Successfully completing many forms of behaviour requires that humans look in the right place at the right time: This has generated a large volume of research aimed at understanding how the eyes are guided. Prominent in research over a number of years now has been the notion that visual salience (low-level conspicuity) is a crucial causal factor in determining where people look. However, there is now a critical mass of evidence, drawn from a range of experimental settings, that visual conspicuity is unlikely to play a prominent role in determining where we look under most viewing conditions: Salience models are disappointingly poor at predicting where humans fixate and can fail completely to account for eye guidance during certain visual tasks. In this talk I will review the current evidence against purely image salience accounts and consider the emerging evidence for (1) the levels at which fixation selection might occur and (2) new models for explaining eye guidance.
In our lab, we are taking somewhat of a different approach to developing models of eye guidance than others currently are. An emerging theme across a range of activities is that motor behaviours display prominent systematic tendencies in how they are executed, and these are highly informative in modelling behaviour. We found that systematic biases in saccade magnitude and direction play an important role in selecting where to look: A model of eye guidance based solely on these systematic biases and ignoring all visual information outperforms existing salience models. While it is often supposed that basic visual features play a major role in determining where we look, our evidence provides strong support that in isolation they play a relatively minor role compared to oculomotor tendencies.