The Future is not Set: Visual Decision-Making with Predictive Models

Predictions, such as COVID forecasts and AI recommendations, are now commonly seen in our daily lives and integrated into workflows. However, these predictions come with inherent imperfections—such as error and uncertainty—that raise significant challenges for individuals trying to understand them and make appropriate decisions. My research tackles the problems of trust in predictive models and decision-making. In this talk, I will focus on two areas: providing visual explanations to identify machine learning errors, and using uncertainty visualizations to build trust in election forecasts, both of which help lead to more appropriate decisions. I will also briefly discuss the application of perceptual science and virtual reality in presenting model predictions and aiding decision-making. The ultimate goal of my research is to ensure appropriate decisions for different user groups.