Abstract
To sample information efficiently, the fovea needs to be directed to the places where it is most needed to resolve the visual input. In this series of experiments, we define efficiency as the proportion of fixations that target locations where higher resolution is required. By our measure of efficiency, some people are near-optimal, some random, and some even counter-optimal. This conclusion comes from simple arrays of line segments, under the assumption that these will minimize the individual differences in experience that could introduce unnecessary variation in behaviour. We then systematically varied surface level properties of the stimuli to test whether individual differences in efficiency generalise from one visual context to another. Across five experiments (with a total N of 105) we measured the efficiency of healthy observers as they searched through simple line segment stimuli, desktop icons, polygons and pens. Objectively, the ease of implementing an efficient strategy should have been the same across all contexts, but we observed large differences between contexts. In particular, the efficiency of searching for a particular orientation among distractors, whether they are pens or lines, is variable across individuals, while eye movements during search for a particular identity among objects are uniformly closer to optimal. These results demonstrate that small changes to surface-level details of the task can lead to large changes in measured behaviour. This result could help explain contradictory evidence about the extent to which eye movements are driven by information gain, and also challenge the assumption that observations based on simple visual stimuli scale up to more complex objects.