Perceiving Uncertainty: how visual encoding, socially mediated doubt, and task complexity influence human decision-making

Human decision-making during autonomous underwater vehicle (AUV) operations is fundamentally influenced by uncertainty in environmental information. This exploratory pilot study investigates how during path planning, task dynamics can influence operator performance and risk tolerance, relative to a 2D visualisation of bathymetric data uncertainty. Using bathymetric data obtained from field trials of prototype AUVs, we visualise uncertainty in these data using Gaussian Processes (GP), manipulating the hyperparameters. Participants (n=18) completed 108 free-form path planning trials, overlayed on a 2D contour map of bathymetric uncertainty, while avoiding marine dangers. Additional uncertainty was introduced via an AI agent with one of three face realism conditions, designed to sow doubt in the participants perceived task performance. Bayesian modelling suggests Contour lowered redraw rates, with a modest U-curve relationship. Variance followed an inverted U-curve effect on redraw rates, with moderate values reducing ambiguity and improving performance. AI agent appearance shaped trust behaviour, while environmental complexity reduced risk tolerance. Results of our experimental pilot study show that visual uncertainty, social agent appearance, and task complexity systematically shape human trust, risk tolerance, and decision-making behaviour during path planning.