This
article summarizes Stuart Kauffmans book At Home in the Universe.
All of the principles that are discussed concerning complex systems are
his, or if attributed to others, may be found in his book. The conclusions
drawn with respect to venture management and transition management in
organizations are a mixture of mine and Dr. Kauffmans. The models
from the Taylor Modeling Language are copyrighted, of course, by MG Taylor
Corporation.
Related Articles:
Part I: Selection, Self-Organization,
and Autocatalysis
Part II: Requirements
for Order and Evolution on Fitness Landscapes
Coevolution
and Coupled Fitness Landscapes
"We all swap our stuff
In doing so, we create
niches for each other."
Now that the organizations made it to the end game
perched smugly on its peak, nature raises the stakes. No enterprise is
an island. Species in ecosystems influence the evolution of other species.
They all coevolve. Their fitness landscapes are linked. Therefore, a change
in overall fitness by one species on its landscape may deform the peaks
on the landscape of a different species.
A few notes here about building a community. We know from
reading Kellys book Out of Control that the composition
of stable communities is highly dependent upon the order in which species
are introduced. We also know that when building a community, at first
its easy to add species but this gets harder and harder as the community
approaches stability. This is more than the food web and who eats who.
Much of an ecosystem is caught up in cooperationmutualism and symbiosis.
Of course, there are host-parasite and predator-prey links as well.
And
we also know that if we remove species at random, the cascade of changes
through the community will follow the form of a power law curve. In other
words, in a few instances, the removal of a species will be catastrophic,
but most of the time the removal of a species will cause a little cascade
of change that quickly dampens out.
It's easy to understand the linkage between species through
a predator-prey model. Imagine a valley of grass inhabited by mice and
lions (OK, well, I drew a mouse and a lion: forget about whether it's
REAL or not ;-). The mice eat the grass and the lions eat the mice. As
the population of mice increases, they provide more food for the lions,
who use the extra energy to breed more lions. More lioins eat more mice
and the population of mice begins to decline. Fewer mice cause the lions
to die out and the overlapping cycles repeat.
|
However, the fitness landscapes of
the mice and the lions are coupled together. If the mice mutate
to develop better cornering ability to outmaneuver the lions, then
the lions' fitness landscape will deform--what was once a peak now
becomes a valley.
Perhaps now the lions develop a coloring thats
harder to see. Now the mice are in trouble and their new peak
sinks to become a valley from which they must emerge.
To emerge from the valley, they'll begin with
wild jumps around their new landscape to find the areas with higher
peaks. Once near a peak, the jumps become shorter and once the
new peak is gained the jumps will cease altogether--unless the
lions have adapted in the meantime.
|
Again, this relationship can be modeled using
the NK network (see Part
I) model with a few variations. Each fitness
landscape represents a species, and each node on the network under the
landscape represents a trait. N is the number of traits per species, represented
as nodes on the network. K is the number of couplings between traits in
a species. C, a new variable, is the number of links to traits in other
species that influence the behavior of the node. K represents internal
connections and C represents connections between species. Theres
a fourth variable, S, which represents the number of species that are
interacting.
N = number of
nodes or traits (genetic possibilities) on the target species' fitness
landscape
K = number of
inputs to each node from within the target species
C = number of
connections to each node in the target species from traits of other
species that influence the behavior of the node
S = number of
other species interacting with the target species
Different types of behavior emerge while running the model.
If K is high, then there are lots of
peaks on a species' landscape. It doesnt much matter what other
species do because if a species gets bumped off of a peak, its
a short climb back up. This is an ordered regime for coevolution and
one of average low fitness as well.
Or if C is low, there are few connections
between species. As a result, each species climbs to its nearest peak
and settles down there. The sparcity of interconnection means that species
will rarely bump each other off of peaks. It's as if the different species
contribute nothing to one another's well being. There is too much order
in this regime to permit further coevolution.
In both of the two previous examples, a species can find
an evolutionary stable strategy in the ordered regime, moving to local
peaks and remaining there. If you perturb a system in the ordered regime,
the disturbance is quickly damped out.
If K is low, there are few peaks to
get trapped on but it takes a long time to climb to the top. So, even
with modest values of C or S, the
species will keep trying to climb peaks that move too fast to attain.
This is known as the Red Queen regime, or the arms race
syndrome. Species are so tightly linked that they engage in a never-ending,
chaotic war to outdo each other. If a system is perturbed in this chaotic
regime, the changes amplify and cascade in waves throughout the ecosystem.
This is the reverse of the power law.
In each of the three previous examples, the average level
of fitness attained by any species is relatively low. Either there are
a lot of modest peaks that are attained quickly and then held, or the
landscape heaves beneath the species so that it can never reach a peak
at all.
There's a fourth option. If the simulation of coupled
fitness landscapes is run long enough, a more interesting phenomenon emerges.
Red Queen regimes migrate toward the more ordered regimes and the evolutionary
stable strategies migrate upward toward the more chaotic regimes. They
meet in the middle, around the complex phase transition once again. It
would seem that nature abhors too much chaos or too much order. In effect,
the ecosystem manages the number of connections between species, between
traits within a species and the number of species involved.
IMPLICATIONS FOR TRANSITION MANAGEMENT
Industries can be examined to discern the relative sizes
of their K, C and S
values. If theres too much interconnectivity between organizational
species, or between divisions and departments, a Red Queen war can break
out. If there is not enough connection, then the organization or division
will stagnate. Stagnation can be programmed if evolutionary stable strategies
are desired for a period of time.
Over time, all industries naturally move away from overly
chaotic or ordered regimes. If an organization comes to rely too much
on complacency or aggressiveness for its success, it will be in trouble.
Assuming a Red Queen strategy on a more docile landscape causes the
organization to disperse away from local peaks (the self-imposed mutation
rate is too high) instead of coalescing. Assuming an evolutionary stable
strategy in a more agitated environment will leave the organization
constantly inhabiting the basins.
Solving
Hard, Conflict-laden Problems with Patch Theory
"As we approach the year 2000, the design of complex
artifacts is plagued with nearly unsolvable conflicting restraints
Organisms, artifacts and organizations all evolve and coevolve on rugged,
deforming fitness landscapes. How do we track peaks on deforming landscapes?"
This is all fine, but in the real world of enterprises,
the values of K, C or S
tend to be high. We need strategies for handling the dense interconnection
and the inevitable conflict such connection brings. Kauffman uses the
example of designing a modern jet. Everything seems to affect everything
else in the design. The size and type of engines affects the size of the
fuel tanks, affects the design of the wings or the center of gravity,
and so on. The traditional method for solving such complex problems requires
some of the variables in the equation to become constants. For instance,
the plane may be designed around the engines, the organization may be
designed around one particular product line or department. It's tempting
to overlay a hierarchy or prioritization scheme over the elements of the
problem to force an artificial resolution of the conflict.
Could there be another way of resolving conflict between
components in design and implementation besides imposing a bureaucratic
hierarchy on the system? Maybe.
Well start with the familiar NK network. Lets
say that a problem has N parts and K inputs to each part (connections
between parts). This system is going to navigate on a fitness landscape.
The diagram shows 36 components to the problem with each component receiving
four inputs which influence its next state, or decision, or design iteration.
For now, well treat the problem/system as one patch.
This means that a part can mutate (change its state, design or answer
to its part of the problem) only if the mutation is good for the entire
system; if it moves the entire system up in the fitness scale. Most mutations
will fail the criteria because of the dense interconnnectivity and the
dampening of changes in the solution that attempt to move through the
system. A part can't adopt a new solution unless it makes everyone else
in the system happy. Therefore, the entire system will move slowly up
to the nearest local peak and squat there. Once on the peak, all changes
would take it down one of the slopes. Since this violates the criteria
of accepting only increased fitness, there will be no movement off of
that local peak. Chances are that local peak is at a relatively low level
of fitness as well. This example illustrates the central government strategy
and the total consensus strategy. No one can change anything unless you
can prove it is good for everyone else. Once a local peak is attained,
further evolution is impossible and the system is merely waiting to be
fractured and destroyed by the competition and other forces in the ecosystem.
Even if you can see a better solution you can't get there from here.
Suppose we now superimpose a patchwork grid over the network.
Were not going to disturb any of the N nodes or
the K connections between nodes. But were going
to overlay a grid of patches and now, if a node, N, mutates,
it can do so if that increases the fitness for its patch. It no longer
has to worry about increasing the fitness of the entire system. But since
the nodes are still all interconnected, that accepted mutation may propagate
across other patches and influence them. In other words, it changes the
problem that those other patches have to solve and now theres a
chance for evolution. An improvement in fitness for one patch may mean
a degradation in fitness for the system as a whole. Now the system can
move up and down on the fitness landscape.
What if we divide the system into N patches, where each
node is its own patch? This leads to chaos and the Red Queen regime where
each node competes viciously against all of the nodes its connected
to. No one gets anywhere. This is an example of total participative management
and decision making.
What if the number of patches is something in-between?
The system encounters a phase transition between chaos and order and just
on the chaotic side of the boundary we find our comfortable land of complexity
where theres enough order to get the work done but enough variability
to allow evolution as well.
IMPLICATIONS FOR TRANSITION MANAGEMENT
Patch theory is different than conventional management
theory where you break problems down into departments and then manage
the departments. Patch theory says you break problems down into departments
and then let them coevolve a solution. Traditional management theory
was on the right track. Organizations were divided into departments
or divisions, each with its own unique but interconnected part of the
problem. Where management theory went wrong was in severing the connections
between nodes and wiring the nodes in each patch up to a single "manager"
node through which all decisions and state change communication had
to pass. This left the door wide open for Red Queen wars between divisions.
It also reduced the efficiency of the organization--especially in times
of rapid change--because most of the nodes in the enterprise had lost
their connections to other patches. To overcome this limitation, people
organized informal "grapevine" structures so that much of
the real work could get done in spite of the formal structure.
If a solution does not coevolve, then one part of the
problem will become fixed, and the rest of the elements will have to
fall into line based on the ramifications of its specifications.
If an enterprise employs too many patchesif members
of the network are too autonomousRed Queen behavior and internal
arms races may emerge. Many sales organizations suffer from this phenomenon
where each salesperson tries to negotiate for a better deal with the
company than his associates. Many companies have vicious internal politics
that resemble this type of behavior as well.
When applying patch theory and coevolution, the first
step is to break the problem into patches. Ensure that the connections
between network nodes remain intact. The underlying network structure
must remain whole. If, instead, the connections get severed, each patch
either acts as an autonomous unit, or as a node. This risks Red Queen
behavior.
For an example of how we use Patch Theory in DesignShop®
events, click here.
Conclusion:
Implications for Venture Management
The following ideas summarize the conclusions from each
of the preceding sections in the light of the venture.
Selection
Selection is insufficient of itself to propel evolution.
We cannot rely solely on the random selection of people, ideas or artifacts
in or out of our organization as a mechanism to further our growth.
Autocatalysis
There are really no measurements we can make concerning
the complexity of our organization. The science has not become that
cohesive or precise. But we can think in general ways about the diversity
of the people, ideas and artifacts in our organization and the probability
that any or all of these actually act as catalysts. We can imagine whether
our organization is indeed autocatalytic or not. I suspect that it is
not yet. The knOwhere® stores are our next major test of an autocatalytic
system.
I also suspect that for many small enterprises there
is a tendency toward less diversity and more probability of catalysis.
In other words, people, ideas and artifacts must facilitate each others
work more in a small company than in a large one. The economics for
having great diversity in a small company are just not there. Or if
the diversity is present, it tends to cluster in certain individuals,
making them cross-functional and many-skilled.
We also need to be aware if the catalysis that we have
going on in MG Taylor is truly feeding back on itself and converging
on autocatalysis, instead of diverging its energy. We may be catalyzing
the production of lots of new "molecules" but are these pathways
linked together in a net or do they dissipate and straggle off? These
questions are related to the policy, logistics and task stages of our
Vantage Points model.
Requirements for Order in Emergent Systems
Since we are a highly networked organization, we should
pay attention to the degree that our network is interconnected in a
causal way. This is tuning the K variable. Naturally we all communicate
with one another but what is the nature of the catalytic connections
that lead to decisions? Im not sure myself how to think about
this but its worth some effort. If there are too many conflicting
messages in the system, it will oscillate in the chaotic regime of state
space. If we put too much control on the system, it will settle down
into the ordered domain out of which we will find it difficult to evolve.
If we choose to operate with a higher K value then we
must tune the P value. Recall that P regulates the way nodes make decisions
by canalizing the decision making function or rules. This means that
regardless of the variability of input, the output of the nodes is fairly
regular. Only a few out of many inputs play a large role in determining
the output. A commander on the battlefield receives a great many inputs
for decision making. In his synthesis, usually a few of the inputs can
determine the outcome: an order from a higher command, a threat that
must be countered immediately, a break in the weather pattern, etc.
Decisions come down to paying attention to a few of the variables while
not ignoring any of them. This is a strategy for resolving conflict.
The Evolution of Living Systems on Correlated Fitness
Landscapes
There must be a level of redundancy in the system to
stabilize its chances for success. This is particularly difficult in
a small enterprise.
We need to use policy and task level tools to help restrict
conflict if the K value is too high, creating a landscape with too many
peaks. If we find ourselves quickly scaling peaks of mediocre fitness,
we may have too high a K value.
Trying to mutate to new, innovative ideas too rapidly
and in too rapid succession can cause the organization to lose focus
and wander all over the fitness landscape.
The way to explore unmapped areas of terrain is to "mate"
with a partner and watch the fitness of the offspring that necessarily
occupies terrain in-between the parents.
If our N and K values become very high, we will be unable
to make radical improvements. The complexity will tie us down and well
have to resort to spin-offs and other mechanisms to inject the necessary
growth.
As we enter new arenas of the economy we should be aware
of what games are being played there and what stages the games are in.
If the game is new, we use long jump strategy to quickly seek more favorable
position. This means trying out many innovations covering a broad arena.
If the game is a middle game, we should not enter unless we can see
a way through to making it a new game. If we are in the middle game
or enter in a middle game in progress the strategy is cautious improvement
step by step until we occupy a local peak. If were in the end
game, were on a local peak and should stay there.
Coevolution and Coupled Fitness Landscapes
We know that building community is easy at the beginning
when everythings just simmering in the pot. As things crystallize
out, it becomes harder for new ideas to take shape in the matrix.
We may find ourselves in one of three scenarios: playing
the Red Queen arms race, stalled in evolutionary stable system, or on
the complex edge of chaos where we have a stable ability to produce
and an ability to evolve as the landscape deforms beneath us. If were
playing the Red Queen scenario, then we may have scant internal connections
but be over-connected to other species in our ecosystem. If were
playing the evolutionary stable system scenario, we may be over-connected
internally but not have enough connections to other species. We can
also tell if were in either of those scenarios if our overall
level of fitness seems mediocre.
Solving hard, Conflict-laden Problems with Patch Theory
Even if we are organized as a network we may be trying
to solve problems with a single patch. If so, then no one can make a
move unless its deemed good for everyone else in the system. The
result will be a slow climb up to a modest level of fitness followed
by stagnation.
If were behaving as if each of our network members
is a patch unto themselves, then were in a chaotic zone playing
an internal version of the Red Queen strategy. Well find that
we can attain modest fitness peaks fairly rapidly but for some reason
the peaks deform so rapidly that were forever climbing.
Or we may divide into patches and divide the network
at the same time along patch lines. This forms factions or departments
that can lead to another form of Red Queen behavior.
concepts concerning complex adaptive
systems come from Stuart Kauffman's book At Home in the
Universe, 1995, Oxford University Press
application to organizational theory copyright © 1997, MG Taylor Corporation.
All rights reserved
copyrights,
terms and conditions
19970724072802.web.bsc
|