An educated guess: how coral reef fish make decisions under uncertainty

Abstract

Making informed decisions is fundamental to intelligent life and is crucial to survival. The problem many organisms face, is how to make optimal decisions in the face of incomplete, unreliable, or conflicting information. In many aquatic environments, fish use visual information to guide key behaviours, but the environment itself can alter or mask the very signals they rely on. Here, we asked how a highly visual species, Rhinecanthus aculeatus, responds to a learned discrimination task as signal reliability decreases, and whether probabilistic information gained during previous experience can be incorporated into their decision strategy. Fish were first trained to select a target (dark grey circle) from three distractors (light grey circles). In the first experiment, the target was more likely to appear in one of four possible stimulus positions. In the second experiment, the target appeared in all positions equally. In a series of trials, the difference in brightness between the target and distractors was reduced until all four stimuli were identical. We found that target selection accuracy decreased with decreasing target and distractor disparity. However, in Experiment 1 where the target was more likely to be in one position, fish increasingly selected stimuli in the biased position as target selection accuracy decreased, but not in Experiment 2. These results demonstrate 1) that fish learned more than a simple select/avoid rule based on stimulus brightness, but also integrated information (stimulus position) that could be considered ancillary to the primary task. 2) Fish can learn probability distributions and apply this knowledge as uncertainty increases, ultimately increasing the overall frequency of correct choices. Our results reveal that probabilistic decision rules can be used by fish when visual information is unreliable, indicating a possible mechanism for decision making given the inherent noise in incoming sensory information.