We compare eye movement strategies across a range of different stimulus sets to test the prediction that eye movements are guided by expected information gain. When searching for a simple target that has been defined based on orientation, interindividual variability is high, and a large proportion of eye movements are directed to locations where peripheral vision would have been sufficient to determine whether the target was present there or not. In contrast, when searching for a target defined based on identity, eye movements are similar across individuals and highly efficient, being directed almost exclusively to the locations where central vision is most needed. The results suggest that for most people, the way they search for a simple feature (orientation) is not directly representative of the way they search for objects based on their identity. More generally, the results highlight that because humans are adaptable, contradictory theories can be accurate descriptions of search in particular contexts and individuals. For a complete and accurate account of human search behavior to be achieved, the conditions that shift us from one mode of behavior to another need to be part of our models.
Public Significance StatementThe current research tests the predictions of theories of eye movement control that are based on a principle of expected information, that is, how much new information one can expect to gain from making each eye movement. We tested people's strategies across a range of different search displays, from simple line segments to computer desktop icons, and designed the displays to ensure a strategy based on expected information should be equivalently easy to implement across all these situations. Our results show two distinct patterns of behavior: When participants are searching for a particular object (e.g., one specific pen among other pens) they clearly and uniformly match the predicted optimal strategy. But when they are searching for an object defined based on its orientation (e.g., a pen tilted a particular way) they exhibit a broad range of different behaviors. The results demonstrate the striking adaptability of human strategies and show us that, in order to devise a unifying theory that can explain and predict eye movements, we need to account for how they change across situations.