67

Abstract

Classic change blindness is the phenomenon where seemingly obvious changes that coincide with visual disruptions (such as blinks or brief blanks) go unnoticed by an attentive observer. Some early work into the causes of classic change blindness suggested that any pre-change stimulus representation is overwritten by a representation of the altered post-change stimulus, preventing change detection. However, recent work revealed that, even when observers do maintain memory representations of both the pre- and post-change stimulus states, they can still miss the change, suggesting that change blindness can also arise from a failure to compare the stored representations. Here, we studied slow change blindness, a related phenomenon that occurs even in the absence of visual disruptions when the change occurs sufficiently slowly, to determine whether it could be explained by conclusions from classic change blindness. Across three different slow change blindness experiments we found that observers who consistently failed to notice the change had access to at least two memory representations of the changing display. One representation was precise but short lived: a detailed representation of the more recent stimulus states, but fragile. The other representation lasted longer but was fairly general: stable but too coarse to differentiate the various stages of the change. These findings suggest that, although multiple representations are formed, the failure to compare hypotheses might not explain slow change blindness; even if a comparison were made, the representations would be too sparse (longer term stores) or too fragile (short-lived stores) for such comparison to inform about the change.

Introduction

Change blindness is a robust phenomenon characterized by observers’ failure to notice seemingly obvious changes in their visual input. In most cases of change blindness in the literature, such unnoticed changes coincide with other visual transients. For example, observers may miss the disappearance or replacement of a scene element that co-occurs with the simultaneous appearance of small markings elsewhere in the scene (so-called “mud-splashes”) (O’Regan, Rensink, & Clark, 1999) or with a brief blanking of the whole scene, either due to a blank frame inserted between frames (interstimulus interval) (Rensink, O’Regan, & Clark, 1997) or due to a blink or eye movement (Grimes, 1996; McConkie & Currie, 1996; O’Regan, Deubel, Clark, & Rensink, 2000). 

The change blindness phenomenon raises questions about the brain’s representation of current and prior visual information. A comparison between representations of the pre-change stimulus state and the post-change stimulus state should readily alert observers to the change, even if they miss seeing the change occur; however, the nature of such representations is heavily debated. Early interpretations of change blindness centered on the idea that only a small fraction of the visual world is stored beyond the present moment (O’Regan, 1992), and the resulting lack of stored information would explain why even substantial changes may go unnoticed. The disruption of attentional mechanisms during classic change blindness suggests that a lack of attention to pre-change information can explain the failure to notice changes (Awh, Vogel, & Oh, 2006; Caplovitz, Fendrich, & Hughes, 2008; Chun & Turk-Browne, 2007). Several other interpretations do allow for the initial formation of a memory representation of the pre-change scene and propose that change blindness occurs when this representation fades or is overwritten by a representation of the new, post-change scene (Beck & Levin, 2003; Irwin, 1992; Noë, Pessoa, & Thompson, 2000; O’Regan & Noë, 2001; Rensink et al., 1997). More recent work has provided evidence that, in fact, both pre-change and post-change representations can exist simultaneously even when the observer does not notice the change (Beck & Levin, 2003; Hollingworth & Henderson, 2002; Mitroff, Simons, & Levin, 2004; Simons, Chabris, & Schnur, 2002), leading to the suggestion that the inability to notice the change may reflect a failure to compare representations rather than their absence (Mitroff et al., 2004; Smith, Lamont, & Henderson, 2012). Relatedly, there is evidence that, during change blindness, a changing item may still be registered implicitly, as indicated by an influence on subsequent behavior such as task accuracy, pupil size, or eye-movement patterns (Chetverikov, Kuvaldina, MacInnes, Jóhannesson, & Kristjánsson, 2018; Fernandez-Duque & Thornton, 2003). This is consistent with the possibility that representations of the changing item may exist, even if a change in the item remains unnoticed. 

Most knowledge on change blindness and, indeed, all evidence summarized above come from studies on so-called classic change blindness: blindness to a sudden visual change that coincides with other visual events. But, observers have also demonstrated blindness to changes that occur without distracting events, provided that the changes unfold too slowly for a visual transient to capture attention (David, Laloyaux, Devue, & Cleeremans, 2006; Frey, Koenig, He, & Brascamp, 2024; Hollingworth & Henderson, 2004; Laloyaux, Devue, Doyen, David, & Cleeremans, 2008; Simons, Franconeri, & Reimer, 2000). In these studies, participants are presented with a stimulus in which a part of a visual scene changes, appears, or disappears slowly over the span of many seconds, and they are asked to report changes that they notice. Crucially, a large proportion of participants do not report the gradual change, thus demonstrating slow change blindness. These studies illustrate how robust the phenomenon is but do not attempt to quantify the visual representations involved. What is especially intriguing about such slow change blindness is that the stimuli are uninterrupted and observers have plenty of time to look for changes. As such, one would intuitively expect observers to be aware at some point that something has changed, even if they did not witness the change as it occurred. The fact that they do not underscores that our awareness may be sparser than our intuitions would suggest. Slow change blindness may also be particularly informative when it comes to furthering the understanding of visual processing in natural viewing conditions, which often involves movement of objects as well as gradual changes in, for example, viewpoint or lighting. 

Very little literature on slow change blindness exists, and it is clear that some findings from work on classic change blindness cannot be directly applied here. The finding that observers maintain both a pre-change and a post-change representation side by side may not generalize to a situation where the change between the initial and final state occurs gradually via numerous intermediate steps and over a much longer period. One study shows that observers do not notice a gradual rotation in the viewpoint of a scene but do notice a reversion back to the initial view, evidence that the observer’s representation of the scene is continually updated as slightly different visual information keeps reaching the senses, despite the observer’s failure to notice that anything is changing (i.e., implicit updating) (Hollingworth & Henderson, 2004). Other work, however, indicates that what is represented by the end of an unnoticed slow change is not simply the most recent state of the scene but rather incorporates information from both recent and earlier steps in the changing sequence (Laloyaux et al., 2008). 

To investigate what sort of representations an observer may form and retain during slow change blindness, we presented observers with images in which a large, centrally located area slowly changed color. Even though these changes are obvious when they happen quickly, when they unfold over many seconds (16 in our case) they are rarely noticed (Frey et al., 2024). These stimuli are particularly interesting for studies of slow change blindness because they elicit change blindness even when the changing scene element covers a large part of the visual field (including the center of fixation). To gain insight into the observer’s representation of the changing region, we probed this representation as soon as the change had finished and the image had been removed from the screen. We did so by cueing attention to the relevant region of the scene and then showing a comparison image that could match the slowly changing region as it was at a particular timepoint during the slow change. Across the experiments that we performed, this timepoint could be the initial timepoint, the final one, or a timepoint midway in the change. For comparison, our later experiments also included conditions where the comparison image did not, in fact, match the color-changing region at any timepoint, as well as ones where the region did not actually change color. Observers reported whether the comparison image matched the cued region as it had been immediately before it disappeared. Although we asked about the final state of the region before its disappearance, the observer’s comparison judgment was indicative of their internal representation of the region, which could in principle be influenced by this final state as well as by earlier stimulus states. Mindful of the fact that the mind harbors multiple memory systems, we performed three variants of the experiment, each using a slightly different cueing procedure, aimed at probing different memory systems. We hypothesized that different memory systems may be influenced by the changing stimulus information to different extents and that reports of “same” or “different” would be influenced by the memory system probed. Overall, our results are consistent with the idea that observers have access to multiple memory representations during change blindness and that some of those representations get overwritten by updated ones as the change happens. Other representations may be more stable over time, yet insufficiently detailed to alert observers that a change has happened. 

Method

Participants

In all experiments, participants were recruited using Prolific (www.prolific.co) (Palan & Schitter, 2018), and the experiments were made available online using Pavlovia (https://pavlovia.org) (Peirce et al., 2019). Participants were paid approximately $12/hour for their participation (the rate of pay for each experiment differed slightly and was based on the payment recommendations of Prolific at the time of the study). The study was approved by the Michigan State University institutional review board, and all participants provided informed consent through an online form via Qualtrics (www.qualtrics.com). Prolific users who self-reported normal or corrected-to-normal vision and hearing, had English fluency, were between 18 and 65 years of age, had more than 10 previous submissions on Prolific, had a Prolific approval rating above 95, and had not participated in previous iterations of our study on Prolific were eligible to participate in the study. Otherwise, no restrictions were imposed on participation. Participants were instructed to complete the experiment in a single sitting using a desktop or laptop computer. There was no direct control of participant environment and behavior because the study was administered online, but participants were instructed to sit at their normal viewing distance and to avoid large movements during the study. Before the study began, participants completed a blind-spot identification procedure and a bank card scaling procedure (Brascamp, 2021) so that we could estimate viewing distance, as well as the size and aspect ratio of the pixels. Using this information, we displayed stimuli so that they would appear square and subtend approximately 20 degrees of visual angle from the participant’s viewpoint. 

General procedure

We performed a series of three experiments, each building on the previous, to investigate the memory representations involved in slow change blindness. In each experiment, each participant completed a single critical trial where they judged whether the color of a cued area was the same or different as it was before a mask. Observers were presented with a 20-second long video in which a large, central, region underwent a slow color change over the span of the central 16 seconds. Immediately after the video finished, the scene was covered up by a mask, and the relevant screen region was cued with a single-word audio file. We then presented a comparison scene that matched a frame that had occurred at some time point during the slowly changing scene, such as the initial or final frame (Experiments 1 and 2) or an intermediate frame (Experiment 2). The observer’s task was to report whether the cued region in the comparison scene matched what they saw immediately before the image was covered. Unbeknownst to our observers, the cued area was always the large colored region that had undergone a color change during the preceding video. 

In addition to the slow color change, these videos each contained three quick changes that happened over 1 second and were relatively easy to notice. These quick changes were included to impress upon the participant the fact that the scene was not static so that it could not be assumed that whatever was seen early on would remain that way, meaning that the comparison image really needed to be compared to the state that immediately preceded the blank. These videos were generated by the present authors using a novel, semi-automatic procedure (Frey et al., 2024) and are available online for use in studies such as this. Our previous work indicates that slow change blindness is not notably affected by the presence of such quick changes. When the number of such changes could vary between one and three, this number did not influence the observer’s ability to detect the slow change (r² = 0.0266) (Frey et al., 2024). In fact, that work also showed that observers performed similarly on the slow change detection task even in the absence of any quick changes. 

Prior to the trial, participants were instructed to fixate on a black-and-white circle at the center of the image at both the beginning and the end of the 20-second presentation. However, the circle disappeared after 2 seconds and did not reappear until 18 seconds into the presentation, and participants were free to look around the image during the absence of the circle. When the circle reappeared, a tone was played via the speakers or headphones as a reminder that the participant should return their gaze to the circle at the center of the image for the remainder of the trial. Observers were told that, during the 20-second viewing period, certain features of the image might change and that their task was to report via keypress whether the covered-up area was the same or different as it was immediately before the image was covered. The image shown after the mask was removed matched the final frame in terms of which quick changes had occurred, and the large color area matched a color that had been shown at some point during the slow change sequence. In Experiment 1, this was either the initial color or the final color of that area during the video (Figure 1). In later experiments, we also included comparison frames in which the large area had either a color that was intermediate between this initial and final color or a color that was substantially different from any color shown during the slow change sequence. 

Each experiment included two conditions that differed in which type of memory stores we aimed to probe: short-lasting visual memory traces or longer term memory representations. To probe the former, we covered up the area of interest with a gray mask, which drew attention to the relevant screen location but with minimal visual disruption. Given that short-lasting visual memory traces are thought to be formed (and overwritten) any time visual input is present (Becker, Pashler, & Anstis, 2000; Landman, Spekreijse, & Lamme, 2003; Sligte, Scholte, & Lamme, 2008), minimizing visual disruption is critical here. We refer to this condition as the retrocue condition. In general, retrocues are spatial markers that are designed to cue an observer to relevant visual information after the information has been removed from the screen but before any disrupting visual information is presented (Griffin & Nobre, 2003; Landman et al., 2003; Sligte et al., 2008). In our Experiment 1, the temporal sequence was slightly different from what is described in this general definition of retrocues, because the mask acts both to remove the relevant visual information and simultaneously to cue the participant to where that information was (the location and extent of the mask indicate this). Still, we refer to this condition as the retrocue condition because the cue is introduced before any new visual information is introduced. 

To probe longer term memory traces, we covered up the area of interest with a full-color and visually rich cartoon image mask, which drew attention to the relevant screen location while also providing substantial visual disruption. The visual disruption was intended to interfere with the storage and maintenance of fleeting visual memory traces, leaving only more robust and persistent traces for the participant to base their response on. We refer to this condition as the postcue condition. In general, postcues in this context are spatial markers that are designed to cue an observer to relevant visual information, after the information has been removed from the screen and also after disrupting visual information has been presented (Astle, Summerfield, Griffin, & Nobre, 2012; Sligte et al., 2008). In our Experiment 1, the temporal sequence was again slightly different because the cartoon cover does three things simultaneously: It removes the original visual content, it cues participants to the relevant scene region, and it also interferes with visual memory traces. We still use the term “postcue” because the cue does not precede visual disruption. In Experiment 2, we directly addressed some of the considerations concerning the sequence of events (see below). 

In both conditions, the cue appeared at the same time and consisted of the mask (either blank or cartooned) with a red outline, as well as an audio file that played a one-word reference to the item in question (see below). Aside from the content of the mask (minimal visual information vs. an abundance of new visual information), the two conditions were the same. 

Because visual memory traces can be very short lived (Sperling, 1960), in each experiment we took care to extensively train participants in quickly accessing their memory upon the presentation of a cue. Accordingly, the critical trial in each experiment was preceded by 100 training trials in which participants reported whether a cued area was the same or different between two still images. These practice trials helped participants learn to expect a cue and practice accessing memory quickly based on the cue. In each training trial, participants fixated on a central point. An image appeared for 500 ms, after which a portion of the image was covered up by a mask outlined in red. In addition to the location of the mask itself, a single-word audio file played simultaneously to further help participants direct their attention to the relevant, and covered-up, item while maintaining fixation at the center dot (Figure 2). Participants were told that they would be comparing an image before and after a portion of it was covered up and that they would be asked whether the image was the same or different. They were instructed to try to remember what was present behind the covered area as soon as the cover appeared and the audio file played. We added the audio cue because the rectangles, although certainly cueing attention to the relevant section of the image, often covered up other elements of the image in addition to the element that the participant would be asked about. In such cases, the audio file could help participants pay attention to the relevant aspect of the memory trace more quickly. The nature of the cover-up and the timing of the cue were different depending on whether the trial was a postcue or a retrocue trial and which experiment the observer was participating in (details of the cueing paradigm for each experiment are described below in their respective sections). After 1300 ms, the cover was removed, and participants responded with a keypress to convey whether the cued part of the image was the same as or different from what it was before it had been covered. The image remained on the screen until the participant entered their response.