A brief history of evidence for change blindness

The literature on change blindness has a long history, dating back at least to the late 1970s (see Simons and Levin, 1997, for a review). Systematic work on change blindness largely began with studies of eye movements during reading (Blanchard et al., 1989; McConkie and Zola, 1979; Morris et al., 1990; Pollatsek and Rayner, 1992) and studies of displacement detection across saccades (Bridgeman et al., 1975; Li and Matin, 1990; Mack, 1970; Wallach and Lewis, 1965). Only more recently has the focus of change blindness research turned to the perception and representation of natural scenes. Early work on displacement detection focused primarily on the issue of whether or not the perception of the visual world is suppressed during eye movements and also whether information from consecutive fixations can be integrated into a single representation. In a typical experiment, observers might view a dot presented parafoveally and then, during a saccade to the location of the dot, the dot would be shifted in position. Typically, shifts of less than 10% of the extent of the eye movement are not noticed by the observer. Displacement detection is quite poor when saccades are long and the displacements are relatively short.

Unlike studies of displacement detection, change detection studies during reading have focused more on the identity than on the position of the changed item. When reading, changes that occur during saccades are rarely noticed. Observers continue reading at the same rate without noticing substantial changes to the appearance or even the presence of words that are not currently fixated. For example, if observers are asked to read text in AlTeRnAtInG cAsE, they do not notice when the case of each letter on the screen changes during every saccade (McConkie and Zola, 1979). In contrast, observers viewing the same displays with changes unconstrained by their fixation patterns cannot easily read the text because the now-visible changes produce massive transient signals throughout the display; it is difficult to read text that we see changing. The presence of a saccade during the change appears to prevent the automatic detection of changes by inhibiting or masking the transient signal that would otherwise be distracting. In order to detect case changes that occur during a saccade, observers must actively attend to the letter, memorize it, and then compare that letter before and after a saccade.

More recent studies of saccade-contingent change blindness have produced parallel results using natural scenes rather than text (Grimes, 1996; Hollingworth and Schrock, 2001; Hollingworth et al., 2001a; McConkie and Currie, 1996). For example, when studying a photograph for a subsequent memory test, observers often failed to notice when two central people in the scene swapped hats during a saccade (Grimes, 1996). Many even failed to notice when two people exchanged heads! Blindness for these scene changes disappeared entirely when the changes occurred during fixation, suggesting that something about the representation or comparison process was disrupted by the saccade.

In addition to studies of saccade-contingent change detection, a number of early experiments used change detection tasks to explore the limits of short-term visual memory (e.g., Pashler, 1988; Phillips, 1974). These experiments briefly presented an array of letters, dots, or other simple stimuli, followed by a brief delay interval and then by a slightly changed array. Unlike most of the saccade-contingent change studies, the observer's primary goal in these studies was to try to detect changes. Consequently, change blindness cannot be attributed to the unexpected nature of the change. Instead, failed change detection was attributed to the limits of short-term visual memory. Observers were unable to hold all of the information in the array in memory during the brief delay interval.

In a seminal study in the change blindness literature, Rensink et al. (1997) adapted these one-shot change detection tasks to explore prolonged search for change. They confirmed that change blindness is not specific to changes made across saccades; it can occur during fixation and even occurs when observers actively search for changes. These failures of visual awareness result whenever the transient signal that would otherwise accompany a change is disrupted. In their visual search flicker task, observers view a rapid alternation of an original and a changed scene, with a brief blank screen following each image presentation. The blank screen, like a saccade, disrupts the localized transient signal that would otherwise be produced by the change, giving the display its flickering appearance. Even though they are actively searching, observers often require many alternations to detect large changes, sometimes searching unsuccessfully for up to a minute. Notably, the level of change blindness does not appear to diminish even when subjects have extra time to study the images or when they have practiced performing the task (Rensink, 2000c; Rensink et al., 1997). Perhaps more importantly, the nature of the change influences the speed of detection: changes to objects of central interest in the scene are detected more rapidly than changes to irrelevant or peripheral objects (Rensink et al., 1997). This center of interest effect reflects the importance of attention in change detection. Changes to objects in a scene that are more likely to be the focus of attention are detected more rapidly than those less likely to be attended. This finding suggests that attention plays a functional role in the eventual detection of changes and that change blindness occurs when observers fail to devote focused attention to the changing item both before and after the change.

However, other work suggests that attention to an object does not guarantee detection, particularly when changes occur unexpectedly (Levin and Simons, 1997; Levin et al., 2002; Simons and Levin, 1998; Simons et al., 2002). For example, nearly two-thirds of observers failed to report any change when the only actor in a motion picture unexpectedly was replaced by a different person across a cut from one camera angle to another (Levin and Simons, 1997). In fact, from 35% to 65% of observers typically fail to notice when a real-world conversation partner is surreptitiously replaced during the middle of an interaction (Levin et al., 2002; Simons and Levin, 1998). These studies suggest that attention to an object is necessary for change detection, but it is not sufficient to guarantee detection. Rather, successful detection requires attention to the specific features of the object that are changing so that observers can encode those features and then compare them following the change.

Given that successful change detection implies the existence of a representation, evidence for change blindness has sometimes been taken to imply the absence of any internal representation of the prechange feature—that our representation of the visual world is more sparse than we might otherwise believe (e.g., O'Regan and Noe, 2001; Rensink, 2000a; Simons and Levin, 1997). Although this conclusion might well be true, it does not follow logically from existing evidence for change blindness. The absence of detection does not necessarily imply the absence of a representation. Change detection could fail even in the presence of a complete and detailed representation of the prechange scene, provided that observers do not compare that initial representation to the postchange scene (Scott-Brown et al., 2000; Simons, 2000b).

An Aside: What Makes Change Blindness Interesting

Findings of change blindness are interesting not because they imply the absence of a representation, but because they run counter to our intuitive belief that unusual or unexpected events will draw our attention. These intuitions reflect an understanding that small, irrelevant changes will not capture attention. For example, nobody would be particularly surprised by the failure to notice a subthreshold change to the color of a single pixel in a photograph of a natural scene. However, people are surprised when the changes are large and supraliminal. To some degree, these intuitions are entirely reasonable based on our daily experiences. After all, most sudden changes in the world do produce a visible change signal and rarely occur quickly enough to start and finish during a saccade. Moreover, we do occasionally notice changes. For example, it is not uncommon to detect an occasional editing error in a motion picture. These examples of successful change detection are readily available, whereas, by definition, we do not become aware of those changes that we fail to notice. Given the tendency to judge the frequency of events based on the information that is most available in recollection (Tversky and Kahneman, 1982), estimates of successful change detection are likely to be inflated. Because some changes are noticed, we have the impression that all changes draw our attention, leading to awareness. Evidence for change blindness shows that this impression is false. Unexpected changes do not necessarily draw our attention.

This mistaken intuition can be demonstrated empirically (Levin et al., 2000). In one change detection study (Levin and Simons, 1997), observers viewed a brief motion picture of a lunch conversation in which an intentional change was made every time the camera cut instantaneously to a different angle. For example, a scarf worn by one of the actors disappeared and then reappeared from one shot to the next, plates changed from red to white and then back to red, and objects on the table switched locations. When unsuspecting observers were asked to view this film carefully, they typically did not report any of the changes (Levin and Simons, 1997). That is, they showed complete change blindness for these unexpected changes. To demonstrate the overestimation of change detection abilities, a different group of observers viewed still frames from the film and predicted how likely they would be to detect the various changes (Levin et al., 2000). These observers predicted far greater change detection for irrelevant changes than actually occurred: for example, approximately 90% of observers thought that they would notice the scarf change, and more than 75% thought they would notice the plate color change when, in reality, no subjects noticed either of these changes.

This massive overestimation of change detection rates reflects the intuitive assumption that important, unusual, or distinctive events will automatically draw attention. This assumption can have important implications, both for theoretical arguments about change detection and for practical applications of evidence for change blindness. Just as observers believe they will detect changes, product designers believe that the subtle change cues they provide will be sufficient to guide attention. One obvious situation in which change blindness can have dire consequences is driving. Brake lights are typically red, and when they come on, they simply brighten. The assumption that the brightening of brake lights will automatically draw attention even if we are focusing attention on some other aspect of the scene (or on other distracting events in the car such as a phone conversation) can have dire consequences. The change from dim to bright is an instantaneous change, and if the change itself does not attract attention, it sometimes will be missed. A more reasonable strategy for ensuring that motorists notice the onset of brake lights would be to have the change occur repeatedly, thereby increasing the probability that observers will be attending to the lights at some point when the change occurs. If drivers are aware of their own poor change detection abilities and the likelihood that other drivers will have mistaken intuitions about change detection of other drivers, then they can adjust. For example, if another car is following too closely and you need to brake, you can lightly pump the brake pedal to cause the brake light to flash repeatedly, thereby increasing the chances that the driver behind you will notice that you are stopping.

Studies of the mistaken belief in the absence of change blindness, something Levin et al. (2000) dubbed change blindness blindness, help reveal the potential consequences of change blindness. They also demonstrate that change blindness findings are surprising: not only are we blind to large changes in our visual world, but we remain largely unaware of this fact. This meta-cognitive failure might well have more real world consequences than the phenomenon of change blindness itself.