Behavioral Ecology Vol. 12 No. 2: 192-199
© 2001 International Society for Behavioral Ecology
Limited attention: the constraint underlying search image
Nebraska Behavioral Biology Group, School of Biological Sciences, University of Nebraska, Lincoln, NE 68588-0118, USA
Address correspondence to R. Dukas, Behavioral Ecology Research Group, Department of Biological Sciences, Simon Fraser University, Burnaby B.C. V5A 1S6, Canada. E-mail: rdukas{at}sfu.ca .
Received 16 February 2000; revised 15 June 2000; accepted 16 August 2000.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSDISCUSSIONREFERENCES |
|---|
Recent models of predator search behavior integrate proximate neurobiological constraints with ultimate economic considerations. These models are based on two assumptions, which we have critically examined in experiments with blue jays searching for artificial prey images presented on a computer monitor. We found, first, that when jays had to switch between searching for two distinct prey types, they showed no reduction in detection rates compared to no-switching to no-switching conditions, and second, that when jays divided attention between searching for two prey types at the same time, they had lower detection rates than when they focused attention on one prey type at a time. Our results suggest that limited attention strongly affects predator search patterns and diet choice, including the ubiquitous tendency to form search images.
Key words: attention, blue jay, constraint, foraging, predator, prey, search image, switching.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSDISCUSSIONREFERENCES |
|---|
The foraging behavior of predators searching for cryptic prey can strongly affect the fitness of both the predators and the prey, and consequently, the dynamics of populations and communities of animals and plants (e.g., Bond and Kamil, 1998
;
Hassell, 1978;
Martin, 1988;
Sutherland, 1996); hence
ecologists must acquire a realistic understanding of predators' search
patterns and their dynamics. In a series of papers, Dukas and colleagues
(Dukas and Clark, 1995a,
1995b;
Dukas and Ellner, 1993)
attempted to integrate mechanistic neurobiological information with foraging
theory in order to explain and predict how constraints on information
processing together with economic considerations shape predators' foraging
patterns. These models included two assumptions. First, the brain can
effectively process only a limited amount of information at any given time.
Substantial evidence indicating a very limited rate of information processing
by the brain comes from neurobiological studies in humans and monkeys, which
have employed either electrophysiological recording from individual neurons or
brain imaging. These studies have revealed that focusing attention on a given
task is associated with increased activation of the neurons processing this
task, suppressed activity of other neurons, and enhanced behavioral
performance on the attended task (e.g.,
Behrmann and Haimson, 1999;
Corbetta et al., 1990;
Desimone, 1998;
Kastner et al., 1998;
Spitzer et al., 1988).
The other assumption in the foraging models that include neurobiological
constraints is that alternation among distinct tasks involves a period of
reduced efficiency immediately following a switch. Some proximate reasons for
such interference between distinct tasks are: (1) the mere passage of time
devoted to one task is accompanied with the decay of memory of the other task,
(2) the newly acquired information interferes with information already in
memory, or (3) the old information already in memory interferes with the
acquisition of new information. Extensive evidence for interference exists
from human and animal studies conducted in the laboratory and in the field
(Anderson, 1990,
1995;
Duncan et al., 1994;
Spear and Riccio, 1994;
Stanton, 1983;
Wickens, 2000;
Woodward and Laverty,
1992).
Optimality analyses including one
(Dukas and Ellner, 1993) or
two (Dukas and Clark, 1995a)
of the neurobiological constraints just mentioned predict that animals should
search only for a single cryptic food type at any given time while ignoring
alternatives that are equally cryptic, rewarding and abundant, a behavioral
pattern sometimes referred to as "search image"
(Dukas, 1998;
Tinbergen, 1960). This
prediction is attractive because it allows integration within foraging theory
of the somewhat separate research on perceptual biases affecting predator
search. In the four decades since Lucas Tinbergen
(1960) coined the term
"search image," researchers have focused mostly on establishing
whether search images do indeed exist (Dawkins,
1971a,b;
Pietrewicz and Kamil, 1979;
Reid and Shettleworth, 1992).
Recent studies have indicated that selective attention is involved in search
images (Blough, 1991;
Langley, 1996). However, these
studies, which aimed to establish the existence of search images, did not
critically test for either the effect of limited attention or switching on
predator searching behavior.
Here we present results of experiments designed to evaluate the costs of switching and divided attention using blue jays (Cyanocitta cristata) foraging for digital images of cryptic prey. We began with a quantification of the cost of switching, because this could provide us with the necessary knowledge for designing a well-controlled experiment measuring the cost of divided attention. In the first experiment, we predicted that because switching between distinct search tasks can decrease performance, increasing the rate of alternation between the tasks would decrease the overall rate of target detection. In the second experiment, we repeated the test for the effect of switching, this time comparing the extremes of no switching at all to switching at an average rate of one alternation every other trial. The second experiment also included another set of sessions, in which we compared target detection rate when jays had to either focus attention on searching for a single target type per display or divide attention between searching for two distinct target types per display. In that set of sessions, we predicted that dividing attention would decrease the frequency of target detection.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSDISCUSSIONREFERENCES |
|---|
Subjects
The eight blue jays (Cyanocitta cristata) used in the experiments were captured as nestlings in Lincoln, Nebraska, USA, approximately a year before the experiments and were hand raised in the laboratory. During the experiment, the jays were maintained at 80% of their ad libitum body weight with controlled daily feedings of turkey starter and Lefeber brand food pellets. The jays were housed in individual cages, with water available, at a constant room temperature of 27°C and on a 14:10 h light:dark cycle.
Apparatus
We trained and tested the jays in an operant chamber (approximately 50
x 50 x 50 cm) with opaque walls located in a small, darkened room.
A white noise generator was played throughout the experiment to mask outside
sounds. Stimuli were presented on a computer monitor embedded in the front
wall of the chamber. A lamp mounted above the monitor provided dim light
throughout the experiment. We attached a clear Plexiglas sheet to the front of
the monitor by springs to prevent damage to the monitor and to the jays'
beaks. An infra-red "touch screen" reported the location of each
peck directed at the screen. A wooden perch was positioned approximately 10 cm
from the touch screen and 15 cm above the chamber floor. Jays standing on the
perch could readily peck at targets presented on the monitor screen and reach
the food rewards, which were half pieces of mealworms (Tenebrio
molitor), delivered via a Davis UF-100 universal feeder into a food cup
mounted to the left of the lower left corner of the monitor. At the moment of
reward delivery, a light above the food cup was turned on and fully
illuminated the food cup for 3 s. All stimulus presentations, schedules of
reward delivery and data recording were controlled by a personal computer
programmed in Borland C.
Training
Prior to each experiment, we trained the jays to peck at targets presented
among non-target background items on the computer monitor. In the final stage
of training, we gradually increased the density of background items in order
to create the desired highly cryptic conditions. In experiment 2, we also
altered the background items to make them more similar to the target. During
the final stage of training, we adjusted task difficulty individually for each
jay in order to maintain the percentage of correct detection by each jay at
approximately 65%. By the end of the training period for each experiment, the
jays were familiar with the experimental protocol and were able to detect what
we perceived as highly cryptic targets.
Experiment 1: the cost of switching
Methods
In this experiment we wished to quantify the cost of switching between two
distinct search tasks by manipulating the frequency of switching events per
session. Specifically, we expected switching to diminish searching performance
and thus predicted that increasing the frequency of switches per session would
result in reduced target detection rate.
We created two artificial prey images and backgrounds. Because bird color
vision is different from that of humans (e.g.,
Jacobs, 1981), we used only
monochromatic images. The images were an artificial caterpillar 15 pixels long
and five pixels wide on a background of a random assortment of segments, and
an artificial moth 20 pixels in maximum length and 17 pixels in maximum width
on a background consisting of various components of the moth's body parts.
Although the visual dimensions of the targets were slightly different, both
were treated by the computer as being at the center of a 20 x 20 pixel
virtual display corresponding to approximately 8 x 8 mm on the screen.
(Faithful reproductions of the images and backgrounds used are technically
unattainable in print format; hence the authors will provide electronic
examples of the displays on request).
A 100-trial session consisted of 50 caterpillar displays and 50 moth displays. Half the displays of each target contained the target at a randomly chosen location and numerous background items, and the other half contained background items only. The jays had two distinct search tasks (searching for either a moth on its background, or for a caterpillar on its background) and we manipulated the frequency of switching between these tasks.
In nature, switching between searching for alternative prey types might occur over various temporal scales. To cover a wide range of possibilities, we compared jays' performances at three switching rates: low, moderate, and high. The low switching rate consisted of a single switch per session, occurring on a randomly chosen trial between the 45th and 55th trial. The moderate switching treatment involved 10 switches per session, with switching occurring randomly every 8 to 12 trials. Finally, the high switching treatment involved 50 switches occurring randomly every one to three trials. Our protocol simulated a natural situation of a forager alternating searching between two prey types each occurring in a distinct patch.
Each trial began with the presentation of a start signal, a red circle at the center of the blank monitor. Upon a single peck to the start signal, the display was presented. The display always contained background and a move signal consisting of a green circle at the center of the display; it also contained a prey item on 50% of the trials. A jay could do one of four things: (1) peck at the target ("correct detection"), (2) peck at the background ("false alarm"), (3) peck at the move circle ("correct rejection" on negative displays and a "miss" on positive displays), or (4) not peck at all. Correct detection resulted in the delivery of half a mealworm and 3 s feeding time followed by a 5 s intertrial interval. A peck at the background resulted in the termination of the trial without reward followed by a long intertrial interval of 30 s. A peck at the move key resulted in the termination of the trial followed by a 1 s intertrial interval. A trial that lasted for 30 s without any response ended with no reward and was followed by a long intertrial interval of 30 s.
The experiment was conducted with a randomized blocks design. Each of the three treatments was tested once in random order within each 3-day block for a total of 10 blocks. The first trial in each session consisted of a randomly chosen display type (caterpillar or moth), counterbalanced within treatments and across blocks. That is, for each treatment, 5 days began with caterpillar and the other 5 days started with moth. In subsequent trials, display type remained the same until the first switch, at which the display type was alternated; then the display type was identical until the subsequent switch (10 and 50 switches) or the session end (one switch). Each display had a 0.5 probability of containing a single prey item.
At the end of each 3-day block throughout the experiment, we monitored the average proportions of correct detection by each jay. If detection by a jay exceeded 70% for either prey type, we decreased that prey conspicuousness by increasing the background density for that jay. This way we could maintain the perceived prey conspicuousness approximately constant between jays and throughout the experiment. Note that we had to adjust difficulty in order to counteract learning by the jays and maintain the key experimental condition of a search for cryptic prey. The adjustment did not compromise our statistical tests because it was carried out between blocks. That is, experimental treatments within a block were always carried out under identical difficulty, allowing us to conduct a comparison between treatments within blocks.
We calculated the detection rate of each target type as the number of targets detected per session over the total time spent searching for that target type. The results were analyzed with a repeated measures ANOVA, which included switching frequency (three values), target type (two values) and block number (10) as fixed factors, and jay (four individuals) as a random factor.
Results
There was no significant effect of the switching frequency on target
detection rate. That is, the jays detected targets at similar rates during
sessions with 1, 10 or 50 switches (repeated measures ANOVA,
F2,6 = 1.4, p >.3,
Figure 1). The caterpillar was
detected at a higher rate (p <.006), but the interaction between
switching frequency and target type was not significant (p >.1).
The jays slightly varied in their behavior across treatments
(F6,54 = 2.7, p <.05 for the jay by treatment
interaction). A power analysis (Cohen,
1988) revealed that the power to detect moderate (20%) differences
in detection rate was close to one, as indeed indicated by the highly
significant effect of target type.
|
We examined performance in detail in the trials immediately before and after a switch. Switching had no effect on the frequency of correct pecks (p >.1), but it had a strong effect on response latency only in the single-switch treatment, with latency being the shortest just before switching, longest in the first two trials after switching, and then rapidly decreasing. The effect of switching on response latency was not significant for the whole data set (p >.1), but highly significant for a data set containing only the single switch treatment (F5,15 = 8.1, p <.001). Overall, switching in the single switch treatment resulted in increased search latencies in the two trials following the switch, adding about 6.5 s to search duration. This amounted to less than 1% of the average total search duration of approximately 715 s for the 100-trial session.
Experiment 2: the partial costs of divided attention and
switching
In Experiment 1, each target type was presented on a unique background,
which informed the jays which single target to search for. Hence the jays
could always focus attention on searching for a single target at a time. In
contrast, in Experiment 2 we wished to compare target detection rate in
sessions where jays were informed which of the two target types to search for
versus sessions where they were not informed and hence had to divide attention
between searching for either type. Because this test of the effects of divided
attention had to include target switching, we included measurement of the cost
of switching in this experiment. This allowed us to separately assess costs of
both switching and divided attention. Even though we expected no switching
costs based on our previous experiment, the measurement of switching costs in
this experiment was a necessary control.
Methods
We used two new pairs of targets and backgrounds: (1) target A was a white
vertical ellipse 24 pixels high and 8 pixels wide, with a background
consisting of numerous white ellipses shorter or narrower than the target, and
(2) target B was a brown horizontal bar 12 pixels wide and 4 pixels high, with
a background consisting of numerous brown bars shorter or narrower than the
target. Although the visual dimensions of the targets were different, the
targets were treated by the computer as being at the center of a sensitive
area of identical size. (Faithful reproductions of the images and backgrounds
used are technically unattainable in this publication format; hence the
authors will provide electronic examples of the displays on request.) In this
experiment, the "start signal" for each trial consisted of the
target(s) that could appear in the subsequent display, presented at the center
of a red circle (see below).
The experiment consisted of two subtests, one quantifying the effect of switching alone and the other measuring the combined effects of divided attention and switching. Overall, there were six session types. The features of each type are detailed below and referred to throughout using a code summarized in Table 1. One set of three session types was designed to quantify the cost of switching. Each trial of these sessions began with the presentation of the start signal, which contained the single target type that would appear in the subsequent display. The display always included a single target, which appeared on its corresponding background (Table 1). In session types A/A (target A, background A) and B/B (target B, background B), a single target type appeared in both the start signals and displays of all 50 trials. In session type A/A||B/B (target A, background A; or target B, background B) each trial consisted of either a start signal containing target A followed by a display containing target A with its matching background, or a start signal containing target B followed by a display containing target B with its matching background. Each trial type appeared equally often and in random order. During these three session types, both the start signal initiating each trial and the background informed the jays which single target to search for. Hence the jays could always focus attention on searching for this target only. The jays did not switch between targets in sessions A/A and B/B, but they often switched between targets in session A/A||B/B. Thus a comparison of the detection rates during session A/A||B/B with those during sessions A/A and B/B would reveal any cost of switching.
|
The other three session types were designed to measure the cost of divided attention, although switching costs could not be ruled out. In trials of these sessions, a single target always appeared on a mixed background consisting of numerous white ellipses shorter or narrower than target A, and numerous brown bars shorter or narrower than target B (Table 1). In session type A/ (A + B) (target A, background A + B) and session type B/ (A + B) (target B, background A + B), a single target type appeared in all the start signals and displays of all 50 trials. In session type A/ (A + B)||B/ (A + B) (target A, background A + B; or target B, background A + B) the start signal always included the two targets, and each of the two targets appeared in the display of half the trials in random order. During A/ (A + B) and B/ (A + B) sessions, the start signal initiating each trial informed the jays which target would be found in the display. Hence the jays could always focus attention on searching for this target only. In contrast, in A/ (A + B)||B/ (A + B) sessions, the start signal indicated that either target may appear in the following display. Thus the jays always had to divide attention between searching for either target. A comparison of the detection rates between A/ (A + B)||B/ (A + B) sessions and the A/ (A + B) and B/ (A + B) sessions would reveal the combined costs of divided attention and switching, and the relative contribution of divided attention could be measured by subtracting the cost of switching evaluated in sessions A/ A, B/ B, and A/ A||B/ B.
Once a day, a jay performed a single session consisting of 50 trials. Each trial began with the presentation of the "start signal" at the center of the blank screen. Pecking at the start signal prompted the presentation of the display depicting a single cryptic target at a randomly chosen location and a background. A single peck at the display terminated the trial. A peck at the target was rewarded with half a mealworm, with the following trial presented after 3 s. A peck at the background resulted in 15 s delay. Finally, when jays did not peck at all, the trial was terminated after 15 s, with the next trial presented after 1 s.
The experiment was conducted with a randomized blocks design. Each of the six treatments was tested once in random order within each 6-day block, with a total of 16 blocks. At the end of each 6-day block throughout the experiment, we monitored the average proportions of correct detection by each jay. If detection exceeded 65% for either prey type, we decreased that prey conspicuousness by increasing the similarity of the background items to the target. This way we could maintain prey conspicuousness approximately constant throughout the experiment. Note that, as in Experiment 1, this adjustment, which allowed us to maintain the required cryptic conditions in spite of jay learning, did not compromise the statistical tests, because the changes in difficulty only occurred between blocks and we tested for treatment effects within blocks. We calculated the detection rate of each target type as the number of targets detected per session over the total time spent searching in trials consisting of that target type. The results were analyzed with repeated measures ANOVA's, which, for each of the subtests, included the number of target types per session (one or two), the target type (A or B), and block number (a total of 16) as fixed factors, and jay identity (eight individuals) as a random factor. Note that we compared jay performance only within each subtest of three sessions because the background difference between the two subtests (A or B in the first three sessions, and A + B in the other three sessions) was an obvious reason for differences in performance between the two subtests.
Results
The target detection rate in session A/ A||B/ B, when
the target types were presented in random order each on their unique
background, was very similar to target detection rates in sessions A/
A and B/ B, when all of the trials of a session consisted of
a single target type on its unique background (repeated measures ANOVA,
F1,7 = 0.3, p >.5,
Figure 2a). The pattern of
similar performance between treatments was persistent across jays
(F7,105 = 1.8, p >.05 for the jay by treatment
interaction).
|
The target detection rates were much lower in session A/(A + B)||B/(A + B), when the two target types were presented in random order on a mixed background, than during sessions A/(A + B) and B/(A + B), when only a single prey type was presented on a mixed background each session type (F1,7 = 76, p <.001, Figure 2b). Overall, the jays detected targets at an average rate of 3.5 ± 0.2 (mean ± SE) per min spent searching in session A/(A + B)||B/(A + B); this was 25% lower than the average rate of target detection of 4.7 ± 0.2 in sessions A/(A + B) and B/(A + B). The pattern of lower performance in session A/(A + B)||B/(A + B) than in sessions A/(A + B) and B/(A + B) was persistent across jays (F7,105 = 1.4, p >.1 for the jay by treatment interaction).
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSDISCUSSIONREFERENCES |
|---|
Experiment 1 indicated no cost of switching. Similarly, there was no cost of switching in the sessions of Experiment 2 when the different target types were presented on the start key and had a target-specific background. However, in the other sessions of Experiment 2, target detection rates were much lower when the jays had no cues, either on the start key or in the background, that would allow them to direct their attention towards a specific target. Thus, our results agree with the prediction that dividing attention between searching for two distinct cryptic target types at any given time causes lower detection rate (Figure 2b) and we can reject the hypothesis that switching between searching for alternative targets alone reduces detection performance (Figures 1 and 2a). Although one might be tempted to identify hints of cost of switching in Experiment 1, we should emphasize that we failed to find significant effects of switching in additional detailed analyses. Moreover, there was not even a slight sign of switching cost in Experiment 2 (Figure 2a), which was the most elaborate and with a high statistical power due to the use of eight jays and 96 daily sessions. The issue of costs of switching and divided attention is highly relevant for optimal foraging decisions (Dukas and Clark, 1995a
; Dukas and Ellner,
1993). However, here we did not allow the jays to choose between
alternative foraging strategies. Rather, we forced the jays to switch or
divide attention in certain sessions but not in others because we wished to
measure costs of switching and divided attention.
Our conclusion about the detection cost of divided attention is based on
the assumption that in A/(A + B)||B/(A +
B) sessions, when either target A or B could appear and there was no
cue of any sort indicating which target might be present, the jays indeed
divided attention between simultaneously searching for the two target types.
We do not possess direct neuronal information to substantiate this assumption.
Thus it is possible that the jays did one of the following two alternatives.
First, they could always focus attention first on one type and then switch to
the other, a behavior that should have resulted in a lower detection latency
for one target type. The results do not agree with this option, as the average
trial duration was similar for the two target types in A/(A +
B)||B/(A + B) sessions: 9.1 ± 0.19 s and
8.9 ± 0.18 s for targets A and B respectively. Individual analyses
revealed that all eight jays showed a similar pattern of just slightly better
performance on target B than A. A second alternative is that the jays rapidly
alternated between searching for targets A and B; for example, they may have
searched for target A for 1 s and then focused on searching for target B for 1
s, and so on. We consider this alternative to be close to true simultaneous
division of attention, because an analysis with a temporal resolution of 2 s
would not distinguish between the two. Moreover, Duncan et al.
(1994) found that, in human
subjects, it takes about half a second to effectively switch attention between
distinct visual tasks. This suggests that rapid switching of attention (at
least in humans) is inefficient due to time loss.
Kono et al. (1998), in a
study that focused on search image more than selective attention, also tested
the effects of background signaling on detection. They used two moth species
(Catocala relicta and C. ilia) which rest on different tree
species and were always shown on their species-typical substrate. When the
display presented to blue jays only contained a single type of tree, and
therefore could only contain a specific moth, performance was no better than
when both tree types were shown. There are a number of methodological
differences between Kono et al.
(1998) and this study, which
might account for the difference in results. One difference particularly worth
further study is that Kono et al.
(1998) always used the same
cue as the start signal. It is possible that most of the effects of divided
attention that we observed in this study were due to the lack of information
on the start signal during A/(A + B)||B/(A +
B) sessions compared to the presence of explicit information allowing
focused attention in sessions A/(A + B) and B/(A +
B).
In sum, the jay's reduced performance in A/(A + B)||B/(A + B) sessions is best explained by limited attention. Because our inference about limited attention was based on behavioral information, we briefly discuss below relevant neurobiological data.
The neurobiology of limited attention
The neurobiology of attention has been studied most directly through
electrophysiological monitoring of individual neurons in monkeys
(Moran and Desimone, 1985;
Spitzer et al., 1988). More
recently, two types of brain imaging, positron emission tomography (PET), and
functional magnetic resonance imaging (fMRI), have been widely used to monitor
attention in large populations of neurons
(Corbetta et al., 1990;
Drevets et al., 1995;
Heinze et al., 1994). For
example, Corbetta et al.
(1990) instructed human
subjects to report whether the moving bars in two successive briefly presented
computer displays were identical. In the focused attention treatment, the
subjects had been told what visual attribute may differ between the bars on
each display (color, shape, or velocity). In the divided attention treatment,
the subjects had only been told that the bars could differ in one of the three
attributes. Through the employment of PET, Corbetta et al.
(1990) found that focusing
attention on a single attribute was correlated with increased neuronal
activation in the area of the visual cortex that processes this attribute
compared to the divided attention treatment. Correspondingly, subjects'
performance on the discrimination task was higher in the focused than divided
attention treatment. In short, hundreds of studies using numerous protocols
have all identified neuronal correlates of limited attention, and research is
now focused on the mechanisms underlying the selective allocation of attention
to the most relevant information (Behrmann
and Haimson, 1999; Desimone,
1998; Desimone and Duncan,
1995; Hillyard et al.,
1998). The neurobiological research, however, implicitly assumes
that attention is an efficient mechanism allowing animals to focus only on
relevant information. This notion overlooks the ecological reality, that
limited attention can be costly, an issue we discuss below.
Limited attention: the constraint underlying search image
Limited attention implies that the amount of information foragers process
at any given time can strongly affect their feeding success. Indeed we have
documented here that jays detected targets at a lower rate when they were
forced to divide attention between searching for two target types than when
they could focus attention on a single target
(Figure 2b). Hence from a
cognitive perspective, it is advantageous to focus attention on a single
difficult task at any given time. However, in an environment where a few types
of visually distinct food types of identical conspicuousness, quality, and
density are randomly distributed, focusing attention on searching for only a
single food type means that the effective density of food is reduced, because
the other food types are overlooked. Thus the benefit from selective attention
must be sufficiently large to compensate for the effective decrease in food
density. A formal model of this foraging problem, which includes a parameter
for limited attention (Dukas and Ellner,
1993), indicates that when food is highly cryptic, focusing
attention on one type while ignoring others is indeed optimal. Hence this
model offers a likely explanation of foragers' tendencies to use search images
when searching for cryptic food. Our results are in agreement with the model
because we have documented a cost of dividing attention during search for
highly cryptic targets. In other words, limited attention is the only
identified neurobiological mechanism that has been shown theoretically
(Dukas and Ellner, 1993) and
empirically (Experiment 2) to explain observational and empirical studies on
search images.
Several previous studies on search image discussed the possible role of
selective attention (e.g., Bond,
1983; Dawkins,
1971b; Langley et al.,
1996; Reid and Shettleworth,
1992). Most notably, Blough
(1989) compared pigeon
performance in sessions in which subjects were either informed or uninformed
which of two targets would appear among non-targets in a computer display. The
information about target identity was provided either with a visual cue
preceding each trial, or by presenting one target type in all trials within a
single session. Either type of information resulted in a slight (5-10%)
increase in response latencies compared to sessions where no information was
provided. The targets used, however, were rather conspicuous, as indicated by
the approximately 85% correct detections and 1 s response latencies. The
sessions in which the pigeons were either informed or uninformed prior to the
trial which target to search for were similar to our sessions with mixed
backgrounds in Experiment 2, in which we documented large costs of divided
attention (Figure 2b) while
ruling out the possible effect of switching costs
(Figure 2a). This strongly
suggests that with the use of highly cryptic targets, our results will be
replicated with pigeons as well as other species.
While we believe that limited attention is the central feature explaining
search image, other factors, especially learning, are likely to be involved in
various field settings. In many cases, items of the same food type are
clustered in time and space. Indeed foragers possess behavioral mechanisms,
such as reducing movement distance and increasing turning angle, which help
them exploit patchily distributed food
(Dukas and Real, 1993a;
Hassell, 1978;
Price and Reichman, 1987).
Hence in a realistic natural setting, where a forager has only limited
knowledge of the available cryptic food items, learning about one type, which
appears to be common at a certain time and place, may be followed with a
period of focused search for this type.
Is switching costly?
Our prior expectation of significant costs of switching between tasks was
mostly based on human studies, which indicate that the learning of one item of
information may interfere with the later recall of another item learned
previously or subsequently (reviewed in Anderson,
1990,
1995;
Baddeley, 1986). Our
experiments, however, addressed alternation between well-learned tasks, which
had been known for weeks or months prior to testing. It is possible that a
well-practiced switching assignment would have little or no cost. Although one
might argue that well practiced switching would incur costs in tasks that are
sufficiently difficult, the fact is that the level of difficulty we employed
was sufficient to reveal large costs of divided attention, but not switching
(Figure 2). Unfortunately, we
cannot provide a meaningful comparison of the difficulty of search tasks in
our experiments and in the field.
Other behavioral consequences of limited attention
Limited attention can have other effects on behavior besides the issue of
searching for cryptic food addressed here. First, in addition to selectively
attending to relevant items or visual attributes such as certain colors and
shapes, animals may also modify the area they attend to at any given time as a
function of the difficulty of a given search task: the attentional scope can
be wide for easy search tasks but narrow for difficult tasks
(Desimone and Duncan, 1995;
Dukas, 1998;
Eriksen and Yen, 1985;
LaBerge, 1983). Narrowing the
focus of attention implies that a smaller area is searched per unit time, that
is, the search rate is reduced. Thus limited attention also provides a
neurobiologically based explanation for the observation that animals reduce
search rate when the difficulty of a search task is increased
(Gendron, 1986;
Gendron and Staddon, 1983).
The ability to modify the spatial focus of attention, however, does not
eliminate the need to selectively attend to a certain target or visual
attribute when the search task is difficult
(Dukas and Ellner, 1993). That
is, the suggestion that alteration of search rate can explain observations on
search image (Guilford and Dawkins,
1987) is in disagreement with neurobiological data, behavioral
observations, and theory (Behrmann and
Haimson, 1999; Bond,
1983; Dukas and Ellner,
1993; Moran and Desimone,
1985; Plaisted and Mackintosh,
1995; Reid and Shettleworth,
1992).
Second, another effect of limited attention is that when animals focus
attention on a difficult foraging task, they may be less likely to notice
approaching predators (Godin and Smith,
1988; Krause and Godin,
1996; Milinski,
1984; Milinski and Heller,
1978). This prediction was critically tested by Dukas and Kamil
(2000) who found that when
blue jays were engaged in an easy central search task they were three times
more likely to detect briefly presented peripheral targets than when engaged
in a difficult central task. This suggests that limited attention could be a
major cause of mortality in nature.
Third, many flower visitors tend to restrict visits to the flowers of one
species while bypassing equally rewarding alternatives
(Waser, 1986). This behavior,
which appears similar to search image, may be caused, at least in part, by
limited attention (Chittka et al.,
1999; Dukas and Real,
1993b). Finally, limited attention could explain why many insect
herbivores have a restricted diet even when they are not limited by deterring
secondary compounds. One explanation for this diet specialization is that the
insect herbivores can make faster and better feeding decisions when they focus
attention on foraging for a single plant species only
(Bernays and Wcislo, 1994).
Recent experiments indeed indicate that specialist insects forage more
efficiently than closely related generalists
(Bernays and Funk, 1999;
Jans and Nylin, 1997), and
that generalist species forage more efficiently when facing items of one type
rather than a few food types (Bernays,
1998,
1999). These studies clearly
indicate a role of some cognitive limitation in favoring specialization, but
the exact mechanisms are yet to be critically examined. In sum, limited
attention appears to have a few ecologically important effects on animal
behavior. This conclusion warrants further analyses of the factors underlying
limited attention.
Why is attention limited?
Neurobiologists focusing on mechanisms take as a given that the brain can
effectively process only a limited amount of information at any given time. In
contrast, ecologists addressing questions on the adaptive value of a trait
cannot accept a constraint as a given, unless it is directly derived from some
fundamental laws of physics or chemistry. Hence it is relevant to ask
"why is attention limited?" At the neurobiological level, it is
established that the recognition of visual patterns is computationally
demanding and cannot be performed at the maximal possible resolution across
the entire visual field (Maunsell,
1995; Van Essen et al.,
1992). That is, the brain does not possess the computational power
to process all the information provided by the sensory organs at any given
time. The underlying reasons for this limited power have not been elucidated.
The cause may simply be a limited number of neurons, but it may involve less
studied issues concerning some limitation on the integration of signals from
numerous neurons. From a functional perspective, we may assume that
attentional capacity is at some optimal level determined by certain costs and
benefits (see Dukas, 1999),
although various fundamental constraints may be important as well. For
example, one may argue that attentional capacity reflects a trade-off between
the marginal cost of maintaining additional neuronal tissue and the marginal
benefit from the added attentional span. An illuminating approach would be to
search, based on relevant ecological knowledge, for species differences in
attentional capacities.
In sum, by integrating proximate neurobiological knowledge with foraging
theory, we believe we have provided a realistic account of search image: a
prey model that includes a limited attention parameter predicts that predators
searching for cryptic prey should focus on a single prey at a time while
ignoring equally cryptic, rewarding and abundant alternatives
(Dukas and Ellner, 1993). Here
and in a companion article (Dukas and
Kamil, 2000), we have shown that limited attention is indeed a
relevant trait that can explain search image and other ecologically important
behaviors.
|
ACKNOWLEDGEMENTS |
|---|
We thank A. Bond, C. Smith, C. Cink, N. Ternus, M. Belik, and B. Gibson for comments and various kinds of assistance, and L. Bernays, C. Clark, and D. Westneat for comments on the manuscript. Our research was supported by NIH grant MH57282-01.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSDISCUSSIONREFERENCES |
|---|
Anderson JR, 1990. Cognitive psychology and its implications. New York: Freeman.
Anderson JR, 1995. Learning and memory. New York: Wiley.
Baddeley A, 1986. Working memory. Oxford: Oxford University Press.
Behrmann M, Haimson C, 1999. The cognitive neuroscience of visual attention. Curr Opin Neurobiol 9: 158-163.[Web of Science][Medline]
Bernays EA, 1998. The value of being a resource specialist: behavioral support for a neural hypothesis. Am Nat 151: 451-464.[Web of Science][Medline]
Bernays EA, 1999. When host choice is a problem for a generalist herbivore: experiments with the whitefly, Bemisia tabaci. Ecol Entomol 24: 260-267.
Bernays EA, Funk DJ, 1999. Specialists make faster
decisions than generalists: experiments with aphids. Proc R Soc Lond
B 266:
151-156.
Bernays EA, Wcislo WT, 1994. Sensory capabilities, information processing, and resource specialization. Q Rev Biol 69: 187-204.
Blough P, 1989. Attentional priming and visual search in pigeons. J Exp Psychol 15: 358-365.
Blough P, 1991. Selective attention and search images in pigeons. J Exp Psychol 17: 292-298.[Web of Science]
Bond AB, 1983. Visual search and selection of natural stimuli in the pigeon: the attention threshold hypothesis. J Exp Psychol 9: 292-306.
Bond AB, Kamil AC, 1998. Apostatic selection by blue jays (Cyanocitta cristata) searching for virtual prey produces balanced polymorphism. Nature 395: 594-596.
Chittka L, Thomson JD, Waser NM, 1999. Flower constancy, insect psychology, and plant evolution. Naturwissenschaften 86: 361-377.[Web of Science]
Cohen J, 1988. Statistical power analysis for the behavioral sciences, 2nd ed. Hillsdale, New Jersey: Erlbaum.
Corbetta M, Miezin S, Dobmeyer GL, Shulman GL, Petersen SE,
1990. Attentional modulation of neural processing of shape,
color, and velocity in humans. Science
248: 1556-1559.
Dawkins M, 1971a. Perceptual changes in chicks: another look at the `search image' concept. Anim Behav 19: 566-574.
Dawkins M, 1971b. Shifts of `attention' in chicks during feeding. Anim Behav 19: 575-582.
Desimone R, 1998. Visual attention mediated by biased
competition in extrastriate visual cortex. Philos Trans R Soc Lond
B. 353:
1245-1255.
Desimone R, Duncan J, 1995. Neural mechanisms of selective attention. Annu Rev Neurosci 18: 193-222.[Web of Science][Medline]
Drevets WC, Harold B, Videen TO, Snyder AZ, Simpson JR, Raichie ME, 1995. Blood flow changes in human somatosensory cortex during anticipated stimulation. Nature 373: 249-252.[Medline]
Dukas R, 1998. Constraints on information processing and their effects on behavior. In: Cognitive ecology (Dukas R, ed). Chicago: University of Chicago Press; 89-127.
Dukas R, 1999. Costs of memory: ideas and predictions. J Theor Biol 197: 41-50.[Web of Science][Medline]
Dukas R, Clark CW, 1995a. Searching for cryptic prey: a dynamic model. Ecology 76: 1320-1326.[Web of Science]
Dukas R, Clark CW, 1995b. Sustained vigilance and animal performance. Anim Behav 49: 1259-1267.
Dukas R, Ellner S, 1993. Information processing and prey detection. Ecology 74: 1337-1346.[Web of Science]
Dukas R, Kamil AC, 2000. The cost of limited attention
in blue jays. Behav Ecol 11:
502-506.
Dukas R, Real L, 1993a. Effects of recent experience on foraging decisions by bumble bees. Oecologia 94: 244-246.[Web of Science]
Dukas R, Real L, 1993b. Learning constraints and floral choice behaviour in bumble bees. Anim Behav 46: 637-644.
Duncan J, Ward R, Shapiro K, 1994. Direct measurement of attentional dwell time in human vision. Nature 369: 313-315.[Medline]
Eriksen CW, Yen YY, 1985. Allocation of attention in the visual field. J Exp Psychol 11: 583-597.
Gendron RP, 1986. Searching for cryptic prey: evidence for optimal search rates and the formation of search images in quail. Anim Behav 34: 898-912.
Gendron RP, Staddon JER, 1983. Searching for cryptic prey: the effects of search rate. Am Nat 121: 172-186.[Web of Science]
Godin JGJ, Smith SA, 1988. A fitness cost of foraging in the guppy. Nature 333: 69-71.
Guilford T, Dawkins MS, 1987. Search images not proven: a reappraisal of recent evidence. Anim Behav 35: 1838-1845.[Web of Science]
Hassell MP, 1978. The dynamics of arthropod predator-prey systems. Princeton, New Jersey: Princeton University Press.
Heinze HJ, Mangun GR, Burchert W, Hinrichs H, Scholz M, Munte TF, Gos A, Scherg M, Johannes S, Hundeshagen H, Gazzaniga MS, Hillyard SA, 1994. Combined spatial and temporal imaging of brain activity during visual selective attention in humans. Nature 372: 543-546.[Medline]
Hillyard SA, Vogel EK, Luck SJ, 1998. Sensory gain
control (amplification) as a mechanism of selective attention:
electrophysiological and neuroimaging evidence. Philos Trans Roy Soc
Lond B 353:
1257-1270.
Jacobs GH, 1981. Comparative color vision. New York: Academic Press.
Jans K, Nylin S, 1997. The role of female search
behaviour in determining host plant range in plant feeding insects: a test of
the information processing hypothesis. Proc R Soc Lond B
264: 701-707.
Kastner S, De Weerd P, Desimone R, Ungerleider LG,
1998. Mechanisms of directed attention in the human extrastriate
cortex as revealed by functional MRI. Science
282: 108-111.
Kono H, Reid PJ, Kamil AC, 1998. The effect of background cuing on prey detection. Anim Behav 56: 963-972.[Web of Science][Medline]
Krause J, Godin JGJ, 1996. Influence of prey foraging posture on flight behavior and predation risk: predators take advantage of unwary prey. Behav Ecol Sociobiol 7: 264-271.
LaBerge D, 1983. Spatial extent of attention to letters and words. J Exp Psychol 9: 371-379.[Web of Science]
Langley CM, 1996. Search images: selective attention to specific visual features of prey. J Exp Psychol 22: 152-163.
Langley CM, Riley DA, Bond AB, Goel N, 1996. Visual search for natural grains in pigeons (Columba livia): search images and selective attention. J Exp Psychol 22: 139-151.[Web of Science]
Martin TE, 1988. On the advantage of being different:
nest predation and the coexistence of bird species. Proc Natl Acad Sci
USA 85:
2196-2199.
Maunsell JHR, 1995. The brain's visual world:
representation of visual targets in cerebral cortex. Science
270: 764-769.
Milinski M, 1984. A predator's costs of overcoming the confusion-effect of swarming prey. Anim Behav 32: 1157-1162.
Milinski M, Heller R, 1978. Influence of a predator on the optimal foraging behaviour of sticklebacks (Gasterosteus aculeatus L.). Nature 275: 642-644.[Web of Science]
Moran J, Desimone R, 1985. Selective attention gates
visual processing in the extrastriate cortex. Science
229: 782-784.
Pietrewicz A, Kamil AC, 1979. Search image formation
in the blue jay (Cyanocitta cristata). Science
204: 1332-1333.
Plaisted KC, Mackintosh MJ, 1995. Visual search for cryptic stimuli in pigeons: implications for the search image and search rate hypotheses. Anim Behav 50: 1219-1232.
Price MV, Reichman OJ, 1987. Distribution of seeds on Sonoran Desert soils: implications for heteromyid rodent foraging. Ecology 68: 1797-1811.[Web of Science]
Reid PJ, Shettleworth SJ, 1992. Detection of cryptic prey: search image or search rate? J Exp Psychol 18: 273-286.[Web of Science]
Spear EN, Riccio DC, 1994. Memory: phenomena and principles. Boston: Allyn and Bacon.
Spitzer H, Desimone R, Moran J, 1988. Increased
attention enhances both behavioral and neuronal performance.
Science 240:
338-340.
Stanton ML, 1983. Short-term learning and the searching accuracy of egg-laying butterflies. Anim Behav 31: 33-40.
Sutherland WJ, 1996. From individual behavior to population ecology. New York: Oxford University Press.
Tinbergen L, 1960. The natural control of insects on pinewoods I. Factors influencing the intensity of predation by songbirds. Arch Neerl Zool 13: 265-343.
Van Essen DC, Anderson CH, Felleman DJ, 1992. Information processing in the primate visual system: an integrated systems perspective. Nature 255: 419-423.
Waser NM, 1986. Flower constancy: definition, cause, and measurement. Am Nat 127: 593-603.[Web of Science]
Wickens CD, 2000. Engineering psychology and human performance, 3rd ed. Upper Saddle River, New Jersey: Prentice Hall.
Woodward G, Laverty TM, 1992. Recall of flower handling skills by bumble bees: a test of Darwin's interference hypothesis. Anim Behav 44: 1045-1051.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  K. A. Jones and J.-G. J. Godin Are fast explorers slow reactors? Linking personality type and anti-predator behaviour Proc R Soc B, February 22, 2010; 277(1681): 625 - 632. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K. Nakata Attention focusing in a sit-and-wait forager: a spider controls its prey-detection ability in different web sectors by adjusting thread tension Proc R Soc B, January 7, 2010; 277(1678): 29 - 33. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. S. Hunter Familiarity breeds contempt: effects of striped skunk color, shape, and abundance on wild carnivore behavior Behav. Ecol., November 1, 2009; 20(6): 1315 - 1322. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. Wennersten and A. Forsman Does colour polymorphism enhance survival of prey populations? Proc R Soc B, June 22, 2009; 276(1665): 2187 - 2194. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
N. M. Marples, M. Quinlan, R. J. Thomas, and D. J. Kelly Deactivation of dietary wariness through experience of novel food Behav. Ecol., September 1, 2007; 18(5): 803 - 810. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
G. D. Ruxton, C. Fraser, and M. Broom An evolutionarily stable joining policy for group foragers Behav. Ecol., September 1, 2005; 16(5): 856 - 864. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
E. A. Hebets Attention-altering signal interactions in the multimodal courtship display of the wolf spider Schizocosa uetzi Behav. Ecol., January 1, 2005; 16(1): 75 - 82. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. W. Clark and R. Dukas The behavioral ecology of a cognitive constraint: limited attention Behav. Ecol., March 1, 2003; 14(2): 151 - 156. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||