Engaging with AI: How Interface Design Shapes Human-AI Collaboration in High-Stakes Decision-Making

Automation bias remains a critical problem in human-AI collaboration: The paper confirms that human + AI teams often perform worse than AI alone because humans tend to follow AI recommendations even when incorrect, highlighting the need for better interface design to counteract this tendency.

Traditional text explanations may not be enough: Since humans rely on quick, intuitive "System 1" thinking, users often ignore or insufficiently engage with text-based explanations of AI reasoning, suggesting that explanation alone is insufficient for responsible AI deployment in high-stakes domains like healthcare.

Cognitive forcing functions can improve decision-making: The study found that mechanisms requiring users to actively engage with AI recommendations—such as displaying AI confidence levels and performance visualizations—enhanced both user engagement and the quality of human-AI collaborative decisions in a diabetes management scenario.

Interface design directly impacts trust calibration: Different decision-support mechanisms had measurable effects on how much users trusted AI suggestions, indicating that how information is presented shapes whether humans appropriately balance AI input with their own judgment rather than blindly accepting recommendations.

Visual and interactive elements outperform passive information: The research suggests that moving beyond static text explanations toward visual displays and interactive elements that force deliberate consideration can help ensure humans remain meaningfully engaged in high-stakes decisions rather than defaulting to automation bias.