Collective decision-making in honey bees: how colonles choose among nectar sources
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Behav Ecol Sociobiol(1991) 28:277-290
Behavioral Ecology
and Sociobiology
© Springer-Verlag1991
Collective decision-making in honey bees:
how colonles choose among nectar sources
Thomas D. Seeley 1, Scott Camazine 1, and James Sneyd 2.
t Section of Neurobiologyand Behavior, Mudd Hall, CornellUniversity,Ithaca, NY 14853, USA
2 Centre for MathematicalBiology,MathematicalInstitute, 24-29 St. Giles', Oxford OX1 3LB,UK
Received April 3, 1990 / Accepted September 30, 1990
Summary. A honey bee colony can skillfully choose suggest that honey bee colonies possess decentralized
among nectar sources. It will selectively exploit the most decisi0n-making because it combines effectiveness with
profitable source in an array and will rapidly shift its simplicity of communication and computation within a
foraging efforts following changes in the array. How colony.
does this colony-level ability emerge from the behavior
of individual bees ? The answer lies in understanding how
bees modulate their colony's rates of recruitment and
abandonment for nectar sources in accordance with the
profitability of each source. A forager modulates its be- Introduction
havior in relation to nectar source profitability: as prof-
itability increases, the tempo of foraging increases, the Recently, numerous authors have noted that colonies
intensity of dancing increases, and the probability of of advanced social insects constitute higher-order cogni-
abandoning the source decreases. How does a forager tive entities, that is, supraorganismal systems capable
assess the profitability of its nectar source? Bees ac- of evaluating situations and producing adaptive solu-
complish this without making comparisons among nec- tions to problems (Hofstadter 1979; Markl 1985; Seeley
tar sources. Neither do the foragers compare different and Levien 1987; Wilson and H611dobler 1988; Franks
nectar sources to determine the relative profitability of 1989). Examples of this colony-level problem-solving in-
any one source, nor do the food storers compare differ- clude a swarm of honey bees selecting the best nest site
ent nectar loads and indicate the relative profitability from a dozen or more possibilities (Lindauer 1955; See-
of each load to the foragers. Instead, each forager knows ley 1982), a colony of honeypot ants assessing the
only about its particular nectar source and independent- strength of a neighboring colony and deciding accord-
ly calculates the absolute profitability of its source. Even ingly whether or not to attack (H611dobler 1981 ; Lums-
though each of a colony's foragers operates with ex- den and H611dobler 1983), and a colony of army ants
tremely limited information about the colony's food choosing a bearing for each day's swarm raid so as to
sources, together they will generate a coherent colony- avoid the previous day's foraging zone (Franks and
level response to different food sources in which better Fletcher 1983; Deneubourg et al. 1989). All such ensem-
ones are heavily exploited and poorer ones are aban- ble cognitive performances present us with the puzzle
doned. This is shown by a computer simulation of nec- of how the problem-solving abilities of a colony arise
tar-source selection by a colony in which foragers behave from the actions and interactions of the insects compos-
as described above. Nectar-source selection by honey ing the colony.
bee colonies is a process of natural selection among alter- We address this mystery in the context of honey bee
native nectar sources as foragers from more profitable colonies choosing among patches of flowers, selectively
sources "survive" (continue visiting their source) longer exploiting the most profitable ones. The ecology of this
and °' reproduce" (recruit other foragers) better than do decision-making can be summarized as follows (reviewed
foragers from less profitable sources. Hence this colonial in Seeley 1985). Each year a colony of bees consumes
decision-making is based on decentralized control. We considerable food, approximately 20 kg of pollen and
60 kg of honey. The thousands of foragers in a colony
* P r e s e n t address." Department of Biomathematics,UCLA School laboriously gather this food from the flower patches dot-
of Medicine, 10833 La Conte Avenue, Los Angeles, CA 90024, ting the countryside around the colony's hive. Typically
USA a colony will know each day about a dozen or more
O f f p r i n t requests to : T.D. Seeley potential food sources, each with its own level of profit-278
ability, determined by such variables as the distance applied to its abdomen. Finally, the bees were gently poured into
from the hive and the abundance and quality of the a cage containing a sugar water feeder and the mother queen.
food. To gather its food efficiently, a colony must deploy This procedure was repeated several times a day for 2 days until
4000 bees were labelled.
its foragers among the flower patches in accordance with At the end of the second day, the labelled workers and their
their profitabilities. queen were transferred to a two-frame observation hive identical
We begin by documenting the skill with which a col- to the one described in Seeley (1989). The bottom frame contained
ony chooses among nectar sources so that most of its a mixture of brood (eggs and larvae), honey, and pollen, and the
nectar comes always from the richest sources. Then we top frame was empty except for a few cells of pollen. On 4 June
the colony was moved to Cranberry Lake and left undisturbed
show how this selective exploitation arises through pre-
for 4 more days, though bees were trained to the feeders to set
cise modulation of the rates of recruitment and abandon- up the experimental array and to provide food for the colony.
ment for different nectar sources in relation to their pro- On the evening of 8 June, the colony's population was censused
fitabilities. This points out the crucial question: how by counting the number of bees inside 10 randomly selected 2.5 x
are the rates of these two processes adaptively regulated ? 2.5 cm grid squares on each side of the observation hive: the colony
After describing the multidimensional modulation of contained approximately 3450 bees. Rainy weather prevailed for
the next 10 days, during which all we could do was provide food
foraging behavior in relation to nectar source profitabili-
at the feeders whenever the bees could fly. This kept the colony
ty, we experimentally analyze how each forager assesses fed and ensured that the dozen or so bees trained to each feeder
nectar source profitability. The two principal competing would continue visiting the feeders. By the time the weather cleared
hypotheses are that each forager judges by itself the pro- (19 June) the colony had grown noticeably; fortunately the foragers
fitability of its nectar source, or that each forager is were still all labelled bees. A census on the evening of 19 June
given this information by the food storer bees in the indicated a population of approximately 4200 bees.
hive. Next we check our understanding of how the joint
Feeders and experimental layout. The feeders were pneumatic feed-
actions of a colony's foragers result in producing colony-
ing dishes identical to those described in a previous paper (Seeley
level decision-making by formulating a mathematical 1989). The observation hive and the two feeders were placed in
model of this process and comparing the decision-mak- an approximately linear arrangement, with the hive in the center
ing patterns of simulated and real colonies. We end by and the feeders 400 m to the north and south, in sinai1 clearings.
discussing the general implications of our investigation. Both food sources contained the scent anise in a vented reservoir
beneath the feeder and also in the sucrose solution (60 gl of anise
extract per liter).
Methods Measuring forager group size, and recruitment and abandonment
rates. We recorded which bees visited each food source during
1. Colony-level analysis: selective exploitation of nectar every half-hour period throughout two 8-h sets of observations.
We accomplished this using roll call sheets listing the identification
sources codes of all 4000 bees and crossing off the identification codes
of all bees seen at a feeder during each half hour. Each sheet
A colony composed of individually identifiable bees was presented therefore showed which individuals visited a particular feeder at
with two nectar sources with different profitabilities. The colony's least once during a 30-rain period, and how many in total did
differential exploitation of the two sources was measured by deter- so. This number we call the "forager group size." Comparisons
mining the total number of different individuals visiting each food between roll call sheets for a feeder also revealed when (to the
source. Also, the different rates of recruitment and abandonment nearest 30 rain) individuals were first sighted at (i.e., were recently
for the two feeders, which ultimately generated the exploitation recruited to) the feeder or were last sighted at (i.e., were soon
difference between the two food sources, were determined from to abandon) the feeder. From these data we calculated the per
records of each bee's visits to the feeders. capita recruitment rate or abandonment rate for each feeder for
each half hour by dividing the number of recruits or deserters
Study site. All observations were made at the Cranberry Lake Bio- counted during that half hour by the forager group size for the
logical Station (44009 ' N, 74°48 ' W) in northern New York State.
previous half hour.
This site is surrounded by forests, bogs, and lakes for more than
20 km in all directions, hence forage for bees is extremely sparse
and no native honey bees survive there. A full-size colony, kept
on platform scales, lost weight (0.50_+0.30 kg) on 15 days and
2. Individual-level analysis: behavior modulation
gained weight (0.15 kg) on just 1 day during the 16-day study in relation to nectar source profitability
period in June 1989. The scarcity of food enabled us to control
the forage collected by our study colony even though its foragers Two colonies were established in adjacent observation hives and
flew freely from the hive. 30 bees from each colony were trained to visit separate feeders
at the same distance from the hives. The profitability of one (the
Study colony. We assembled a colony of individually identifiable experinaental) colony's feeder was systematical(y changed during
bees over a 2-day period, 1-2 June 1989, in Ithaca, as follows. the day and the behavioral adjustments of its foragers were re-
First, at the beginning of each day, approximately 2500 worker corded. MeanwNle, the profitability of the other (the control) col-
bees from a colony of Italian honey bees (Apis mellifera ligustica) ony's feeder was left unchanged and the behaviors of its foragers
were shaken into a wire cage. From this, we periodically shook were recorded as a check for possible confounding changes in the
groups of approximately 50 bees into plastic bags and placed them ambient conditions.
in a refrigerator to immobilize them. After at least 15 min of cool-
ing, the bees in a bag were removed from the refrigerator and Study site, bees, experimental layout, and feeders. This experiment
poured onto a container of "blue ice", where they stayed chilled was conducted in July 1987 in connection with other experiments
during the labelling operation. A bee tag (Opalithpl/ittchen; with already published (Seeley 1989). It involved the same study site,
500 number and color combinations) was glued on the thorax of the same two colonies in observation hives, the same experimental
each bee, and a dot of one of eight different colors of paint was layout, and the same feeders as were described in the prior report.279
The only difference between the present experiment and the one in a previous study (Seeley 1989). Both feeders were positioned
reported in Fig. 2 of Seeley 1989 was that, in the present case, along a rarely used road (one 50 m and the other 1250 m from
the fullness of the experimental colony's honey combs was kept the hive) that traverses flat, treeless agricultural fields; hence, both
constant and the profitability of it's feeder was varied, whereas, feeders had virtually identical exposure to wind and sun. Fifteen
before, the fullness of the experimental colony's honey combs was bees were trained to each feeder using standard techniques (von
varied and the profitability of it's feeder was kept constant. Frisch 1967). Each bee was labeled with paint marks coding indi-
vidual identity and which feeder she was visiting. Recruits to each
Measuring recruitment and recording the behaviors of foragers. Re- feeder were captured in plastic bags to prevent forager build-up
cruitment from each colony was measured by training 30 bees at the feeders.
to a feeder using standard techniques (von Frisch 1967), labeling
these bees (the recruiters) with individually identifiable paint Recording the behaviors of foragers. Labelled foragers were fol-
marks, and counting the number of unlabeled bees (the recruits) lowed one at a time inside the observation hive from time of arrival
appearing at the feeder every 15 rain. Each recruit was captured to departure. Measurements were made, as previously described,
in a plastic bag (to minimize release of alarm pheromone) shortly of the time to start of unloading, the dance circuits per bee per
after its arrival at the feeder, and was frozen at the end of the return to the hive, and the probability of dancing.
day.
Roll calls of the labeled bees at 30-min intervals indicated the Statistical tests. We used t-tests, either Student's or the test for
number of recruiters at each feeder; this number dropped below equality of two proportions using arcsine transformations (Sokal
30 for the experimental colony when the profitability of its feeder and Rohlf 1981).
was lowered and some bees abandoned it. The probability of visit-
ing the feeder was calculated by dividing the number of labeled
bees visiting the feeder by 30, that is, the number of labeled bees
that visited the feeder when all did so. Results
Measurements were also made of several variables of the in-
hive behavior of the labelled bees from the experimental colony. 1. Colony-level analysis
This involved following one (randomly selected) labelled bee at
a time from when it entered the hive to when it exited it. One Pattern: selective exploitation o f nectar sources. F i g u r e 1
variable was the "time to start of unloading," which is the time
shows t h a t w h e n a c o l o n y was given a series o f choices
interval between entering the hive and beginning to transfer nectar
to a food storer bee. The "time to end of unloading" is the time b e t w e e n t w o feeders w i t h different profitabilities, it c o n -
interval between entering the hive and completing the transfer of sistently f o c u s e d its c o l l e c t i o n efforts o n the m o r e p r o f i t -
nectar to food storers. The "time in hive after unloading" is the able feeder. T h e net result was t h a t the c o l o n y s t e a d i l y
time interval between finishing unloading nectar and leaving the t r a c k e d the richest source o f n e c t a r in the c h a n g i n g ar-
hive. Another variable was "dance circuits per bee per return to ray. N o t e t h a t the m e a s u r e o f f o r a g i n g effort s h o w n
the hive," which was measured by counting the dance circuits each
in Fig. 1 - the t o t a l n u m b e r o f different i n d i v i d u a l s visit-
bee performed during its time inside the hive. The "probability
of dancing" was calculated by dividing the number of bees that ing the feeder in a h a l f h o u r - u n d e r e s t i m a t e s the col-
danced by the total number of bees that were followed (i.e., sample o n y ' s r e s p o n s e differential to the two feeders. This is
size). Finall¢, the "foraging tempo," measured in units of trips because the p o o r e r feeder (0.75 mol/1), relative to the
to the feeder per bee per 30 min, was calculated by first dividing richer feeder (2.50 tool/l), was n o t o n l y visited b y fewer
the number of labelled bees visiting the feeder by the number of bees, b u t also was visited at a l o w e r rate b y each bee
labelled bees entering the hive per rain (measured with ten l-rain
(see d a t a on f o r a g i n g t e m p o in Table 1).
counts of labelled bees coming in the entrance). This quotient indi-
cates the average time per roundtrip to the feeder. Next, 30 min
was divided by the roundtrip time to determine the number of Process: tuning the rates o f recruitment and abandon-
trips to the feeder that each bee was making in 30 rain. ment. Because the size o f a g r o u p is d e t e r m i n e d b y the
h i s t o r y o f a d d i t i o n s to a n d s u b t r a c t i o n s f r o m the g r o u p ,
Statistical test. A single classification ANOVA (Sokal and Rohlf it was clear a p r i o r i t h a t the c h a n g e s in f o r a g e r g r o u p
1981) was performed for each variable for which there was an
size s h o w n in Fig. 1 reflect c h a n g e s in the rates o f recruit-
estimate of the mean and variance, i.e., all variables except for
the probability of visiting the feeder, the probability of dancing, ment (additions) and abandonment (subtractions) for
and foraging tempo. each feeder. T h e c o m p l e t e p i c t u r e o f the r e c r u i t m e n t a n d
a b a n d o n m e n t d y n a m i c s for the two feeders is s h o w n
in Fig. 2. O n the m o r n i n g o f 19 June, there was essential-
3. Individual-level analysis: assessing nectar source ly no r e c r u i t m e n t to o r a b a n d o n m e n t o f the n o r t h feeder
profitability (1.00 mol/1), a n d the f o r a g e r g r o u p size here rose o n l y
to 12 bees, n a m e l y the ones t h a t h a d f o r a g e d here the
Two feeders were established, one with a relatively dilute sugar p r e v i o u s day. A t the s o u t h feeder (2.50 mol/1), h o w e v e r ,
solution but located near the hive, and the other with a more
concentrated sugar solution but positioned far from the hive. Bees there was s t r o n g r e c r u i t m e n t , especially e a r l y in the
were trained to forage from each feeder, and their behaviors inside m o r n i n g w h e n the feeder was still lightly e x p l o i t e d , a n d
the hive, especially the strength of their recruitment dances, were little a b a n d o n m e n t , so the f o r a g e r g r o u p size i n c r e a s e d
monitored to determine which feeder was judged more profitable. d r a m a t i c a l l y , rising f r o m 0 to 91 bees in 4 h o u r s . T h e n
The answer reveals whether it is the food storers or the foragers in the a f t e r n o o n , the s o u t h feeder ( n o w 0.75 tool/l) re-
that assess nectar source profitability. ceived negligible r e c r u i t m e n t b u t intense a b a n d o n m e n t ,
Study site, bees, experimental layout, and feeders. This experiment so t h a t its f o r a g e r g r o u p size fell b a c k to j u s t 10 bees
was performed in Ithaca, New York, using a small colony of Italian b y the e n d o f the a f t e r n o o n . M e a n w h i l e , the n o r t h feeder
honey bees in a two-frame observation hive housed inside a labora- ( n o w 2.50 tool/l) h a d b e c o m e the t a r g e t o f s t r o n g recruit-
tory building. The hive and feeders were identical to those described m e n t b u t little a b a n d o n m e n t , with the net result t h a t280
North Feeder Hive South Feeder unexpectedly large forager group size - as m a n y as 32
• bees - for the poorer (south, 0.75 mol/1) feeder on the
T T morning of 20 June. This morning surge in foragers to
8 AM
the poorer feeder arose because bees that had foraged
(9 ® here the previous morning came out to reconnoitre the
site of their prior success, not because bees were recruit-
1.00 M 2.50 M ing nestmates to the 0.75 re•l/1 solution at this feeder.
Virtually all of the bees (46 out of 48) that visited the
south feeder on the morning of 20 June had foraged
...: ::... there the previous morning. Also m a n y (18 out of 48)
LU
Z
-~ Noon
T of these bees ceased visiting the feeder within an hour
® ® of their first visit, so that the bees were actually abandon-
ing this feeder at a rather high rate f r o m 9:30 a.m. to
12 noon (see Fig. 2).
2.50 M ••:o 0.75 M These results indicate that colonies are highly skilled
..'. .... ::..
o • o O o • eO o • o o~°• °
• oOo°•• • • eeee• ooe
at adjusting the rates of recruitment and a b a n d o n m e n t
:::::
@ • ~ oeOOo • • •
• o
000
eoe
oO•
00 eel looooe
•
°
• • ooeo
°oooooe for each nectar source in relation to its profitability,
• •0••o • oo•eeo
• o~ • TO•O • and that it is the precise modulation of these rates which
4 PM
l G T generates a colony's selective exploitation of superior
® nectar sources. This raises the next question: what do
individual bees do to produce this adaptive tuning of
their colony's rates of recruitment and a b a n d o n m e n t ?
• •
2. Individual-level analysis
8 AM
T f~ T
Pattern: behavior modulation in relation to nectar source
(9 (9 profitability. The data presented in Table 1 illustrate the
ability of bees to finely adjust several components of
2.50 M 0.75 M
their foraging behavior in accordance with nectar source
... quality. On the one hand, when the quality was high
(as when the feeder was loaded with the 2.0 re•l/1 sucrose
LU solution), the bees continued visiting the food source,
z T (3 T
~
.-~ Noon worked quickly, and danced vigorously, thereby bring-
® ® ing additional nestmates to help exploit their rich find.
On the other hand, when the nectar source quality was
0.75 M 2.50 M
low (0.5 mol/1 solution, or less), the bees tended to aban-
don the source or, if they did continue foraging there,
they behaved relatively slowly and did not perform re-
cruitment dances. This set of responses resulted in a
y t3 Y drop in the n u m b e r of bees at the feeder. Setting the
4 PM feeder at intermediate levels of quality elicited behavioral
® responses of intermediate strength, with correspondingly
Fig. 1. The basic decision-making phenomenon. When given a intermediate rates of recruitment to or a b a n d o n m e n t of
choice between two food sources with different profitabilities, the the feeder. Note that simultaneous measurements of re-
colony consistently directed most of its foraging effort onto the cruitment rate for a control colony showed no significant
richer one. The number of dots above each feeder denotes the changes, which implies that the ambient conditions were
forager group size for that feeder, i.e., the number of different stable during the observations and, therefore, that the
bees that visited the feeder in the half hour preceding the time behavioral changes documented in Table 1 were solely
shown on the left. For several days prior to the start of observa-
tions, a small group of bees was trained to each feeder (12 and in response to changes in food source profitability.
15 bees for the north and south feeders, respectively), thus on
the morning of 19 June, the two feeders had essentially equivalent
histories of low-level exploitation. The feeders were located 400 m Process: assessing nectar source profitability. H o w does
from the hive and were identical except for the concentration of a forager assess the profitability of its nectar source?
the sugar solution Three hypotheses come to mind. First, each forager
might acquire this information by making direct com-
the forager group size here rose from 12 to 121 bees. parisons of nectar sources. This would require that a
Similar patterns of reciprocal recruitment and abandon- forager visit several nectar sources to judge the relative
ment, in accordance with food source profitability, were value of the source f r o m which it is foraging. Second,
recorded on the second day as well. One difference be- each forager might rely instead upon indirect compari-
tween the data for 19 and 20 June, however, was an sons made by the food storer bees inside the hive. Each281
NORTH FEEDER 1
sucrose sol'n: 1.00 2.50 0.75
(rnol / L)
~-
& 100
-N
~.
o~ 50
o
Ia_ v
E
~ 0.5
.13 Fig. 2. The colony's responses to
[ m ~
the two nectar sources in detail.
,.5
Values of forager group size
denote the number of different
~i N 1.0 individuals that visited each
~' 0.5 feeder during the previous half
hour. Per capita rates of
I I recruitment or abandonment
10 12 2 4 8 10 12 2 4 represent the number of recruits
or deserters recorded for a 30-
min interval divided by the
forager group size for the start of
the interval. At the start of the
I SOUTH FEEDER I observations, 12 and 15 bees had
sucrose sol'n: 2.50 0.75 2.50 experience at the north and south
(rnol / L) --
feeders, respectively, and so
,-~ provided an initial link between
~ 100 m the colony and each feeder. These
bees returned to their respective
-~ feeders on the morning of 19
# June, but were not counted as
o
B~ 5O recruits. Likewise, any bee that
$ had visited a feeder during the
afternoon of 19 June was not
o
IJ. counted as a recruit if it appeared
I I I at the same feeder on the
1.0 morning of 20 June, hence the
increase in forager group size at
E 0.5 the two feeders without strong
..g~
~'~21
O ~ . ~ 3~ m m recruitment on the morning of 20
~ I__ I .,_1
~.13~ June. This figure illustrates the
E~ ~ .~ 1.5
precise regulation of a colony's
recruitment and abandonment
rates in relation to nectar source
0.5 profitability, with the result that
the colony's foraging efforts
I I__J 1__/, tracked the richer nectar source
8 10 12 2 4 8 10 12 2 4
TIME OF DAY
19 June 20 June
food storer bee unloads foragers from various nectar nervous system can calculate the absolute quality of its
sources, and ithas been suggested (Boch 1956; Lindauer source. This hypothesis is closely analogous to Lin-
1961, 1975) that food storers compare the nectar loads dauer's (1955) idea that a scout bee, without previous
they receive and preferentially unload foragers bearing experience in house-hunting, can assess the quality of
the most profitable nectar. A forager's ease in getting a nest site through reference to an "angeborenes Schema
unloaded would, according to this hypothesis, inform eines Nistplatzes." [Note that in the case of foraging,
it of the relative quality of its nectar source. A third food source quality is not the sole factor influencing
possibility is that no comparisons are made, neither by a bee's behavior. Other factors, such as the colony's
foragers nor by food storers; instead, each forager need for food and the seasonal availability of forage,
knows only about its particular nectar source, and its determine the levels of food source quality that are the282 Table 1. Information on the protocol, ambient conditions, and results of the experiment documenting the modulation of foraging behavior in relation to nectar source profitability Time period (hours) 0915-1045 1100-1230 1245-1415 1430-1600 Significance Experimental colony Sucrose solution (tool/l) 2.00 1.50 1.00 0.50 Probability of visiting feeder 1.00 1.00 1.00 0.60 Foraging tempo (trips/bee/30 min) 6.1 6.2 6.0 3.9 Probability of dancing 0.80 0.59 0.26 0.00 Dance circuits/bee/return to hive 9.8 _+15.4 5.4_+4.9 1.6 _+3.9 0.0
283
50 m 1250 rn strongly than did the bees from the 1250-m, 1.00-tool/1
0.75 mol/L 1.00 or 2.50 mol/L
feeder. These results strongly contradict H2.
This rejection of H2 is reinforced by the average
T times to start unloading for the different forager groups.
I I I First, note that the bees from the 1250-m, 1.00-mol/1
Variable
feeder took more time to begin unloading than did the
Time to start of 11 + 4 a 37 + 26 b 12 + 4 a bees from the 50-m, 0.75-mol/1 feeder. This runs con-
unloading (s)
trary to H2, which assumes that food storer bees prefer-
Probability of 0.50 a 0.10 b 0.73 a entially unload foragers bringing in more concentrated
dancing nectar. Second, note that foragers from the 50-m, 0.75-
Dance circuits per 4.5+6.3 a 0.6 + 0 . 3 b 14.6 + 19.6 c tool/1 feeder and the 1250-m feeder, when reloaded with
bee per return to a 2.50-moi/1 solution, began unloading equally quickly,
hive
despite the huge difference in concentration between
Sample size 22 20 22 their sugar solutions. This too contradicts the basic as-
(bees followed)
sumption of H2. However, both of these patterns in un-
Fig. 3. Experimental design and results of the test for how foragers
loading times are consistent with H3. The first pattern
assess the profitability of a nectar source. Fifteen bees were trained evidently is the result of the foragers from the 1250-m,
to forage at each of the two feeders, and their responses to these 1.00-tool/1 feeder concluding that their feeder's profit-
feeders were observed. The bees from the 50-m, 0.75-mol/1 feeder ability was low and slowing their foraging tempo (see
unloaded their nectar more quickly and danced more strongly than Table 1), in part by not seeking to unload their nectar
did those from the 1250-m, ~.00-mol/1 feeder. This implies that immediately upon entering the hive. The second pattern
assessments of nectar source profitability are made by the forager
bees, which have knowledge of the many variables affecting the
apparently reflects the bees from both the 50-m, 0.75-
energetic profitability of a nectar source, rather than by the food tool/1 feeder and the 1250-m, 2.50-mol/1 feeder judging
storer bees, which know only one variable, the sugar concentration that their feeder was high in profitability and deciding
of the nectar to forage with high tempo, which included seeking to
unload nectar immediately upon entering the hive.
3. Model of collective decision-making
food storer bees acquire information about the other
variables influencing the energetic profitability of a nec- We now formulate a mathematical model of nectar-
tar source such as nectar abundance, direction in rela- source selection by honey bee colonies, based on the
tion to the wind, and wind speed since food storers empirical work just presented. The model enables us to
remain inside the hive where they have no direct access see more clearly the colony-level consequences of the
to information about these variables, and they have no individual-level behavior patterns described above. It
known means of gaining information about these vari- therefore provides a test of the idea that the colony-level
ables indirectly from the foragers. decision-making can be accounted for simply in terms
In contrast to H2, H3 implies that a nectar source's of an assemblage of bees, in which each bee independent-
profitability is assessed as a function of many variables. ly modulates its behavior in accordance with the profit-
This is because this hypothesis assumes that the foragers ability of its nectar source. We check the validity of the
are making the assessments, and foragers of course have ideas expressed in the model by comparing the pattern
direct access to information about the multiple variables of exploitation of two unequal nectar sources that is
affecting the energetic profitability of a nectar source. predicted by this model with the pattern of exploitation
The experimental layout shown in Fig. 3 is such that that was actually observed.
H2 and H3 give opposite predictions about the relative
profitabilities (as judged by the bees) of the two feeders, Biological basis of the model. We will start with a pool
and hence about the bees' relative responses to these of foragers that are not yet committed to one of the
feeders. Consider what each hypothesis predicts about nectar sources available outside the hive. Each of these
the bees' responses to the 50-m, 0.75-mol/1 feeder relative uncommitted bees (hereafter called "follower bees ") lo-
to those for the 1250-m, 1.00-tool/1 feeder. H2 predicts cates a nectar source by following the recruitment dances
that the bees foraging from the 50-m, 0.75-mol/1 feeder of a nestmate that is already committed to a patch of
will assess its profitability as relatively low because its flowers. Now consider the behavior of one follower bee
sugar concentration is relatively low, and thus predicts as it begins its day by following dances (Fig. 4). Within
that these bees will dance relatively little. H3, however, the dance floor region of the hive are bees dancing for
predicts that the bees visiting the 50-m, 0.75-mol/1 feeder two nectar sources (A and B). Our bee follows a dancer
will assess its profitability as relatively high because its for nectar source A and is recruited to this source. Hav-
much shorter distance from the hive will more than com- ing located source A, it gathers a load of nectar and
pensate for its lower sugar concentration, and therefore returns to the hive. After unloading its nectar to a food
predicts that these bees should dance relatively strongly. storer bee, the forager proceeds with one of three op-
Which prediction is correct? As we see in Fig. 3, the tions, illustrated by the arrows after the two branch
bees from the 50-m, 0.75-mol/1 feeder danced much more points (diamonds). The first is (1) to abandon the nectar284
( At hiveunloading |
nectar from A
(HA) )
I At hiveunloading 1
nectar from B
(HB)
a m o n g these options. But in this model, as in the study
of behavioral modulation above, we will work with nec-
tar sources that differ only in their sugar concentration.
Note too that this model does not consider changes in
L the response thresholds of the foragers in relation to
P5 P7
changes in foraging conditions (Lindauer 1948; Seeley
P3 t Pl
fxB ~ - 1986, 1989); hence, it represents the process of nectar-
source selection under a fixed set of foraging conditions
1- fxA 1- fx ~3 (weather, forage abundance, colony need, etc.).
Mathematical structure of the model. The essential fea-
tures of the biology just described can be incorporated
fdA (1- fxA) ,~ in a mathematical model of a colony choosing between
two nectar sources as follows. First, we assume that at
any m o m e n t each forager must be in one of the seven
for B (DB) c o m p a r t m e n t s shown in Fig. 4. These compartments
are:
i:~'
~,
~,
1. HA : unloading nectar from nectar source A
(1-fdA) (1- fxA) (1-fdB) (f- fxg)
2. HB: unloading nectar f r o m nectar source B
3. DA: dancing for nectar source A
4. DB: dancing for nectar source B
5. A : foraging at nectar source A
P2
6. B: foraging at nectar source B
7. F: following a dancer
flB
More formally, a bee is considered in one of these seven
i nectar source A
(A)
nectar source B
Fig. 4. Graphical representation of the mathematical model of how
(B)
1 c o m p a r t m e n t s until the m o m e n t she enters her next com-
partment. Thus, for example, the time spent in DA in-
cludes both the time spent dancing for anc't the time
needed to return to nectar source A. Note that the dance
honey bee colonies choose between two nectar sources (A and B).
floor (shaded area in Fig. 4) holds three compartments,
The model is based on the following empirically determined cycle
of foraging behavior. A bee begins foraging by following nestmates one (DA) containing bees dancing for source A, one (DB)
performing recruitment dances, then arrives at a food source and containing bees dancing for source B, and one (F) hold-
gathers a load of nectar, and finally returns to the hive and unloads ing bees following a dancer. Note too that Fig. 4 consists
the nectar. At this point it repeats the cycle, though now with of two separate cycles, one for each food source, with
the option of continuing to forage from its current nectar source the follower c o m p a r t m e n t (F) the only intersection point
(and perhaps even dancing for this source), rather than following
for the two cycles. Thus bees f r o m one nectar source
dances to locate a new source. The black diamonds along the arrows
denote decision points. The grey circle represents the dance floor can switch over to the other source only by passing
area inside the hive. At any given moment, each forager is in one through the dance floor and following a dancer for the
of the seven compartments shown (HA, HB, F, etc. denote both other nectar source. Figure 4 suggests that the dance
the respective compartments and the number of foragers in these floor lies at the heart of the decision-making process.
compartments). The model predicts a distribution of the foragers Whatever information transfer occurs a m o n g the for-
among the seven compartments over time as a function of the
rate constant (p~)associated with each compartment and the proba- agers is assumed to happen here.
bilities (f~A,fx B, etc.) associated with each of the five branch points. Two variables affect the proportion of the total for-
Empirical studies have revealed the behaviors of individual bees. ager force that is in each of the seven compartments:
A computer simulation of this model enables us to see the colony- (1) the rate at which individual foragers move f r o m one
level implications of these individual-level behavior patterns and c o m p a r t m e n t to another and (2) the probability that
thereby test our understanding of the way that these behavior pat- a forager takes one or the other fork at the five branch
terns combine to produce a colony's ability to choose among nectar
points (diamonds) in Fig. 4). The fraction of bees leaving
sources
a c o m p a r t m e n t in a given time interval~ is denoted by
the appropriate rate constant Pi. For example, the rate
source and return to the pool of follower bees. If, how- constant for bees leaving c o m p a r t m e n t A is P3 and the
ever, she elects to continue with her nectar source, then fraction of the bees at nectar source A that leave in
she has two other options, either (2) to perform recruit- a time interval At is equal to p3At. The values of the
ment dances before heading back to the flowers, or (3) rate constants pi will be calculated f r o m experimental
to return to the flowers directly without attempting to data in the next section.
recruit nestmates. M a n y variables of nectar sources - N o w let us consider what determines the probabilities
including nectar sweetness, nectar abundance, nectar ac- of the different behaviors at each of the five branch
cessibility, and distance from the hive (Seeley 1986 and points in Fig. 4. The first branch point occurs after a
references therein) - influence foragers as they choose bee has unloaded her nectar in the hive. At this point285
Table 2. Parameter values for the model of a colony choosing between two nectar sources, as shown in Fig. 1. A and B correspond
to the 2.50-mol/1 and the 0.75-mol/1 feeders, respectively. For full definitions of the parameters, see the text and Fig. 4
Parameter: definition Value Reference
T1 : time from start of unloading to start of following, dancing, or foraging, A foragers 1.0 min Table 1"
T2 : time from start of dancing to start of foraging, A foragers 1.5 rain Table 1
T~ : time from start of foraging to start of unloading, A foragers 2.5 min Table 1
T~ : time from start of following dancers to start of foraging, A and B foragers 60 rain Seeley and Visscher 1988 b
Ts : time from start of unloading to start of following, dancing, or foraging, B foragers 3.0 min Table 1 ~
T6 : time from start of dancing to start of foraging, B foragers 2.0 rain Table 1
T~ : time from start of foraging to start of unloading, B foragers 3.5 min Table 1
f ~ : probability of abandoning A, per foraging trip 0.00 Table 1
f ~ : probability of abandoning B, per foraging trip 0.04 Table 1 ~
f d A : probability of dancing for A 1.00 Table 1
fe~: probability of dancing for B 0.I5 Table 1
a Since the probability of dancing upon return from A is 1.00, T1 can be more precisely defined as the time from start of unloading
to start of dancing. Conversely, since the probability of dancing upon return from B is near zero, Ts can be defined as essentially
the time from start of unloading to start of following or foraging
b This paper reports that in order to find a rich patch of flowers by following dances, on average a bee spends a total of 121 rain
outside the hive searching for targets indicated by dances. The total time spent inside and outside the hive will be somewhat greater
still. The value given here, 60 rain, is less than these estimates because in the Fig. 1 experiment the recruitment targets (feeders) were
only 400 m from the hive, whereas in Seeley and Visscher (1988) the recruitment targets (natural flower patches) were on average
1600 m from the hive
c fx for the 0.75-mol/1 feeder was calculated as follows. Table I shows that a bee's probability of continuing to visit (=not abandoning)
a 0.75-mol/1 feeder after 30 rain was approximately 0.80. Because each forager made approximately 5 trips to the feeder in 30 min,
this implies that (I - p ) S = 0.80, where p = the probability of abandoning the feeder per trip. Solving this equation yields p = 0.04
she m a y a b a n d o n the nectar source and return to the o f encountering and following a dancer for nectar source
dance floor to follow a n o t h e r dancer. The probability A (fl A) at any particular m o m e n t can be r o u g h l y estimat-
that a bee does so is d e n o t e d by the function fx, which ed by DA/(DA+ De). However, because Di includes b o t h
we call the a b a n d o n i n g function. In general, the value bees dancing for i and bees that have ceased dancing
o f f x will depend on the profitability o f the nectar source; but are still in the dancing c o m p a r t m e n t , the ratio DA/
thus, fx A denotes the probability that a bee leaving HA (DA+ D~) will often be a p o o r estimate off1A. F o r exam-
will a b a n d o n nectar source A and b e c o m e a follower ple, if nectar source A is scarcely d a n c e d for but its
bee (F). A b a n d o n m e n t diminishes the n u m b e r o f bees foragers linger in the hive w i t h o u t dancing, then the
c o m m i t t e d to a nectar source and provides a pool o f ratio DA/(DA+D~) will greatly overestimate the likeli-
u n c o m m i t t e d bees that follow dancers for one nectar h o o d o f encountering a dancer for A. It can be m u c h
source or the other. i m p r o v e d by multiplying each Di in the ratio by the
The second b r a n c h point applies to bees that did p r o p o r t i o n o f time that foragers in the dancing c o m p a r t -
n o t a b a n d o n their nectar source. It determines w h a t pro- m e n t i are actually dancing, which we will denote as
p o r t i o n o f these bees will dance for their nectar source. "~i'
A l t h o u g h at this b r a n c h point there is no filtering o f For simplicity in m a k i n g calculations with the model,
bees a w a y f r o m nectar sources, the o u t c o m e o f this we m a k e three further assumptions: (1) the total n u m b e r
b r a n c h point affects the probability o f a follower bee o f bees foraging is fixed, (2) all bees begin foraging si-
following dances for each nectar source, as described multaneously, a n d (3) all the foragers go to either one
hereafter. The probability that a bee becomes a dancer o f two nectar sources. In the appendix we present the
for her nectar source is d e n o t e d by the function re. The set o f differential equations that describes this m o d e l
function fd is called the dancing function and, as with o f colony-level decision-making.
the a b a n d o n i n g function, its value for a given bee de-
pends on the profitability o f the bee's nectar source, Model parameters. We n o w calculate estimates o f the
with feA d e n o t i n g the probability o f dancing by a bee model's parameters for the situation s h o w n in Fig. 1,
foraging at nectar source A. namely a c o l o n y c h o o s i n g between two equidistant su-
The t h i r d , b r a n c h p o i n t arises when bees follow crose solution feeders that are identical, except that one
dancers for one or a n o t h e r nectar source. The probabili- contains a 0.75-mol/1 solution and the other contains
ty o f a follower bee following dances for nectar source a 2.50-mol/1 solution. Let us first consider the rate con-
A or B is d e n o t e d by the functionf~ A or f l ~, respectively, stants. These can be calculated f r o m the time required
which we call the following function. A recent study for a bee to get f r o m one c o m p a r t m e n t to a n o t h e r (i.e.,
(Seeley and Towne, in preparation) has s h o w n that fol- the time spent in the c o m p a r t m e n t plus the time to get
lower bees simply follow the first dancer they encounter to the next c o m p a r t m e n t ) . E a c h rate constant, p~, is
and that encounters occur at r a n d o m . Hence, in the situ- equal to l/T,., where each T; is the time required to get
ation o f just two nectar sources, A and B, the likelihood f r o m the relevant c o m p a r t m e n t to next. Values o f T~286
appropriate to the foraging situation shown in Fig. 1 North: 0.75 mol/L 2.50 mol/L
are given in Table 2. South: 2.50 m o l / L 0.75 mol/I_
The abandoning function (fx) and the dancing func-
tion (fe) include many variables - such as the profitability
of the nectar source, the colony's need for nectar, and ~ 100 North feeder bees
&
the weather conditions - hence it is difficult to give a
general description of these functions. However, the R
._
co
values of these functions have been measured (see Ta- o_
,o 50
ble 1) for a set of conditions virtually identical to those ob
under which the Fig. 1 data were recorded: feeders
400 m from the hive, high colony need for nectar, sparse Follower bees
~_ South feeder bees
ambient forage, and fair weather. We used the data in
Table 1 (with appropriate interpolations and extrapola- 10 12 2 4
tions) to determine the values offx andfe that are appro- Time of day
priate to simulating the colony's choice between 2.50- Fig. 5. An example of the nectar-source selection by honey bee
tool/1 and 0.75-mol/1 nectar sources as shown in Fig. 1 colonies predicted by the model shown in Fig. 4. For this simula-
(see Table 2). tion the modeI's parameters were adjusted to correspond to the
Values of the following function, f l , were calculated situation shown in Figs. 1 and 2, namely a colony choosing between
for the two nectar sources using the more accurate of two equidistant (400-m) feeders offering either a 0.75 or a 2.50-mol/
1sucrose solution ad libitum. The forager group size for each feeder
the two methods just described. Thus, each Di was first equals the sum of the bees dancing for the feeder (DA or De),
discounted by the factor % the proportion of time in the bees at the feeder (A or B), and the bees unloading nectar
Di that was actually spent in dancing. Each q was calcu- from the feeder (HA or H,). In this simulation, as in reality, the
lated by taking the product of the average number of colony always focused the vast majority of its foraging effort on
dance circuits and the average circuit time, and dividing the richer feeder, and was able to respond rapidly to a reversal
it by the total time a bee is in the compartment D~ (T~ in the location of the richer feeder. This result supports the model
of colony-leveldecision-makingshown in Fig. 4
and T6 for compartments DA and D , , respectively). Gi-
ven that the bees from the 2.50-mol/1 and 0.75-mol/1
feeders danced on average 14 and 1 circuits, respectively
(see Table 1), and each dance circuit for a 400-m target creased sigmoidally while those foraging from the south
requires approximately 2.4 s (von Frisch 1967), and gi- feeder decreased exponentially. The maximum rates of
ven values of Tz=90 s and T6 = 120 s (see Table 2), this increase at the north feeder were 34 and 18 bees/30 min
yields values of ZA=0.38 and z,=0.02, where A and for the experimental and simulation results, respectively,
B represent the 2.50- and 0.75-mol/1 feeders. and the corresponding maximum rates of decrease at
the south feeder were -19 and -30 bees/30 min. The two
sets of results show a less close match in the morning.
Results of the model. Figure 5 shows the pattern of simu- This is because in the field the north feeder was loaded
lated colonial decision-making obtained by numerical with a 1.00 rather than a 0.75-mol/1 solution, in order
integration of the model's equations with the following to prevent total abandonment of the north feeder,
starting conditions: A and B = 11, DA and DB= 1, HA whereas in the simulation the north feeder was loaded
and HB = 0, and F = 101. These values correspond closely with a 0.75-mol/1 solution, in order to keep the simula-
to the starting conditions for the experiment shown in tion simple. Also, in the field the bees did not begin
Figs. 1 and 2 in that at 8 : 00 a.m. on 19 June there were foraging simultaneously, whereas in the simulation they
approximately 12 bees committed to each feeder, and all began at 8:00 a.m. sharp.
during the course of the day a total of approximately
125 different bees visited the two feeders. The pattern
generated by the model closely resembles the one gener- Discussion
ated by the colony of bees; compare Fig. 5 with Fig. 2.
As in the case of the experimental data, the plot of the How does the ability of a honey bee colony to choose
model's results shows consistent increases in the number among nectar sources emerge from the behaviors of
of foragers on whichever feeder provided a 2.50-mol/1 thousands of bees? This question, which is the heart
sugar solution, and consistent decreases at the 0.75-mol/1 of this study, is just one form of the general question
feeder. What is especially important is the comparison of self-organization in biology (Pattee 1976; Nicolis and
between the experimental data and the model's predic- Prigogine 1977; Haken 1978): how does the adaptive
tions for the afternoon of 19 June. This is because the structure and functioning of a living system arise out
starting conditions were nearly matched for the field of the joint actions of its subsystems? Colonies of the
and the model for this time period - 12 and 3 bees advanced social insects express this puzzle especially
committed to the real and simulated north feeder strongly since they present a clear juxtaposition of two
(2.50 mol/1), and 90 and 119 bees committed to the real levels of adaptive organization, the colony and the or-
and simulated south feeder (0.75 mol/1). As the after- ganism, yet it is rarely obvious how the properties of
noon progressed, both in the field and in the computer, a colony emerge from the collective activities of its con-
the number of bees foraging from the north feeder in- stituent insects. Fortunately, colonies of social insects287
are unusually favorable subjects for studies of self-orga- several kilometers from the hive (Visscher and Seeley
nization because, compared with organisms or cells, their 1982) and are not easily located, even by bees that are
dissection and experimental analysis is relatively easy. directed to the patches by recruitment dances (Esch and
Nearly 20 years ago Wilson (1971, p 227) identified the Bastian 1970; Mautz 1971; Seeley and Visscher 1988).
special challenge and opportunity that the social insects More importantly, our data contradict HI, at least as
offer to understanding biological organization when he a mechanism commonly used by bees. Of 117 bees in-
wrote "the reconstruction of mass behavior from a volved in generating the pattern of selective exploitation
knowledge of the behavior of single colony members on the morning of 19 June (see Figs. 1 and 2), only
is the central problem of insect sociology." 2 visited more than one nectar source.
What about the second hypothesis (H2), that the
food storer bees compare the profitabilities of different
1. Synthesis of the analysis nectar loads and indicate (by the willingness to accept
nectar) to each forager the relative profitability of its
This investigation began with careful documentation of forage? According to this hypothesis, nectar source qual-
the adaptive functioning of the whole system, i.e., nectar- ity is assessed by bees simply as a function of the sugar
source selection by honey bee colonies. Previous studies concentration of the nectar, since this is evidently all
(Darwin 1877; Butler 1945; Weaver 1979; Visscher and that a food storer bee knows about a nectar source.
Seeley 1982) had suggested that colonies can selectively But as the experiment reported here (Fig. 3) has shown,
exploit rich food sources and can rapidly shift foragers bees evaluate nectar source profitability as a function
among food sources following changes in the food of variables besides sugar concentration, such as distance
source array. The present findings strongly confirm these from the hive. This result falsifies H2. This conclusion
ideas. Even a cursory glance at the exploitation patterns is reinforced by the reports of Boch (1956) and Wad-
shown in Figs. 1 and 2 reveals these skills. dington (1982, 1985) that foragers adjust the strength
The next step was to explain these colonial foraging of dancing in response to multiple variables affecting
patterns in terms of the behaviors of individual bees. the energetic profitability of a nectar source - such as
Fortunately, these two levels of description are directly distance from the hive, difficulty of obtaining the nectar,
bridged by the fact that the colony-level patterns and nectar abundance - not just sugar concentration.
(numbers of foragers allocated to each food source) are By process of elimination, our results support the
a direct result of two processes (recruitment to and aban- third hypothesis (H3) that each forager independently
donment of food sources) that are rather easily under- assesses the profitability of its nectar source and does
stood in terms of bee-level behavior patterns. Hence the so without reference to other nectar sources. This infor-
key to linking the colonial and organismal descriptions mation processing evidently works as follows. In visiting
of nectar-source selection was understanding how indi- a flower patch, a nectar forager takes in information
vidual bees modulate their colony's rates of recruitment about the energetic gains and costs associated with for-
and abandonment to different nectar sources in accor- aging at the patch - as determined by such variables
dance with their different levels of profitability. as the distance from the hive, sugar concentration of
By recording the behaviors of bees visiting a feeder, the nectar, and nectar abundance - and integrates this
and periodically changing the profitability of the feeder, information to arrive at an estimate of the overall ener-
we were able to assemble a detailed picture of the multi- getic profitability of its patch. Each forager's nervous
dimensional modulation of foraging behavior in relation system is evidently calibrated so that it knows whether
to nectar source profitability (Table 1). Behaviors that this level of profitability is low, high, or some point
the bees vary include the strength of waggle dancing, in between. Nectar sources that rank low will generally
the tempo of visits to the food source, and the likelihood elicit abandoment of the source, and nectar sources that
of returning to the food source. Other investigators (von rank high will generally elicit recruitment to the source,
Frisch 1967, p 45; Nunez 1966, 1970, 1982) have ana- though the precise levels of profitability that are the
lyzed subsets of these behavioral variables in a similar thresholds for abandonment and recruitment are ad-
fashion and have reported comparable patterns of be- justed by the foragers as a function of other variables
havior modulation in accordance with food source prof- independent of any one nectar source, such as the col-
itability. Therefore, we feel that the patterns shown in ony's nutritional status (Lindauer 1948; Seeley 1989),
Table 1 are general for bees. In broad terms, as food the seasonal forage availability (Lindauer 1948; Seeley
source profitability increases, the duration of dancing 1986), and the weather conditions (Schufi 1952; Boch
increases, the tempo of foraging increases, and the prob- 1956). If all the foragers in a colony have shared rules
ability of abandoning the food source decreases. for determining the profitability of nectar sources and
This begs the question: how does a forager assess for adjusting the response thresholds, then even though
the profitability of its nectar source? Three hypotheses the foragers operate independently and with limited in-
(H1-H3) were examined. The first (H1) is that each bee formation, they will generate a coherent colony-level re-
makes this assessment by visiting several nectar sources sponse to different food sources, wherein better food
and directly comparing its nectar source with others to sources are recruited to and heavily exploited while
estimate its relative profitability. A priori this seems im- poorer ones are abandoned. To be sure, the details re-
probable because in nature the flower patches exploited main obscure about the information processing inside
by a colony are widely scattered over an area extending a bee as it assesses the profitability of a nectar source,288
especially how this calculation is performed by the cen- thermoregulation in honey bees (Baird and Seeley 1983;
tral nervous system; however, what is now clear is that Heinrich 1985), nest building in polybiine wasps (Jeanne
each nectar forager makes its own, independent assess- 1986), food-source selection in pavement ants (Pasteels
ment of the profitability of its nectar source and that et al. 1987a, b), and raiding in army ants (Deneubourg
it makes this assessment based only on local information et al. 1989; Franks 1989), to cite just a handful of exam-
from her particular nectar source. Hence, paradoxically, ples. It seems that distributed control will prove the rule,
global order emerges from the actions of many indepen- at least in the advanced eusocial species where colony
dent actors, and moreover, none of the actors possesses populations are generally on the order of l03 or larger
an overview of the total operation (though they do pos- [there is good evidence of centralized control in the pri-
sess shared information about the rules for assessing nec- mitively eusocial insects such as paper ~asps, Polistes,
tar source profitability and about such variables as the where colonies usually contain less than 102 individuals
colony's nutritional state and the weather). (Reeve and Gamboa 1983)]. Why might this be?
We now know a good deal about how bees modulate More precisely, what factors have favored or disfa-
their behavior in relation to nectar source profitability, vored the evolution of decentralized control for large
but is this behavioral modulation, performed by all the societies? Considering first the reasons that would seem
foragers in a colony, really sufficient to account for the to disfavor decentralized control, there is no question
nectar-source selection ability of the colony as a whole? that omitting a central authority leaves a group prone
Evidently it is. The computer simulation of decision- to conflicting actions among its members as individuals
making by a colony wherein each forager simply adjusts respond to their different, local situations rather than
her behavior according to the profitability of her nectar the shared, global condition of the group as a whole.
source and, when being recruited to a new nectar source, Colonies of social insects frequently demonstrate this
follows a dancer chosen at random from among the problem, for example, when an ant colony changes nest
dancing foragers, yields patterns of nectar source exploi- sites and some workers carry brood out of the old site
tation that are remarkably similar to what was observed while others carry them back in again (Wilson 1971,
experimentally (Figs. 2 and 5). This does not imply that p 224). Also, without anyone supervising a group, redun-
the rather simple model (Fig. 4) underlying these simula- dancies within the group are likely to arise, especially
tions contains a complete description of the decision- for any activity that requires a long time to produce
making process. Nevertheless, the match between the negative feedback effects to regulate the number of indi-
model's predictions and reality certainly suggests that viduals undertaking the activity. Furthermore, the ab-
this model captures the essence of the decision-making sence of a strong central authority within a group prob-
process. This can be thought of as a kind of natural ably hampers its ability to amount a rapid, global re-
selection among alternative nectar sources as foragers sponse to a critical stress, such as an attack by a preda-
from higher profitability sources "survive" (continue tor. On the other hand, distributing the control of a
visiting their source) longer and "reproduce" (recruit group widely among its members almost certainly en-
other foragers) better than do foragers from lower prof- hances its ability to make a rapid, local response to a
itability sources. The net result is that the percentage stress because it eliminates the need for time-consuming
of a colony's forager population associated with a nectar communication between the peripheral sensor/effector
source will automatically increase if the profitability of individuals and the central decision-making individuals.
the nectar source is relatively high, and will inevitably Also, distributing the control may often make the group
decrease if its profitability is relatively low. Hence, inside less prone to incapacitating damage since the group's
a bee hive, as in the natural world in general, the blind functioning will not depend critically on the status of
process of natural selection generates impressive order, a small subset of supervisory individuals.
and we need not invoke any higher-level decision-maker We suspect, however, that the principal reason that
(no overseer) to explain this order. decentralized control has evolved in colonies of ad-
vanced social insects is that it is a means for a society
to achieve a high degree of organization without highly
sophisticated communication and computation devices
2. The rationale for decentralized control (Hayek 1945; Simon 1981). If the large colonies of social
insects were run by a central authority, then each colony
This conclusion is important, for it points out that the would have to possess an elaborate communication net-
control of a colony's foraging behavior is highly decen- work so that information about the entire colony could
tralized. Indeed, it seems clear that it is distributed flow adequately to the central planners, and so that
among all of the foragers in the colony, with each bee orders from the central planners could be transmitted
independently adjusting its behavior to what is appro- back to the appropriate members of the colony. Equally
priate to its own food source, and the colony-level de- importantly, the central planning body would have to
cision-making arising automatically out of the resultant possess extremely sophisticated information-processing
dynamics in recruitment and abandonment. Decentral- abilities to perform the difficult calculations needed to
ized control of colony functioning has been reported integrate information about the entire colony and pro-
in a number of contexts, including recruitment to food duce orders that would result in a rational allocation
sources in fire ants (Wilson 1962), queen rearing and of the colony's resources. Both of these complex devicesYou can also read
