A Method for Investigating Change Blindness in Pigeons (Columba Livia)

The method presented here is inspired by the so-called "flicker paradigm" commonly used by cognitive psychologists to study change blindness in humans⁹. In this human research, change blindness is operationally defined as the impairment in change detection produced by the presence of an ISI that interrupts transitions between stimulus displays. The same is true for the pigeon implementation described here. Furthermore, humans tend to approach the flicker task using a serial search strategy, considering possible locations one at a time and at each location either successfully identifying the change or rejecting that location and moving along to another. The method presented here produces results consistent with the same interpretation. Pigeons' change detection accuracy improves with additional repetitions³², as one would expect from a serial search that progresses from one possible change location to the next. Note however, that while the primary results are congruent with a serial search process, it is not the only logical possibility. Alternate interpretations (such as limited-capacity parallel search) should be considered, and may even motivate variations on the procedure. For example, if pigeons do engage in a location-by-location search process, they should be able to take advantage of base rate differences by selectively initiating their search at high-probability locations (enhancing accuracy at those high probability locations while reducing accuracy at others). A parallel search process would not make such a prediction.

Despite these important similarities, there are also some notable differences that must be acknowledged. Whereas human research often utilizes complex visual stimuli such as photographic images, the method presented here relies on simpler elements that can be projected onto the response keys in an operant chamber. One consequence of this difference is that the number of possible change locations in a standard operant chamber is limited to three (the number of response keys), and this number is much smaller than the innumerable possible change locations in a photographic display. So while both pigeons and humans may use a serial search strategy to locate changes, accurate performance on the procedure described here could be supported by a much simpler search process. Second, human research often measures response times to assess change detection. More difficult-to-find changes (such as those obscured by an ISI) require additional search time and produce longer response times. This pigeon method uses accuracy as a stand-in for response times because it was virtually impossible to prevent pigeons from pecking the stimulus displays during presentation (presumably before they had a chance to identify any change). Counting these stimulus pecks as responses led to consistently fast but meaningless response times and unacceptably low change detection accuracy. Nevertheless, the logic for interpreting accuracy as a measure of performance is similar to that for interpreting response times. Easy-to-find changes can be identified in a short time (producing good accuracy after even a small number of repetitions), whereas more difficult changes take longer (only reaching high levels of accuracy after many repetitions). Because the method requires that all repetitions be presented in their entirety before a response is recorded, there is no way to pinpoint precisely when, during stimulus presentation, a pigeon identified the change. Presumably some changes are identified well before the recorded response whereas others are identified only just before. This undermines the usefulness of any resulting response times, but not the usefulness of analyses based on accuracy.

There are several practical matters to consider when implementing this task. While the method has shown itself to be replicable, some scenarios have produced no change blindness effect despite good overall accuracy (see Figure 4, top panel). Thus, stimulus details should be chosen carefully to maximize the likelihood of producing the hypothesized effects. In particular, it appears that very short ISI durations may weaken or even eliminate change blindness³⁰. The recommended 250 ms ISI specified in the protocol should be effective, but note that shorter durations can be used as well; a change blindness effect has been confirmed with an ISI as short as 7 ms³². Note also that the difficulty of ISI trials increases with the duration of the ISI. Thus, if animals struggle to learn the procedure, shortening the duration of the ISI may be an easy way to increase accuracy, especially early in training. Conversely, lengthening the ISI (even beyond the recommended 250 ms) may be a way to increase the effect size. One of the most powerful factors may be the salience of a change, and this is true for both humans³⁶ and pigeons^29,31. Larger or more prominent changes are easier to detect, but the effect is larger on the more difficult ISI trials than on no-ISI trials. Finally, one should consider the number of stimuli that can be generated given the available equipment. With large numbers of change detection stimuli, pigeons will likely see novel, never-before-seen stimuli in every session, and so a general change detection rule is the only strategy that can produce consistently good accuracy. On the other hand, with smaller numbers of stimuli available, memorization becomes a feasible approach, as the same stimuli may be presented repeatedly across and even within sessions. In such cases, it is especially important to reserve stimuli for transfer, in order to demonstrate that the results are due to rule learning and not memorization²⁹. Note that even a small number of available stimulus elements can combine to produce a relatively large set of possible displays. For example, Herbranson and Jeffers²⁹ (Experiment 1), used just four different color elements, yielding 4³ = 64 unique original displays. Each original display could then be paired with one of nine unique modified displays, yielding a total training set of 576 paired change displays. With more elements come even more possible displays. Herbranson and colleagues³² used combinations of eight-line orientation elements on each of three keys to produce a set of over 400 million unique paired displays.

To conclude, the flicker paradigm is exciting because it produces a robust effect that highlights a striking failure of selective attention. The protocol presented here can be implemented using readily available operant chambers to produce results in a nonhuman animal that parallel those previously demonstrated in humans. Given that operant equipment is widely available and not specifically limited to pigeons, the method may prove to be just as useful for the myriad other species of animal that can exert control over attentional processes. While research on attention in animals is not new³⁷, much of it has (quite understandably) focused on the successful or adaptive aspects of attention. Change blindness highlights a consistent and systematic failure of attention that deserves the same level of scrutiny. Thus, change blindness is more than a fascinating quirk of human visual attention. It implies for example, that much of our visual experience relies on representations that are less detailed than we might expect¹⁰. The existence of a parallel effect in pigeons further suggests that this sparseness of representation may be evolutionarily quite old and widespread. Change blindness thus may be a general consequence of selective attention that can provide some valuable insights into the origins of attention and the reasons for its limitations and imperfections.