Abstract
Memorability is an image property that is remarkably consistent across viewers, as people tend to remember and forget the same images. In spite of this consistency, however, people have poor ability to predict what they will later remember (Bainbridge et al., 2013, Isola et al., 2014). Additionally, the specific factors that influence image memorability are still largely unknown. In this experiment, we tested whether participants receiving training could over time learn the difference between memorable and forgettable images. Participants (N=100) on the online platform Prolific viewed a sequence of randomly ordered face images, for a total of 180 images. Stimuli included an equal number of highly memorable and forgettable images as computed by prior memory experiments, with both image categories matched for demographic characteristics such as age, gender, and ethnicity. On each trial, participants were shown a new face and were asked whether they thought the face would be remembered or forgotten by most people. Immediately following their response, they were told whether they had answered correctly or incorrectly. Participants started at an average accuracy of 61%, suggesting some initial ability to distinguish between memorable and forgettable faces. When looking at subject performance across time, we observed a significant increase in accuracy over the first half of trials (90 trials). This trend was strongest in participants who started the task with low accuracy, yet participants with initial above average performance did not show additional improvement, suggesting a ceiling effect. Overall, these results suggest that with repeated exposure and feedback, people may be able to increasingly learn the difference between memorable and forgettable face images, a distinction which has thus far been opaque. Further investigation of the specific stimuli that drive learning and comparison with other image categories may offer insight into the mysteries behind why we remember what we do.