Trusting AI: Does Uncertainty visualization affects decision-making?

Jonatan Reyes ^* Anil Ufuk Batmaz Marta Kersten-Oertel

• Concordia University, Montreal, Canada

Decision-making based on AI can be challenging, particularly when factoring in the uncertainty associated with AI predictions. Visualizing uncertainty in AI solutions refers to techniques that use visual cues to represent the level of confidence or uncertainty in an AI model's outputs, such as predictions or decisions. To investigate the impact of visualizing uncertainty in AI solutions, we considered human factors (e.g., visual perception and cognition) during the design of model outputs. We conducted a user study with 147 participants using static classic gaming scenarios as a proxy to show human-AI collaboration in decision-making. Our study measures changes in decisions, trust in AI, and decision-making confidence when uncertainty is visualized in a continuous format in comparison to a binary output of the AI model. Interestingly, we found that visualizing uncertainty significantly strengthens trust in AI for 58% of participants with negative attitudes towards AI, and 31% of these participants found the visualization of uncertainty useful. Additionally, size was identified as the visualization method that most impact in individuals' trust in AI and confidence in their decisions. We also found a strong association between gaming experience and decision changes when uncertainty was visualized, and a strong association between trust in AI and individuals' attitudes towards AI. Our study provides insights into understanding the psychology of participants, specifically how individuals perceive uncertainty in AI models. These findings provide significant implications for the design of human-AI based decision support systems.