

Users want explanations but may overlook AI system information: Both people with high and low cognitive engagement preferences prioritized seeing explanations over other AI information like confidence scores and accuracy rates, yet they still didn't always make better decisions, highlighting a gap between what users want and what actually helps them decide wisely.

Personality differences don't determine AI decision-making ability: The study found no significant differences between people with different cognitive preferences in their accuracy or mental effort when using AI systems, suggesting that one-size-fits-all AI interfaces may be appropriate across diverse user populations.