Collective Behaviour and Swarm Intelligence · Swarm Intelligence ? "September 2008 – MyFC...
Transcript of Collective Behaviour and Swarm Intelligence · Swarm Intelligence ? "September 2008 – MyFC...
Collective Behaviour and Swarm Intelligence
Jens Krause Humboldt University & IGB
Talk Structure
1. Part
Models of collective behaviour: predictions
Empirical tests: fish and humans
2. Part
Swarm intelligence
Applications
Mechanisms: How do animals group?
Iain Couzin
zx
y
zoa
zor
zoo
αº
(360 - α)º
Couzin, Krause, James, Ruxton & Franks 2002. JTB 218: 1-11
Metric modelsCouzin, Krause, James, Ruxton & Franks 2002.
JTB 218: 1-11Huth & Wissel 1992. JTB 156: 365-385
Topological modelsBallerini et al. 2008 PNAS 105: 1232-1237
Neurobiological modelsLemasson et al. 2009 JTB 261: 501-510
Predictive modelsMoussaïd et al. 2011
Split effect
CouzinCouzin & Krause 2003 Advances in the Study of Behavior 32: 1& Krause 2003 Advances in the Study of Behavior 32: 1--6767
Vacuolation
Advances in the Study of Behavior (2003) 32: 1Advances in the Study of Behavior (2003) 32: 1--6767
■■ 5-10 Individuals
How many leaders to guide a group?
00.10.20.30.40.50.60.70.80.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1ti f i f d i di id l (%)
00.10.20.30.40.50.60.70.80.9
1
0 5 10 15 20 25 30 35 40 45 50number of informed individuals
accu
racy
00.10.20.30.40.50.60.70.80.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1ti f i f d i di id l (%)
00.10.20.30.40.50.60.70.80.9
1
0 5 10 15 20 25 30 35 40 45 50number of informed individuals
accu
racy
aac
cura
cy
Couzin, Krause, Franks& Levin 2005 Nature 433: 513-516
proportion of informed individuals p
group size = 10
group size = 30
group size = 50
group size = 100
group size = 200
N = 10
N = 30
N = 50
N = 100
N = 200
Group sizeGroup size
1122
33
1616
44
101099
88
77
66
55
1414
1515
1313
1212
1111
11
AA
DDEE
CCBB
FFGG
HH
Experimental Tests: Human Groups
When experiment commences…….
Start walking at a normal speed
Remain within arm’s length of others and
Do not talk or gesture with each other.
Dyer et al. 2008, Animal Behaviour 75: 461-470
John Dyer
‘Stay within arm’s length of others’
‘Go to number X, without leaving the group’
Two private behavioural rules:
1122
33
1616
44
101099
88
77
66
55
1414
1515
1313
1212
1111
11
AA
DDEE
CC
BB
FF
GG
HH
Experimental Setup
Time and deviation from target are recorded.
Dyer et al. 2008, Animal Behaviour 75: 461-470
Human Groups
CologneCologne
■■ 200 People200 PeopleFreiburgFreiburg
■■ 100 People100 People
Group size = 20010 Leaders
Dyer et al. 2009 Phil Trans Roy Soc 364: 781-789
No Leaders → Torus
Dyer et al. 2009 Phil Trans Roy Soc 364: 781-789
Torus
Diameter fluctuates
Location keeps changing
Dyer et al. 2009 Phil Trans Roy Soc 364: 781-789
Conflict 10 vs 20
Dyer et al. 2009 Phil Trans Roy Soc 364: 781-789
Conclusions from experiments
Study of human crowd behaviour:
People being led without knowing that they were being led
Anyone with information can be a leader (self-organisation)
Only small number of leaders required
ApplicationsEvacuation of buildings and public squaresOrganisational principles for large crowd events
Robofish
Faria, Dyer, Holt, Waters, Ward, Clement, Goldthorpe, Couzin & Krause 2010. BES 64: 1211-1218
Research with a computer-controlled fish robot
Robotic fish (London/Exeter)
Robotic fish (MIT)
Robofish - Design
x
y
Faria et al. BES 64: 1211-1218
Decision-making in fish
Ashley WardAshley Ward
Decision-making in fish
8815193
9815192
125181
12110robofish
8421
Group sizePredation
91015192
6711181
11120robofish
8421
Group sizeFood
N=20
Ward, Sumpter, Couzin, Hart & Krause 2008. PNAS 105: 6948-6953
Fast & accurate decisions in groups
1, 2, 4, 8, 16 fish
Ward, Herbert-Read, Sumpter & Krause 2011. PNAS 108: 2312-2315
Circles: all fishCrosses: focal fish
Fast & accurate decisionsdecision accuracy
Ward, Herbert-Read, Sumpter & Krause 2011. PNAS 108: 2312-2315
Approach zoneDecision zone
Fast & accurate decisionsdecision time
Ward, Herbert-Read, Sumpter & Krause 2011. PNAS 108: 2312-2315
Fast & accurate decisionsshared vigilance
Approach zoneDecision zone
Ward, Herbert-Read, Sumpter & Krause 2011. PNAS 108: 2312-2315
Swarm Intelligence (SI)(animal and humans)
Krause, Ruxton & Krause 2010. Trends in Ecology and Evolution 25: 28-34
Main characteristics:1. Individuals independently acquire information
2. information is combined and processed through social interaction
3. a cognitive problem is solved in a way that cannot be implemented by isolated individuals.
• Naturally evolved SI (often used without realising it)
• Purposefully applied SI (biomimetics)− Evaluation of information contained in set of independent
individual estimates (“SI potential”)
− Utilisation of both independent individual estimates and social interaction
Swarm Intelligencein humans
Krause et al. 2010. Trends in Ecology and Evolution 25: 28-34
... ?... ?
Experiment (2008) at biomimetics-exhibition in Liebermannhaus, Brandenburg Gate Foundation, Berlin
Two Questions:
1.Estimate the number of marbles in a large glass jar
2.Estimate how many times a coin needs to be tossed for the probability that the coin will show heads each time to be as small as that of winning the German lotto
Swarm Intelligence Potential
Krause, Faria, James, Ruxton & Krause 2011 Animal Behaviour 81: 941-948
Swarm Intelligence Potential
Correctvalue = 562
Mean = 553.57Median = 516 Mean = 498.30
Median = 100Correctvalue = 24
Error of mean = 1.5% Error of mean = 1976%
Krause et al. 2011 Animal Behaviour 81: 941-948
Swarm Intelligence PotentialHow many estimates does it need to achieve a
prediction of a certain quality with high probability?
Swarm Intelligence PotentialHow many estimates does it need to achieve a
prediction of a certain quality with high probability?
Swarm intellgence potentialProbability of obtaining a good answer depends on number of estimates
Number of estimates
Prob.
• Electronic media− Access to knowledge, memory and creativity of huge
collections of people
− Interaction of large groups possible regardless of distances
• Expert knowledge can be insufficient in an increasingly complex and rapidly changing world
Companies and organisations that know how to harness SI are likely to have a competitive advantage
Why has SI becomesuch a hot topic?
Krause et al. 2010. Trends in Ecology and Evolution 25: 28-34
DemocratsDemocrats RepublicansRepublicans
2008 U.S. Presidential Election real2008 U.S. Presidential Election real--money vote share market money vote share market ~~~~ ~~~~
2006 2007 2008
Applications of SI
Source: http://iemweb.biz.uiowa.edu/graphs/graph_Pres08_VS_jpg.cfm
“Prediction markets”: Predicting future events by trading virtual shares.
Example: Iowa Electronic Markets (University of Iowa) http://www.biz.uiowa.edu/iem/index.cfm
Performance of Iowa Electronic Markets IEM:2008 U.S. Presidential Election Forecast
Source: http://tippie.uiowa.edu/iem/media/08pres.html
Applications of SI
User driven content
tapping intocollective creativity
Example: LEGO®
On an internet platform users can talk about design ideas, showcase their work and vote on each other’s designs.
Creative new designs that have the potential to reach the mass-market.
Goal
Applications of SI
Applications of SI
Collective management
Example: Ebbsfleet United football club owned by the members of MyFootballClubhttp://www.myfootballclub.co.uk/
Swarm Intelligence ?
"September 2008 – MyFC members voted 82% in favour of EUFC selling John Akinde to Bristol City for £150,000 (plus add-ons). This was a world's first, where the fans had the final say in a transfer deal. Over 7,000 members voted in 48 hours."
Swarm Intelligence
Does SI spell out the end of leadership in favour of socially self-organised systems?
Probably not...
• Many problems (including utilisation of SI itself) require expert knowledge
• Someone needs to take responsibility for decisions
SI will mainly be a tool that aids decision-making and the organisation of processes
Krause, Ruxton & Krause 2010. Trends in Ecology and Evolution 25: 28-34
Acknowledgements
Jolyon Faria
Darren CroftBen Chapman
Colin Tosh
Iain Couzin
Dean Waters
Ashley Ward
David Sumpter
Dick James
Graeme Ruxton
Lesley Morrell
ChristosIoannou
Stefan Krause
The Money:
The Brains:
The Media: Claudia Ruby, Jacob Kneser,Ranga Yogishwar, Ismeni Walter, Markus Peter
Stiftung Brandenburger Tor
John Dyer
Chantima Piyapong