Change detection has no foresight: Measuring advanced knowledge of changes across displays

David E Fencsik 1,2, Todd S Horowitz 1,2, Stephen J Flusberg 1, Jeremy M Wolfe 1,2

1 - Brigham and Women's Hospital
2 - Harvard Medical School

In a visual change detection task, observers search for a change between two displays presented in alternation. When do observers have knowledge about the presence of a change? Does knowledge accumulate gradually, or is the change detected swiftly, but only after the changed region is selected by attention? While several researchers have reported evidence of awareness of a change prior to or in the absence of explicit report ("implicit change detection" or "mindsight"; e.g., Rensink, PsychSci, 2004), others have found contrary results (e.g., Mitroff & Simons, VisCog, 2002). We applied a new, event-related signal detection method in order to supply converging evidence. Observers viewed alternating displays repeatedly in brief exposures ("frames"). Following each frame, observers indicated their confidence along a rating scale ranging from "highly confident same" to "highly confident different". Each trial ended when the observer indicated response certainty by pressing a button. Ratings were averaged across trials after being aligned relative to the final response frame. From these data, we generated ROCs for each frame relative to the time of response. This procedure has produced evidence for gradual increases in sensitivity in control tasks (e.g., simulated diffusion; motion coherence detection). In change detection, however, we observed a step function. Stimulus sensitivity was high for the final response, but near zero just two frames prior to response. Observers have no information about the presence of a change prior to directing attention to the region that is changing across displays.