"#$%&'()*+,-.'#/0+123-'4561-*6'
"#$%&&'$()*+,-.%/'01)
$%!&'(#)*&+'%,#-.,#/012-&+#34#5%2&.#
Key Concepts: Visual and Narrative Models of Important New Thinking
<-.-8&'9#72%#:-1&#90<0+-'#;'%!+&<9D#"%?#1-.#9;'&-,#
72-(#"%?#2->&#+&-'.&,#(%#-#;%;?+-(0%.#%:#1%=7%'C&'9D#
-.,#"%?#1-.#9&&#"%?'#1%<;-."#-9#%.&#;-'(#%:#-#;%;?=
FRAME: Complex Adaptive +-(0%.#%:#!?90.&99&9#-.,#1%.9?<&'9#(2-(#"%?#-,-;(#(%#
Systems
&>&.#720+&#(2&"#-,-;(#(%#"%?4#
)*+)7-716')8'19-'$)/-,'
Harnessing Complexity
#
++%70.8#09#-#+09(#-.,#!'0&:#,&91'0;(0%.#%:#(2&#1%<=
by Robert Axelrod and Michael D.
Cohen
2)70#"0%(1)
.&.(9#%:#(2&#<%,&+#
M>&'"#-8&.(#2-9#8%-+9#-.,#&<;+%"9#9('-(&80&9#(%#;?'=
9?&#(2&<4#).#&<;+%"&&#<082(#2&+;#-#1%=7%'C&'#0.#
)2(%$0)
(2&#2%;&#(2-(#(2&#1%=7%'C&'#70++#,%#(2&#9-<&#0.#'&=
#-8&.(#09#-.#0.,0>0,?-+#@-1(%'A#0.#-#1%<;+&*#&.>0=
(?'.4#L%<&%.&#0.#.&&,#%:#<%.&"#<082(#-9C#:'0&.,9#
Components
of the Model
.<&.(4#B%'<-++"#7&#(20.C#%:#;&%;+&#-9#-8&.(9D#!?(#
:%'#-#9<-++#+%-.4#E2&9&#-'&#9('-(&80&94#
Following
is a list and brief description of the components of the model
<0+0&9D#!?90.&99&9#-.,#8%>&'.<&.(9#1-.#-+9%#!&#
#
.90,&'&,#-8&.(94#E'&&9D#!0',9#-.,#0.9&1(9#-'&#9%<&#
An
Agent
E2&#9('-(&=
(2&#-8&.(9#0.#(2&#1%<;+&*#9"9(&<#7&#1-++#@-#:%'=
An
agent is an individual “actor” in a complex!"#!$%"&2*
environment. Normally we
80&9#(2-(#-.#
A4#).#-8&.(#2-9#(2&#-!0+0("#(%#0.(&'-1(#70(2#0(9#&.>0=
5+0/ and governments can
think
of
people
as
agents,
but
families,
businesses
-8&.(#?9&9#
.<&.(D#-.,#70(2#%(2&'#-8&.(9#0.#(2&#&.>0'%.<&.(4#
also
be considered agents. Trees, birds and insects are some of the agents
70++#12-.8&#
#-8&.(#1-.#'&9;%.,#(%#72-(#2-;;&.9#-'%?.,#0(#-.,#
!"#!$%"&2*#3&40&%$(
in%>&'#(0<&4#
the complex system we
call “a forest”. An agent has the ability to
.#,%#(20.89#<%'&#%'#+&99#;?';%9&:?++"4#F)#'%1CD#
interact
with
its
environment,
and with other agents in the environment.
J.&#9%?'1&#
.D#7%?+,#.%(#G?-+0:"#-9#-.#-8&.(#0.#(2&#:%'&9(#&1%=
An
agent can respond to what happens around it and can do things more
%:#12-.8&#09#
(&<#!&1-?9&#0(#1-..%(#'&9;%.,#(%#72-(#2-;;&.9#
or(2&#-8&.(N9#
less purposefully. (A rock, then, would not qualify as an agent in the
?.,#0(4H#
forest
ecosystem because it cannot respond to 6%0*.4%#+,
what happens around it.)
?.,&'9(-.,=
3.77%**
0.8#-!%?(#
31-%)+4)2(%$0)
A2%7#7&++#-#;-'(01?+-'#9('-(&8"#09#7%'C0.84#O:#-#120+,#
Type of Agent
&#1-.#,0::&'&.(0-(&#!&(7&&.#@(";&9A#%:#-8&.(9#?90.8# We
can differentiate between “types” of agents using a variety of criteria.
+$&7$89)&4*0'*$3":$;<.'=.'>$
+&-'.9#(2-(#720.0.8#.&>&'#8&(9#20<#72-(#2&#7-.(9D#2&#
-'0&("#%:#1'0(&'0-4#I&#<082(#7-.(#(%#,0::&'&.(0-(&#
We might want to differentiate schoolchildren into “shy” and “aggressive”
70++#F&>&.(?-++"H#9&-'12#:%'#%(2&'#9('-(&80&9#(%#-((-0.#
2%%+120+,'&.#0.(%#@92"A#-.,#@-88'&990>&A#(";&94#J'#
types. Or we may prefer to categorize them by hair color, or reading skill,
209#8%-+94#)8&.(9#-+9%#+&-'.#.&7#9('-(&80&9#!"#1%;"=
#<-"#;'&:&'#(%#1-(&8%'0K&#(2&<#!"#2-0'#1%+%'D#%'#
or athletic ability. An agent is part of the same population if it can use the
0.8#(2&#9?11&99:?+#9('-(&80&9#(2-(#(2&"#%!9&'>&#0.#%(2=
,0.8#9C0++D#%'#-(2+&(01#
same
strategy that is being employed by another agent.
-+)./0&1+"
&'94#
+0("4#).#-8&.(#09#;-'(#%:#
+,#!$%"&*
#9-<&#;%;?+-(0%.#0:#0(#
A# Population of Agents
8"#'"0'+$)
.#?9&#(2&#9-<&#9('-(&8"#
The
concept of a population or collection of agents is important for three
)9#9('-(&80&9#-'&#1%;0&,#-.,#92-'&,#(2'%?82#-#;%;?=
(#09#!&0.8#&<;+%"&,#!"# !"#!$%"&
reasons.
First, by being part of a population, an individual (agent)
<-.-8&'9#72%#:-1&#90<0+-'#;'%!+&<9D#"%?#1-.#9;'&-,#
+-(0%.D#9%<&#>-'0-(0%.#70++#%11?'4#P-'0-(0%.#9&'>&9#(%#
%(2&'#-8&.(4# 72-(#"%?#2->&#+&-'.&,#(%#-#;%;?+-(0%.#%:#1%=7%'C&'9D#
increases
the number of possibilities for learning new and successful
0.1'&-9&#(2&#,0>&'90("#%:#(2&#;%;?+-(0%.#%:#9('-(&80&94#
strategies.
Second, a population can serve as the recipient for a newfound
-.,#"%?#1-.#9&&#"%?'#1%<;-."#-9#%.&#;-'(#%:#-#;%;?=
P-'0-(0%.#1-.#Third, the population serves as part of the environment in
'()%*#+,
improvement.
5+-6."0'+$)+4)2(%$0&)
+-(0%.#%:#!?90.&99&9#-.,#1%.9?<&'9#(2-(#"%?#-,-;(#(%#
!$%"&*
%11?'#(2'%?82#
which
an agent lives. For example, as a business manager, you can learn
&#1%.1&;(#%:#-#;%;?+-=
&>&.#720+&#(2&"#-,-;(#(%#"%?4#
<?(-(0%.#
from
the population of managers who face similar problems, you can
.#%'#1%++&1(0%.#%:#
#
1%<=
F72&'&#;-'(#%:#
80410&1+"of co-workers, and you can
spread
what you have learned to a population
&.(9#09#0<;%'(-.(#:%'#
2)70#"0%(1)
(2&#9('-(&8"#09#
see
your company as one part of a population of businesses and
&&#'&-9%.94#60'9(D#!"#
M>&'"#-8&.(#2-9#8%-+9#-.,#&<;+%"9#9('-(&80&9#(%#;?'=
12-.8&,#!&=
consumers
that you adapt to even while they adapt to you.
.8#;-'(#%:#-#;%;?+-(0%.D#-.#0.,0>0,?-+#F-8&.(H#0.=
9?&#(2&<4#).#&<;+%"&&#<082(#2&+;#-#1%=7%'C&'#0.#
1-?9&#%:#&'=
-9&9#(2&#.?<!&'#%:#;%990!0+0(0&9#:%'#+&-'.0.8#.&7#
(2&#2%;&#(2-(#(2&#1%=7%'C&'#70++#,%#(2&#9-<&#0.#'&=
#&.>0=
A'%'HD#<08'-=
Strategy
,#9?11&99:?+#9('-(&80&94#L&1%.,D#-#;%;?+-(0%.#1-.#
(?'.4#L%<&%.&#0.#.&&,#%:#<%.&"#<082(#-9C#:'0&.,9#
(9D#!?(#
Every
agent has goals and employs strategies to pursue them. An
(0%.#F72&'&#
>&#-9#(2&#'&10;0&.(#:%'#-#.&7:%?.,#0<;'%>&<&.(4#
:%'#-#9<-++#+%-.4#E2&9&#-'&#9('-(&80&94#
#!&#
.&7#-8&.(9# employee might help a co-worker in the hope that the co0',D#(2&#;%;?+-(0%.#9&'>&9#-9#;-'(#%:#(2&#&.>0'%.=
#
#9%<&#
worker will do the same in return. Someone in need of
70(2#.&7#
.(#0.#72012#-.#-8&.(#+0>&94#6%'#&*-<;+&D#-9#-#!?90=
E2&#9('-(&=
money might ask friends for a small loan. These are
#:%'=
9('-(&80&9#
99#<-.-8&'D#"%?#1-.#+&-'.#:'%<#(2&#;%;?+-(0%.#%:#
!"#!$%"&2*
80&9#(2-(#-.#
strategies.
(9#&.>0=
5+0/
-8&.(#?9&9#
<&.(4#
70++#12-.8&#
The strategies that an agent uses will change over time.
#0(#-.,#
!"#!$%"&2*#3&40&%$(
%>&'#(0<&4#
One source of change is the agent’s understanding about
1CD#
how well a particular strategy is working. If a child learns
J.&#9%?'1&#
9(#&1%=
that whining never gets him what he wants, he will
%:#12-.8&#09#
;&.9#
(eventually) search for other strategies to attain his goals.
(2&#-8&.(N9#
6%0*.4%#+,
Agents also learn new strategies by copying the
?.,&'9(-.,=
3.77%**
successful strategies that they observe in others.
0.8#-!%?(#
2%7#7&++#-#;-'(01?+-'#9('-(&8"#09#7%'C0.84#O:#-#120+,#
9#?90.8#
+&-'.9#(2-(#720.0.8#.&>&'#8&(9#20<#72-(#2&#7-.(9D#2&#
.(0-(&#
70++#F&>&.(?-++"H#9&-'12#:%'#%(2&'#9('-(&80&9#(%#-((-0.#
;&94#J'#
209#8%-+94#)8&.(9#-+9%#+&-'.#.&7#9('-(&80&9#!"#1%;"=
'D#%'#
0.8#(2&#9?11&99:?+#9('-(&80&9#(2-(#(2&"#%!9&'>&#0.#%(2=
&'94#
#
8"#'"0'+$)
)9#9('-(&80&9#-'&#1%;0&,#-.,#92-'&,#(2'%?82#-#;%;?=
-8&.(9#?90.8#
:&'&.(0-(&#
&A#(";&94#J'#
1%+%'D#%'#
0&1+"
"&*
'()%*#+,
!$%"&*
-8&.(H#0.=
'.0.8#.&7#
+-(0%.#1-.#
'%>&<&.(4#
#&.>0'%.=
&D#-9#-#!?90=
?+-(0%.#%:#
0.8#-!%?(#
2%7#7&++#-#;-'(01?+-'#9('-(&8"#09#7%'C0.84#O:#-#120+,#
+&-'.9#(2-(#720.0.8#.&>&'#8&(9#20<#72-(#2&#7-.(9D#2&#
.'/,#"0-)*+,%&1"-!23%"/!"#$%&'("%()$%*"*+,-(&"'./%"0%()0"+1)%2"!3&'-(&"'%
70++#F&>&.(?-++"H#9&-'12#:%'#%(2&'#9('-(&80&9#(%#-((-0.#
B%>696($!?%&'2,+8$6%"'$%"0%!"0$%*"*+,-(&"'6%"7%
45)$0$%(5"%"0%!"0$%6+22$667+,%6(0-($1&$6%-0$%3,$'8$8%
209#8%-+94#)8&.(9#-+9%#+&-'.#.&7#9('-(&80&9#!"#1%;"=
-1$'(6%-'8%-,,%"7%()$%6(0-($1&$6%()-(%()"6$%-1$'(6%$!=
Key
Concepts:
Visual
and
Narrative
Models
of
Important
New Thinking
("%20$-($%-%'$5%6(0-($19.:%
0.8#(2&#9?11&99:?+#9('-(&80&9#(2-(#(2&"#%!9&'>&#0.#%(2=
*,"9:%B%>2"!*,$D?%696($!%&6%"'$%&'%5)&2)%()$%-2=
%
&'94#
(&"'6%"7%-1$'(6%-0$%(&$8%#$09%2,"6$,9%("%()$%-2(&"'6%"7%
!"#"$%&'(%
#
"()$0%-1$'(6%&'%()$%696($!:%;)$'%()$%-1$'(6%&'%-%696=
8"#'"0'+$) ;&()%-,,%"7%()$%8&#$06&(9%20$-($8%39%#-0&-(&"'/%-1$'(6%
($!%-0$%-2(&#$,9%(09&'1%("%&!*0"#$%()$!6$,#$6%
5&,,%3$%7"02$8%("%2)""6$%6(0-($1&$6%()-(%-**$-0%("%3$%
)9#9('-(&80&9#-'&#1%;0&,#-.,#92-'&,#(2'%?82#-#;%;?=
4>-8-*(?./%()$'%()$%696($!%&6%-%E"!*,$D%B8-*(&#$%
6+22$667+,:%<)&6%5&,,%!$-'%()-(%,$66%6+22$667+,%6(0-($=
+-(0%.D#9%<&#>-'0-(0%.#70++#%11?'4#P-'0-(0%.#9&'>&9#(%#
@96($!:%
1&$6%5&,,%>8&$%"+(?%-6%7$5$0%-'8%7$5$0%-1$'(6%2)""6$%
0.1'&-9&#(2&#,0>&'90("#%:#(2&#;%;?+-(0%.#%:#9('-(&80&94#
Variation%
()$!:%<)&6%&6%-%7"0!%"7%>6$,$2(&"'?%39%5)&2)%6+22$66=
P-'0-(0%.#1-.#
)(-41"51&"6-'7-%8"-.'/,#"0-)*+,%&1"-!23%"/3As strategies
are copied and shared through a population,
7+,%6(0-($1&$6%-0$%2"*&$8%-'8%$!*,"9$8%5)&,$%,$66%
%11?'#(2'%?82#
95+/"6'5:some variation
will occur. Variation serves to increase the
<?(-(0%.# 6+22$667+,%6(0-($1&$6%-0$%,$7(%("%8&$%"+(:%@$,$2(&"'%
>!"#$%&/%"7%-%'()*#%+,"7%%+-#&/%+6$%()$&0%&%)(%#"*#&/%
(-A$6%*,-2$%()0"+1)%$&()$0%(0&-,=-'8=$00"0%"0%&!&(-(&"'%
F72&'&#;-'(#%:#
80410&1+"
diversity &'%*-(($0'$8%*$%#)(.%*/$/%5&()%$-2)%"()$0%-'8%5&()%
of the population of strategies. Variation can occur
"7%()$%6+22$667+,%6(0-($1&$6%$!*,"9$8%39%"()$0% through mutation (where part of the strategy is changed
(2&#9('-(&8"#09#
-0(&7-2(6:%0#)1/)2($.#,2#(&3)#&,"'%()$%0$6+,(&'1%
12-.8&,#!&=-1$'(6:%<)&6%($'86%("%0$8+2$%#-0&-(&"':%
because of
error), migration (where new agents with new
$#$'(6%80&#$%()$%&#4#.%*/$,"7%-1$'(6%-'8F"0%6(0-($1&$6%
1-?9&#%:#&'=%
strategies
move
into the population), or through combination
()0"+1)%*0"2$66$6%"7%$00"0=*0"'$%./-+*$",-'8%)#5
'%'HD#<08'-=%
(where
two
or
more
successful strategies are blended to create
./26*$(%*/$/%()+6%2)-'1&'1%()$%70$G+$'2&$6%"7%()$%
(0%.#F72&'&#)*+,%+%&'((9*$6%5&()&'%()$%&+&%#2:?%
a
new
strategy).
;)$'%()$%6$,$2(&"'%*0"2$66%,$-86%("%-'%&!*0"#$!$'(%
.&7#-8&.(9#
70(2#.&7# -22"08&'1%("%6"!$%!$-6+0$%"7%6+22$66%41"-,./%5$%2-,,%
Selection
9('-(&80&9# ()&6%-8-*(-(&"':%B8-*(-(&"'%&'20$-6$6%()$%*$07"0!=
-'2$%"7%()$%&'8&#&8+-,%-1$'(%-22"08&'1%("%&(6%"5'%
With all of the diversity created by variation, agents will be
6+22$66%20&($0&-:%C$2-+6$%()$%6+22$66%20&($0&-%!&1)(%
forced to choose strategies that appear to be successful. This
3$%8&77$0$'(%7"0%8&77$0$'(%-1$'(6/%-%*-0(&2+,-0%6$,$2(&"'%
will mean that less successful strategies will “die out” as fewer
"0%2)-'1$%!&1)(%3$%-'%-8-*(-(&"'%7"0%6"!$%-1$'(6%
and fewer agents choose them. This is a form of “selection” by
3+(%'"(%7"0%"()$06:%
which successful strategies are copied and employed while
less successful strategies are left to die out. Selection takes
place through either trial-and-error or imitation of the
successful strategies employed by other agents. This tends to
reduce variation.
%
Adaptation
When the selection process leads to an improvement
according to some measure of success (goal), we call this
!"#"$%&'()*(+),+*-%*%&'(
adaptation. Adaptation increases the performance of the
individual agent according to its own success criteria.
Because the success criteria might be different for different
agents, a particular selection or change might be an adaptation
for some agents but not for others.
Complex Adaptive System
A “system” includes one or more populations of agents and all of the
strategies that those agents employ. A “complex” system is one in which
the actions of agents are tied very closely to the actions of other agents in
the system. When the agents in a system are actively trying to improve
themselves (“adapt”), then the system is a Complex Adaptive System.
An Overview of the Complex Adaptive Systems Framework “Agents, of a
variety of types, use their strategies, in patterned interaction, with each
other and with artifacts. Performance measures on the resulting events
drive the selection of agents and/or strategies through processes of errorprone copying and re-combination, thus changing the frequencies of the
types within the system.”
For more information contact:
257 Castle Glen Road
Walnut Creek CA 94595-2642
(925) 934-1786