FEATURE ARTICLES

Group Decision Makingin Honey Bee Swarms

When 10,000 bees go house hunting, how do theycooperatively choose their new nesting site?

Thomas D. Seeley, P. Kirk Visscher and Kevin M. Passino

The problem of stKial choice has chal-lenged stKial philosophers and po-

litical scientists for centuries. The fun-damental decision-making dilemmafor groups is how to turn individualpreferences for different outcomes intoa single choice for the group as a whole.This problem has been studied mainlywith respect to human groups, whichha\'e developed a variety of voting pro-cedures to single out one option froma list of possible choices: majority rule,plurality wins, various weighted-votingsystems and others. Social choice in ani-mal groups is less well studied, althoughexamples are abundant: A babixm troopmust decide where to go following a restperitxi; an ant colony decides whether ornot to attack a neighboring colony.

A striking example of decision mak-ing by an animal group is the choiceof a nesting site by a swarm of up to10,000 honey bees. This process involves

Tlioims D. Scchy is n proffSMr of uctirobioh^i/ uiui

tkiiavior at Corudl Uiiiwrsiti/. He rccciivd his Ph.D.

hi biologi/from Harvard Uniivrsit}/ in 1978. His

research Reuses on fhc functional organization of

homy bee colonies. P. Kirk Visscher earned hif: Ph.D.

in t'ntoitiologif from Cornell in 1985and i? currenthf

an associate professor ofcntomolo^i/ at the Univer-

sity of California. Riivrsidc. He Sfvcializi.'s in the

study of homy iiee SiKial Miai'iorami commuiiiai-

tion. Kevin M. Passino is a professor of electrical and

computer aigineering at Tlie Ohio State Universiti/.

He receiivd his Ph.D. in electrical engineering (sys-

tems and applied mathematics) from the Uniitersitx/

of Notre Dame in 1%9. His research Hx-iises on s\fs-

h'ntsbiologif and design of hio-iiispired techitologies.

Address for Scclei/: Dqiartniciit of Neiirohiologi/and

BclkivioK Cornell University, Mitdd Hall, Ithaca,

NY 14853. Internet: tds5@cornell.edii

several hundred bees from the swarmworking together to find a dozen ormore candidate nesting cavities in treesand then selecting the best one of theseoptions for their new home. We've beeninvestigating this process for the pastdecade using a variety of observational,experimental and mathematical-model-ing studies. This work has revealed a setof behavioral mechanisms in a swarmthat consistently yields excellent collec-tive decisions. It has become clear thatthis group intelligence is a product ofdisagreement and contest, not consensusor compromise, among different groupsof bees representing different alterna-tives in the decision-making task. Wehave found that evolution has suppliedan intriguing answer to the question ofhow to make a group function as an ef-fecti\ e decision-making unit.

Pioneering WorkFor centuries beekeepers have knownthat in late spring or early summer astrong colony of honey bees will divideitself by swiiruiiiig, a process in whichthe queen and approximately half theworker bees leave their hive to establisha new colony; meanwhile a daughterqueen and the balance of the workersremain behind to perpetuate the old col-ony. Beekeepers also have known thatafter a swarm leaves its parental hive,the bees will coalesce into a beardlikecluster on a nearby tree branch, conducta search for a home and, if left alone,eventually launch into flight and moveoff together to their new abode, usuallya far-off hollow h-ee. People have longcaptured the bivouacked swarms that

they have found and installed them inmanmade hives, cutting short the bees'nest-site search. Thus it is not surprisingthat this decision-ma king process longremained a deep mystery.

This situation began to change in the1950s when Martin Lindauer, a Ger-man zoologist, published his seminalpaper on house hunting by honey bees.Lindauer was then a postdtx:toral stu-dent at the University of Munich, shjdy-ing with the famous Karl von Frisch,who had shortly before decoded thelon^glc dance of honey bees. This com-munication behavior allows successfulforagers to inform hi\'e mates of the k>cations of rich food sources through aspecific series of movements. A dancingbee runs forward and performs the wng-glc run, vibrating her abdomen laterally,then circles back to her starting point,producing one dance circuit. A singlebout of dancing contaiiis many of thesecircuits. Von Frisch found that the lengthof a bee's waggle run traiislates intt) thedistance to the food source, and the an-gle of the dance represents direction.

Lindauer was a keen observer. On oneoccasion when he was using his skills ona swann of bees that had settled outsidethe university's Zoological Institute, henoticed that bees on the surface of tbeswarm were performing waggle dances.Moreover, he observed that these danc-ers on the swarm, uiilike those in a hi\ e,did not bear loads of nectar or pollen.Evidently, these were not foragers ad-vertising profitable food sources. Mightthey be scouts reporting potential nestsites? This was a previously unknownuse of waggle dancing.

220 Americtin Scientist, Volume 94

Figure 1. Bivouacked on a tree branch, a honey bee swarm of some 10,000 workers and one queen delegates the job of finding and choos-ing a new nesting site to just a few hundred scout bees. The other bees remain quiescent during the process, conserving energy. How thescout bees select candidate sites, deliberate among choices and reach a verdict is a process complicated enough to rival the dealings of anycorporate committee. Once the scout bees have selected a new home, they stimulate the swarm to launch into flight and then steer it to itsnew domicile. (All photographs courtesy of Thomas D. Seeley.)

Lindauer answered this question bypatiently observing all the dancers onseveral swarms, a marathon task that de-manded many days of steady bee watch-ing and frantic note taking. Whenever hesaw a new dancing bee, Lindauer noted

the location coded in her dance and gaveher a paint dot to avoid repeatedly re-cording her dance information.

This paijistaking work yielded sever-al remarkable discoveries. One was thatduring the decision-making process.

only a few hundred of the thousands ofbees in a swarm were active—flying toand from the swarm, presumably find-ing and inspecting potential nest sites,then performing and following dances.Most bees remained quiescent, prob-

www,amencanscientist,org 2006 May-June 221

Figure 2. Honey bees' real-eslate preferences are for a cavity perched high off the ground witha volume of at least 20 liters, accessed by a hole located at the base of the cavity that is smallerthan 30 square centimeters and faces south. In this tree (left) the entrance is a knothole in theleft fork of the trunk. After the tree was cut down, the nest inside was exposed (right).

ably to conserve the swarm's energysupply, until a decision had been madeand it was time to fly to the chosen site.A second curious find was that, at first,the bees' dances indicated various sitesaround the swarm, but hour by hour thenumber of sites ad\-ertised by the danc-es declined until just one site remained,which was excitedly reported by dozensof dancing bees. Lindauer also foundthat shortly after the bees' dances hadbecome ftjcused on one site, the entirecluster of bees would suddenly take offand fly toward this site. Sometimes hemanaged to sprint beneath the swarmthroughout its cross- country flight andso learned its precise destination—al-ways a cavity in a tree or building andalways at the spot indicated in the fi-nal dances. There was no doubt thenthat the dancing bees were reportingnest sites. Indeed, it seemed these beeswere holding a kind of plebiscite on the

swarm's future home, although exactlyhow they conducted their deliberationswas still unknown.

Renewed AnalysisIn the mid 199()s we decided to lookmore deeply at this intriguing exampleof animal democracy. In the years sinceLindauer's work, several in\'estigatorshad studied the real-estate preferencesof honey bees and had found that a first-rate home for a honey bee colony hasa cavity volume greater than 20 litersand an entrance hole that is smaller than30 square centimeters, perched severalmeters off the ground, facing south andlocated at the floor of the cavity. But noone had figured out exactly how thescout bees in a swarm implement thesehousing preferences during their collec-tive choice of a new home.

Our first step in renewing the analysiswas to repeat Lindauer's observations of

the scout bees' dances, but using modem\'ideo equipment to get a more completepicture than had been possible in the1950s. We worked with small swarmsof about 4,0(X) bees and labeled each beefor individual identification, so we couldattribute each dance to a particular indi-\ idual and thus ascertain her contribu-tion to a swarm's decision making.

From our recordings of every danceperformed by each scout bee, we founda pattern of dancing by nest-site scoutsthat closely resembles what Lindauerreported based on his records of onlyeach scout's first dance. For example, ina swarm we obser\'ed on July 20 to 22,1997, the enfire decision-making prwessrequired about 16 hours of dance activityspread over three days. During the firsthalf of the process, the scouts reportedall 11 of the potential nest sites that theywould consider, and no one site domi-nated the dancing. During the secondhalf, however, one of the sites graduallybegan to be advertised much more thanthe others and ultimately became thechosen site. Indeed, during the last fewhours of the decision making, the sitethat had emerged as the frontrunner be-came the object of all the dances.

Consensus or Quorum?Gi\'en tbe striking way that the danc-es on a swarm come to represent onesite and then the swarm mo\'es to thissite, it was tempting to conclude thata swarm's decision-making process isessentially one of consensus building,rather like the arrival of the "Sense ofthe Meeting" aniong Quakers. By thishypothesis, a scout bee "votes" in favorof a site by dancing for it, somehow thescouts act and interact so that graduallytheir votes come into agreement in fa-vor of a superior site, and somehow thevoting pattern of the scouts is steadilymonitored so that they know whenthey've reached an agreement and canstart acting on their decision.

There were, however, two factorsthat cast doubt on this appealing hy-pothesis. First, neither Lindauer nor wehad seen any sign of scout bees pollingtheir fellow dancers, something thatsurely they must do to know whenthey've reached an agreement. Second,both Lindauer and we had occasion-ally seen a swarm launch into flightwithout a dance consensus, that is,when there were two strong coalitionsof dancers advertising two distinctsites. Were these rare cases of takeoffswith dissent simply bizarre anoma-

222 American Scientist, Volume 94

3 -

T: O

o w

o•^ 1

0 1,000 3,000 5.000

distance to food source (meters)

Figure 3. Honey bees employ waggle dances to inform others about food sources, as shown here, but the same dances are also used to describe thelocation of nest sites. Here, flowers lie along a line 40 degrees to the dght of the Sun as the bees leave their nest (left). To report this food source, a beeruns through a figure-eight pattern on a vertical comb (center). As she passes through the central portion of the dance, she performs the wa^^^te run,vibrating her body laterally, and the angle of the run indicates the direction to the food source. The duration of the waggle run relates to the distanceto the food source (right). When waggle dancing refers to nest sites, it occurs on the surface of a swarm rather than on the combs inside a hive.

lies that we could ignore, or were theys'aluable clues that we should lieed?

We chose to heed them, because wehad long wondered whether tlie essenceof a swarm's decision maldng might besensing a quorum (sufficient number ofscouts) at one of tbe nest sites rather tbansensing a consensus (agreement of danc-ing scouts) at the swarm cluster. By thisquorum-sensing hypothesis, a scout bee"\'otes" for a site by spending time atit, somehow the scouts act and interactso that their numbers rise faster at su-perior sites, and somehow the bees ateacb site monitor their numbers tbere sothat they know whether they've reachedtbe tbreshoid number (quorum) and canprcKeed to iiiitiating tbe swarm's moveto tbis site. This bypotbesis can explaijTthe cases of liftoff witb dissent as in-stances where a quorum was reached atone site before tbe competition betweendancers from different sites had elimi-nated tbe dancing for all but one site.

We tested tbese two bypotbeses withexperiments performed on AppledoreIsland, site of the Shoals Marine Labora-tory of Cornell University. Tbis island,off the coast of Maine, is nearly treelessand so is lacking in natural nesting cavi-ties for honey bees. Each swarm that weferried to this island was thus compelledto pay attention to tbe special nest boxesthat we provided. In our first experi-ment, we presented several swarms,one at a time, with two identical nestboxes, each one a superb nest site. Theswarm was positioned at the island'scenter, wbereas botb nest boxes wereplaced near the rocky shore, eacb one250 meters from tbe swarm but in dif-ferent directions. We found tbat whenswarms were forced to cbtxise between

two first-rate nest sites, they would rou-tinely take off when scout bees were stilldancing strongly for both sites. Consen-sus among dancers was certainly notnecessary for tbese swarms to start fly-ing to one of tbe sites, hence we couldreject the consensus-sensing hypothesis.At tbe same time, we gained supportfor the quorum-sensing hypothesis, he-cause we noticed that swarms consis-tently started preparing for flight once15 or more bees were seen togetber atone of the nest boxes. It sboLild be noted,bowever, tbat because the bees spendthe majority of tbeir time at the swarm.

seeing at least 15 bees at a nest site atany one time means that approximately150 bees overall are visiting the site.

In our second experiment on Ap-pledore Island, we explicitly tested thequorum-sensing hypothesis by cbeck-ing a fa]sifiable prediction of it: Delayingthe formation of a quorum at a swarm'schosen nest site, wl\i]e lea\'ing the restof the decision-making process undis-turbed, will delay the swarm's flight tothe site. To delay quorum formation, weplaced five desirable nest boxes closetogetber at one location on tbe island.Tliis caused tbe scouts visiting tbe site to

Figure 4. Bees are labeled with a colored and numbered plastic tag affixed to the thorax and,in some cases, also a paint dot on the abdomen. These markings allow investigators to distin-guish between individual scout bees when observing a swarm's behavior.

www.americanscientist.org 2006 May-June 223

11:00-13:00bees: 18dances: 68waggle runs: 547

F=1

•^^ I \ D-3E=1 » \

B=4 \

C-1

7:00-9:00bees: 29dances: 53waggle runs: 868

K=1

B=12

13:00-15:00bees: 30dances: 70waggle runs: 2,376

H=1

B=1i\

C=2

July 21

9:00-11:00bees: 52dances: 99waggle runs: 1,697

K=3

July 20

A-6i>

15:00-17:00bees:dances:waggle runs:

G-10

11:00-11:54bees:dances:waggle runs:

K-2\

Ng

G=20 ^

38662.400

1=1

^ ^

B=17

(rain starts)2743713

B=4

A=5

17:00-19:00bees:dances:waggle runs

K— 1

G

:9—

= 9 ̂ PJ

• — •

9:00-11:58bees:dances:waggle runs

27 T45 /1,877/

J * L

B-13

0 1 2

kilometers

July 22

733523,100

0

Figure 5. This schematic summarizes a swarm's process of reaching a quorum for a nest site. Each panel summarizes a one- to three-hour interval ofactivity and lists the total number of bees, dances and waggle njns during this interval. The white circle represents the swarm. Candidate nest sites{black dots) are assigned lettere in the order in which the bees reported them. Each arrow indicates direction and distance to a site; the thickness of thearrow correlates with how many bees are dancing in support of that site during the interval, as shown by the number next to each site's letter designa-tion. In this case, the swarm considered a total of 11 sites over three days, but none was advertised much more strongly than the others during the firsthalf of the decision-making process. During the second half, however, site G gradually gained support and became fhe subject of all the dances.

be dispersed among five identical nestcavities rather than concentrated at one.We then saw how long it took a swarm,once it had discovered the site of thenest boxes, to make its decision and takeotf to fly to the site. We also performedwith each swarm another control trialwith just one nest box. The two trials foreach swarm were performed using twodifferent sites on the island, so each trialbegan in the same way, with one scoutbee discovering an attractive nest cavityin a new site. In all four swarms that wetested, there was indeed a marked de-lay to takeoff in the five-nest-box treat-ment (442 minutes on average) relativeto the one-nest-box treatment (196 min-utes on average). Thus this experimentyielded strong support for the quorum-sensing hypothesis.

Exactly how scout bees sense a quo-rum remains an enigma. They mayuse visual, olfactory or even tactile in-formation to assess the number of fel-

low scouts at a site, but this remains asubject for future study.

Once the quorum threshold is reachedat one of the sites, the bees start a behav-ior that is well understood. The scouts attliis site will return to the swarm clusterand begin to produce a special, high-pitched acoustical signal that stimulatesthe nonscouts in the swarm cluster tobegin warming their flight muscles, byshivering, to the 33 to 35 degrees Celsiusneeded for flight. In producing this sig-nal, which we call worker pipi)% a scoutscrambles through the swarm cluster,pausing every second or so to press herthorax against another bee and activateher wing muscles. Although most of thevibrational energy probably transfersdirectly into the contacted bee, this ac-tion prtxiuces an audible vibration thatis reminiscent of the re\'\'ing of a race-car engine. The piping signal lasts about0.8 seconds and has a fundamental fre-quency of about 200 hertz. Because the

stimulus for worker piping is a quorumof scouts at the chosen site, not a consen-sus among the scouts for this site, theprocess of swarm warming generallybegins before the scouts have reacheda consensus. But becatise the warm-upusually takes an hour or more, there isusually time for the scouts to achieve aconsensus for the chosen site before theentire swarm launches into flight.

Choosing the Best HomeBy eavesdropping on the decision mak-ing of swarms through observation ofthe dances of their scout bees, Lindaueriind our group have shown clearly that aswarm cluioses one nest site from an ar-ray of five or more alternatives. The nextquestion that naturally arises is whethera swarm chooses the best site, and if so,how? To assess the accuracy of nest-sitechoice by swarms, we presented swarmson Appledore Island with a five-alterna-tive choice in wliich four of the altema-

224 American Scientist, Volume 94

tives were mediocre nest sites and onewas a superb nest site. The four so-sonest boxes were attractive in all waysexcept that each box provided only 15liters of living space. The excellent nestbox was identical to the other four exceptthat it pro\ ided 40 liters of room, a vol-ume that better meets a colony's spaceneeds for its various activities (rearingbrood, storing fcxxi et cetera).

Nearly all of the test swarms chosethe excellent nest box. Specifically, weobserved that although the excellentsite was never the first one to be found,once a scout bee had discovered theprime site, the number of bees visit-ing this site rose more rapidly than atthe other sites and reached the quorumthreshold first. Moreover, as the numberof bees Increased at the excellent site, itdecreased at each of the mediocre sites,indicating that rising interest in the top-quality site depressed interest in theothers. This inhibition of buildup at thepoorer sites by buildup at the best siteis important, because it helps to ensurethat the quorum threshold is crossedfirst at the best site and to generate thepattern of consensus among dancersthat almost always appears shortly be-fore a swarm flies to its new home.

What are the behavioral mechanismsat the level of individual scout beesthat underlie these dynamics? One is

one nest box

Figure 6. Appledore Island, Maine, has few trees, ensuring that bees would focus on the pro-vided nest boxes. Shelters gave each box the same exposure to sun, wind and rain. The size ofthe nest-box cavify and its entrance opening could be adjusted to change its affractiveness tobees. Observers (here, Seeley) counted the scout bees that visited each of the nest boxes.

the scout bees' careful tuning of dancestrength, in terms of the number ofwaggle dance circuits they perform fora site, as a function of site quality. Westudied this phenomenon by presentinga swarm on Appledore Island with twonest boxes simultaneously, one excel-lent and one mediocre, and analyzing

the waggle dances for the two boxes asthey were performed side-by-side onthe swarm. We found that the first timea scout returns to the swarm from a first-rate site, she is apt to perform a waggledance containing 100 or more dance cir-cuits. Scouts also report mediocre butacceptable nest sites, presumably in case

five nest boxes

14:00 15:00 16:00 7:00

July 4, 2003 July 5, 2003

Figure 7. Results from one trial testing fhe quorum-sensing hypothesis for how scout bees know when the swarm has decided on a nest siteshow the effect of diluting the scout bees' attention with five identical nest boxes at the same site. In the one-nest-box scenario, the number ofscout bees at the site (top left) and the rate of waggle dancing on the swarm (middle left) rose rapidly, and piping signals alerting the swarm toready itself for takeoff became strong after only 79 minutes of dancing (bottom left). In the five-nest-box situation, the number of scout bees atany site did not rise above 10 because bees spread themselves across the identical boxes (top right), dancing at significant levels was prolonged(middle right), and piping became strong after 244 minutes of dancing (bottain right). For this swarm, the duration of dancing before takeoffwas 138 minutes and 277 minutes for the one- and five-nest box treatments, respectively.

www.americanscientist.org 2006 Mav-June 225

4

3

1 -

00.5 1.0 1.time (seconds)

30

nothing better is located. But the firsttime a scout returns from a so-so site,we found that she is likely to performa waggle dance contairung only a doz-en or so dance circuits. The greater thestrength of dancing for a particular site,the larger the stream of newcomers to it,hence the buildup of scouts will be mostrapid at the best site.

The difference in recruitment-signalstrength between scouts from excellentand mediocre sites is amplified by an-other curious feature of their behavior.If a scout bee commits herself to a site,she will make multiple visits to the site(probably both to show support for"her site" and to stay informed aboutthe buildup there of fellow scouts),and after each visit she will advertiseher site with a waggle dance. She will,however, decrease the strength of herdance advertisement by about 15 dancecircuits each time she returns to theswarm and performs a dance. TTie re-sult is that the tiverall difference in therecruitment-signal strength between twosites is a nearly exponential function ofthe difference in quality between thesites. If two bees advertising excellentand mediocre sites perform 90 and 30dance circuits, respectively, on their firstretum to the swarm, then the total dif-ference in their recruitment signal willnot be merely threefold, but sevenfold

Figure 8. Piping scout bees run ina random pattem over a swarm,stimulating worker bees to readyfor takeoff [top left). In these 15-second tracings, dots indicatepiping events and white boxesdenote the time until swarm lift-off. When a bee pipes, she pullsher wings together, presses herthorax onto the substrate andactivates her wing muscles toproduce a vibration (bottonr left).The scouts will pipe on a hive'scomb, but in a swarm, they al-most always pipe directly ontoanother bee. A sonogram showsthat the piping signal goes fromlow to high frequency and con-tains harmonics (top right).

(90+75-^60-^45+30-^15+0 = 315 circuitstotal versus 30+15+0 = 45 circuits total).Moreover, there is strong positive feed-back in this recruitment process, suchthat the greater the number of bees com-mitted to a site, the greater the numberof recruiters, which in tum gives rise to astill greater number of bees committed tothe site. Consequently, small differencesin nest-site quality and waggle-dancestrength between two sites can snowballinto large differences in the number ofscouts affiliated with these sites.

The differences in strength of wag-gle dancing and the positive feedbackinherent to this recruitment processexplain the variance in the number ofscouts committed to candidate sites,with the best site gaining scout beesthe fastest. But what causes the collapsein the number of supporters at inferiorsites as it balloons at a superior one?The fundamental basis for the drop inthe number of scout bees affiliated withinferior sites is the reality that all scouts,even ones that are committed to excel-lent sites, will eventually abandon theirsites. Usually, a bee ceases making visitsto a site shortly after she has ceased per-fonnijig dances for the site, hence beesabandon poor sites more rapidly thanthey do excellent ones.

Once a scout abandons a site, she"resets" and can be recruited to another

site, or even re-recniited to the same site.Howe\'er, when a bee finishes dancingfor a site, about 80 percent of tlie time shewill cease dancing entirely. Scout beestherefore depend on the recruitment ofother scouts who were unable to findany candidate sites on their searches andso remain uncommitted to iiny site. Butwhen a bee is recruited to visit a site, ifshe feels it is pix)r, she may not immedi-ately commit to the site by dancing for itupon her retum. An uncommitted scoutmay therefore visit several sites beforefinding one she feels is worthwhile.

As long as the rate of recruitment toa site exceeds the rate of abandonment,the number of scouts affiliated with thissite will grow. Eventually, however, therate of recruitment for the highest qual-ity site will snowball, at which time therate of recruitment for each inferior sitewill melt away: the pool of uncommit-ted scout bees is finite, and most are be-ing recruited to the best site. When therecruitment rate falls below the aban-donment rate at each inferior site, thenumber of scouts committed to thesesites starts to shrink. In short, as thegroup committed to the best site growslarge, it automatically excludes fromthe competition the groups affiliatedwith the inferior sites.

Mary R. Myerscough, a mathemati-cal biologist at the University of Sydney,has created mathematical models of thepopulation dynamics of scout bees per-forming dances for different nest sites.She has elegantly demonstrated that,given enough time, the dancing scoutsin a swarm will almost always becomefocused on the one best site that hasbeen found. This certainly matches whatLindauer and we witnessed in the scoutbees' debates: Almost always, a consen-sus among the dancers arises before theswarm takes off to fly to its home.

Although unanimity among thedancers shortly before takeoff is a con-spicuous feature of the dance recordsof swarms, we now understand thatreaching a quorum, not building a con-sensus, is the essence of the bee's groupdecision-making process. Nevertheless,we should not view the dancer consen-sus as an unimportant, incidental by-product of the bee's decision-makingprocess. On the contrary, consensus isnecessary for a swarm to perform a suc-cessful flight to its new home. Occasion-ally we have seen a swarm take off withthe scouts dancing strongly for multiplehome sites, and each time the airborneswarm has been unable to fly away.

226 American Scientist, Volume

g 15

10

:00 12:00 16:00 8:00 12:00 16:00 8:00 12:00 16:00 8:00 12:00 16:00 8:00 12:00 16:00

Figure 9. Do bees always select the best site out of the available choices? To find out, four mediocre boxes of 15-liter volume and one superbbox of 40-liter volume were arranged in a fan-shaped array so each was 250 meters from the swarm. The scout bees at each site were counted ev-ery 30 minutes. The superior site was not discovered first, but once it was located, interest in the best site grew rapidly and ultimately excludedattention to the other sites, thus a quorum was reached there first. Bees chose the best site in four out of five trials.

The mechanisms of swarm flight guid-ance remain pcxirly understood, btit itis clear from such observations that thesteering preKess depends on a sufficientnumber of scouts providing coherentdirectional information to the rest of theflying swarm bees. When a split deci-sion happens, the swarm seems to needto resettle and continue deliberationsuntil one site predominates.

Speed and Accuracy Trade-offA fundamental problem faced hy anydecision maker is finding a suitable com-promise between swift decisions andgood decisions. If an animal, or a group,must make a quick decision, it is sus-ceptible to making a poor one becauseit either cannot sample its options suf-ficiently broadly, cannot evaluate themsufficiently deeply or both. Assumingthat a honey bee swarm experiences sucha trade-off between speed and accuracyin choosing a nest site, we wonderedwhether the behavioral parameters ofthe bees' process of group decision mak-ing have been tuned by natural selec-tion so that a swarm incurs low timeand energy costs while minimizingits chances of choosing a poor site. Tosee whether this is the case, we built astochastic, discrete-time mathematicalmodel of the decision-making process ofswarms and then used our model to cre-ate "pseudomutant" swarms, ones withdifferent values for various behavioralparameters. This enabled us to see howincreases or decreases in particular pa-rameters affect the speed and accuracyof a swarm's choice of a home.

An obvious candidate parameter foralteration was quorum size, since quo-aim sensing lies at the heart of a swarm'sdecision making. When we varied thequorum size in the model, while hold-ing everything else at normal le\'els, tliemodel made it clear that a low quorum

yields relatively rapid but often inaccu-rate decisions and that a high quorumproduces slower but more accurate ded-sions. It was especially noteworthy thatthe model's prediction of the quorumsize that achieves a good balance be-tween speed and accuracy, some 15 to 20bees, essentially matches the empiricalfinding that scout bees initiate the pro-cess of swarm warming, in preparationfor takeoff, when the number of bees atone of the sites has reached 10 to 20,

We also examined one of the curi-ous features of scout bee behavior thatpresumably contributes to a swarm'sdecision making, namely the way that ascout reduces the strength of her danc-ing for a prospective nest site over re-peated visits to the site. It is striking thateach fime a scout visits a potential nestsite and then returns to the swarm clus-ter to advertise the site, she produces adance with fewer waggle dance circuitsthan before and so advocates for her siteless and less strongly. Varying the rateof dance-circuit reduction in our modelrevealed just how critical this factor isto the decision-ma king process. If thenumber of dance circuits is reduced ata rate faster than what is observed innature, then the time needed to reacha decision steadily increases, becausea rapid decay in the number of circuitsmakes it difficult for a swarm to reacha quorum at any one site. Conversely,if the number of circuits is reduced ata rate slower than what is obser\'ed innature, then an even greater problemarises: Swarm decision making fails al-together as split decisions (that is, quo-rums reached quickly at multiple sites)become common. Again, it is notewor-thy that our model's prediction of therate of dance-circuit reduction that pro-vides a good balance between speedand accuracy—15 to 20 dance circuitsper nest-site visit—essentially matches

the empirical finding that, on average,scout bees shorten their dances by 15dance circuits per visit to a nest site.

Given these findings about quorumsize, rate of dance-circuit reductionand other parameters, we concludethat the behavior of the scouts in hon-ey bee swarms has indeed been tunedby natural selection to create a groupdecision-making process with a favor-able balance between the competingdemands of speed and accuracy.

Swarm SmartsHeiuy David Thoreau lamented in oneof his journal entries from 1838 aboutthe difficulty that human groups ha\'e inachieving a collective intelligence: "The

lOOn

sixth trom lastreturn with dancing

end ofdancing

6 5 4 3 2 1 0remaining rettjrns

to swarm with dancing

Figure 10. Scout bees quickly and linearly de-crease their number of waggle-dance circuitsperformed for a site on successive returns tothe swarm. However, bees supporting excel-lent sites start with a larger number of dancecircuits, whereas bees that visited mediocresites start lower on the line. Hence scout beesgrade the strength of their recruitment inrelation to the quality of each site.

www.americanscientist,org 2006 May-June 227

10:00

13:00

16:00

Figure 11. Scout bees tune their strength of waggle dancing in relation to site quality, whichaccelerates the reaching of a quorum at the hest available site. Here, scouts locate two potentialnest sites, one with a large opening (left) and one with a more desirable small opening (right).Each scout bee then returns to the swarm (green arrows) and performs a waggle dance for hersite (top center), but the scout from the superior right tree performs more waggle dance circuits(red symbol) than the scout from the left tree (hhtc symbol). The result is that three hours later,the number of bees committed to the right tree has increased sixfold, whereas support for theleft tree has increased only threefold, and the majority of dances favor the right tree (middle).After three more hours, the number of scouts at the right tree has ballooned, and dances insupport of this site have excluded the left-tree site from the competition (bottom).

mass never comes up to the standardof its best member, but on the contrarydegrades itself to a level with the low-est." Likewise, Friedrich Nietzschewrote in Beyond Cood and Evil: "Mad-ness is the exception in indi\ iduals butthe rule in groups." Although it is truethat groups can make bad decisions, itis also the case that groups can makegood decisions. What are the circum-stances under which groups will behighly intelligent and able to act col-lectively to make good choices? Wesuggest that bees' nest-site selectionbehavior can provide guidance on thistopic, for it is clear they are successfulat making collective judgments.

The first relevant factor is that thescout bees are organized in a way thatpromotes diversity of knowledge with-in the group. Specifically, they are notled or dominated by a small number ofbees; instead, the decision-making pro-cess is broadly diffused among all thescout bees in a swarm. Consequently,a swarm's decision-making process isbased on the actions of hundreds of indi-viduals, each one an autonomous agentcapable of providing unique informa-tion for solving the house-hunting prob-lem. As an example, note how the beesaccomplish the first stage of their deci-sion-making task—uncovering the pos-sible altematives from which to choose.

Searching independently, widely andsimultaneously, the hundreds of scoutbees from a swarm bring back to thegroup diverse information—knowledgeof superb, mediocre and even lousysites—which can be shared with theother scouts by means of waggle dances.All discoveries of potential nest sites arefreely reported; no scout is stifled. Thus aswarm takes full advantage of its inher-ently collective nature to assemble ratherquickly—often in just a few hours—aprofusion of alternatives from whichto choose. The larger this set, the morelikely it includes a first-rate site. Thus,we see that one key feature of a swarm'sdecision making is its decentralized or-ganization, which helps ensure that ithas a broad set of options.

A second feature of the bees' behav-ior that promotes their collective intel-ligence is that the scouts show no ten-dency toward conformity or slavishimitation of others as they contribute tothe decision-making pRxess. We haveexplained that the heart of this prcKessis a competition among the various co-alitions of scouts affiliated with differentsites, each one vying to attract uncom-mitted scouts to her site. The membersof each coalition recruit additional mem-bers by performing waggle dances thatvary in strength in relation to site qual-ity, so that the higher the site quality,the stronger the waggle dance and thegreater the stream of newcomers. Whatis critically important here is that whenan uncommitted scout is recruited to asite, she does not blindly support thebee whose dance she followed. Instead,she examines the advertised site her-self, and only if she too judges it to bea worthy site does she perform a dancefor it and thereby recruit still more beesto the site. Through this independenceof opinions, the scouts avoid propagat-ing errors in the assessments of sites.Only at a truly good site will dancersattract more dancers, hence will therebe a strong addition to the number ofscout bees at the site. The net result isthat scout bees avoid mass maruas overpoor options.

The third key to the swarm's successis how the quorum-sensing process ag-gregates the diverse and independentopinions of the scouts in a way that bal-ances the competing needs of decision-making accuracy and speed. The quo-rum level is high enough that many beesmust independently assess a site's qual-ity before it is chosen. Quick selectionof a home based on only one or a few

228 American Scientist, Volume

bees' favorable assessments is not pos-sible. The quorum-sensing prtxiess filtersout extreme or inaccurate opinions andprovides a balanced, group-level asses.s-ment of the chosen site. This assessmentprocess takes time but ensures that thereis enough of an interval for true diver-sity of opinion to arise and for discov-ered sites to be independently e\'aluatedbefore one of them is chosen. Thus, thequonim-sensing method of aggregatingthe bees' information allows diversityand independence of opinion to thrive,but only long enough to ensure that adecision error is improbable.

These considerations illustrate howthe study of group decision making byhoney bees might help human groupsachieve collective intelligence and thusavoid collective folly. Good group de-cisions, the bees show us, can be fos-tered by endowing a group with threekey habits: structuring each delibera-tion as an open competition of ideas,promoting diversity of knowledge andindependence of opinions among agroup's members and aggregating theopinions in a way that meets time con-straints yet wisely exploits the breadthof knowledge within the group.

BibliographyBcekman, M., R. L. Pathke and T. D. Seeley.

20(16. How does an informed minorityof scouts guide a honey bee swarm as itflies to its new home? Animal Behaviour71:161-171.

Black, D. 1986. The Theory of CommiiWcs andElections. Dordrecht: Kluwer.

Camazine, S., P. K. Visschcr, J. Finiey and R. S.Vetter. 199*̂ . House-hunting by honey beeswarms; collfctivo decisions and individualbehaviors, /HSCL/CS Sociiiux 46:348-360.

Conradt, L., and T. J. Roper. 2005. Consensusdecision making in animals. Trends in Ecol-ogi/ and Ei'olntion 20:449-456.

Franks, N. R., S. C. Pratt, E. B. Malion, N. F.Britton and D. J. T. Sumpter. 2002. Informa-tion flow, opinion polling and collectiveintelligence Ln house-hunting social insects.Philosophical Transactions of the Roi/al Societyof London B337:\567-15H3.

Lindauer, M. 1955. Schwarmbienen auf Woh-nungssuche. Zeitschrifl fiir zvrgleichendePln/siobgie 37:263-324.

Myerscough, M. R. 201)3. Dancing for a deci-sion: A matrix model for nest-site choice byhoney bees. Proceedings ofthe Royal Society ofLondon B 270:577-582.

Passino, K. M., and T. D. Seeley 2006. Model-ing and analysis of nest-site selection byhoney bee swarms: The speed and accuracytrade-off. Behavioral Ecologu atid Sociohiotogy59:427-U2.

Seeley, T. D. 2003. Consensus building duringnest-site selection in honey bee swarms:

The expiration of dissent. Behm'ioral Ecologymid Sociobiology 53:417^24.

Seeley, T D., and S. C. Buhrman. 1999. Groupdecision making in swarms of honey bees.Behai'ioral Ecolog]/ and Sociobiology 45:19-31.

Seeley, T. D., and S. C. Buhrman. 2001. Nest-site selection in honey bees: How well doswarms implement the "best-of-N" deci-sion rule? Behavioral Ecologifand Sociohiology49:416-427.

Seeley, T D., and }. Tautz. 2(W1. Worker pipingin honey bee swarms and its role in prepar-ing for liftoff, journal of Comparative Physiol-ogy A \87:bb7-676.

Seeley T D., and P. K. Visscher. 2003. Choos-ing a home: How the scouts in a honey beeswarm perceive the completion of theirgroup decision making. Behavioral Ecologyand Sociobiohg}/ 54:511-520.

Seeley, T. D., and P. K. Visschcr. 2004. Quorumsensing during nest-site selection by honeybee swarms. Behaz'ioral Ecology and Soeiobiol-ogy 56:594-60\.

Surowiecki, J. 2004. Tlie Wisdom of Crowds. NewYork: Doubleday.

For relevant Web links, consult thisissue of American Scientist Online:

http://www.americanscientist.org/lssueTOC/issue/841

"Okay, first, I really need to know if you're dead.

www.america nscientist.org 2006 May-Jime 229