Countless change blindness studies have showed that we’re extremely bad at noticing when a scene has changed. We fail to notice objects moving, disappearing, or changing color, seemingly right before our eyes. But sometimes we do notice the change. What sorts of changes are we more likely to notice? I’ve created a simple demo that may (or may not) help answer that question.
Take a look at this movie (QuickTime required). It will show a scene for six seconds. Then it will briefly flash white, and the same scene will be shown for another six seconds. Can you spot what has changed?
I’ve put up a poll at the end of the post so we can see if this demo works (I’m not at all sure it will!). [Update: It looks like I've made the task too difficult. If you don't spot the change the first time you watch, play the movie again. You can repeat until you spot the change or you get bored with the task.]
There are certain types of changes that people are more likely to spot than others. Drug users will notice changes related to their substance of choice more reliably than non-users. In 2004, a team led by Melissa Beck found that that viewers are better at spotting probable changes (a flag starting to flap in the breeze) than improbable ones (a window changing size). Now, with a new team, Beck has started to uncover how this process occurs.
The team showed volunteers pictures like the ones in the movie above (it was actually the same set of photos used in the 2004 study, but this time the researchers also tracked the viewers’ eye movements), for six seconds pre-change and six seconds post-change. They then asked what had changed in the photo (in fact most of the time nothing at all had changed!).
As in the original study, viewers were better at spotting changes that were likely to occur in the real world than at spotting improbable changes like an object changing size or color. But several twists in this study allowed Beck’s team to uncover when the discrepancy was occurring. After the viewers completed the initial change-blindness task, they were asked to recall which scenes they had seen previously. There are two possibilities for what they saw:
Everyone saw the original image, but some viewers saw the probable change, and some viewers saw the improbable change. If a viewer had seen the probable change previously, then they were tested by being shown the original version and the improbably changed version, and asked which image they had seen previously. In every case, viewers had only seen one of the two images they were shown. Here are the results:
Viewers were much more accurate at remembering which scene they had seen before compared to their accuracy detecting the change in the first place. More importantly, there was no difference in their long-term memory for probable versus improbable changes, even though they had been more accurate in detecting probable changes.
So people accurately remember the original scene, but somehow aren’t able to report the change they see. Why is that? Take a look at this graph, showing where viewers were looking after they had failed to detect a change in a scene:
Even when they hadn’t correctly spotted the change, they looked at the place where the change occurred for longer when it was a probable change compared to an improbable change. When they did spot the change, they looked at the changed area longer, too — but for both probable and improbable changes.
What does all this mean? Beck’s team argues that the bias toward spotting probable changes occurs during retrieval: even though they remember what images they’ve seen, viewers aren’t as good at accessing those memories when a change is improbable. The memory of an image, unfortunately, doesn’t do us any good if we can’t retrieve it when we need it!
If we replicated Beck’s results, more people should have seen the taxi move (probable) than the jeans changing color (improbable). If we didn’t replicate, perhaps readers can explain why in the comments.
Beck, M.R., Peterson, M.S., & Angelone, B.L. (2007). The roles of encoding, retrieval, and awareness in change detection. Memory & Cognition, 35 (4), 610-620.