Content uploaded by Anne-Marie Grisogono
Author content
All content in this area was uploaded by Anne-Marie Grisogono on Apr 23, 2015
Content may be subject to copyright.
The Adaptive Stance
- steps towards teaching more effective complex decision-making
Anne-Marie Grisogono and Vanja Radenovic
Defence Science and Technology Organisation, Australia
Anne-Marie.Grisogono@defence.gov.au
Complex decisions arise in dealing with both complex situations and complex systems,
and there are a number of factors that place limits on our natural human capacity to deal with
them effectively. Many studies have identified failure paths stemming from our evolutionary
heritage of cognitive predispositions to respond in certain ways to various cues. These were
adequate heuristics to guide decisions in our much less complex evolutionary past, but now
often lead to predictable failures in our faster, richer and more inter-connected contemporary
environment. On the other hand, there is also a body of laboratory and real world evidence
supporting the thesis, underpinned by both complexity science and a theoretical neuro-
psychological model, that it is possible to cultivate an Adaptive Stance which counters these
predispositions and offers a more effective methodological framework for managing, creating,
shaping and interacting with complex systems and situations. An experimental approach to
teaching more effective decision-making through coaching in the Adaptive Stance has been
trialled with encouraging results, and has led to the development of a more detailed research
agenda.
1 Introduction
The problem of how to influence the development of complex systems and complex
situations towards favourable outcomes, and away from unfavourable ones, is
arguably at the core of most of the important and difficult challenges we face.
However the problem is also notoriously difficult, and the familiar litany of the best-
intentioned interventions being undermined by unexpected and unwanted
consequences provides ample evidence. To what extent are these difficulties inherent
and insurmountable? And what scope exists for advances in our ability to overcome
or mitigate those difficulties in some way?
Examples at global and national scales where such advances could yield
significant benefits include serious issues in economics, social welfare, public health,
national security, climate change and many other domains. At a more personal scale,
countless complex decisions are made every day by doctors, farmers, parents,
entrepreneurs, military officers, and social workers, to name but a few, with varying
degrees of success. As the complexity of our environment continues to grow
exponentially, improving the quality of complex decision-making (denoted as CxDM
for brevity in this paper) at all scales becomes more than ever a critical issue.
The motivation of the work reported here is to explore these questions, and as a
result, to identify ways of improving complex outcomes through better individual1
decision-making. We are interested therefore in understanding how to define and
judge effective decision-making in complex situations, how various factors contribute
to or detract from its quality, and what can be done to improve an individual’s overall
effectiveness in dealing with complex systems and situations.
The present thesis had its genesis in two independent threads of development, a
long range research program in DSTO into the theory and applications of complex
adaptive systems (CAS) science [1] on the one hand, and on the other, Dörner’s
microworld-based experimental research [2] into CxDM behaviour, supported by
theoretical neuropsychological models.
To begin the exploration, Sections 2 and 3 present a CAS perspective on the
nature of complex decisions and how their effectiveness can be judged, followed by a
brief account of a conceptual framework for understanding and exploiting adaptation
which suggests a methodological approach to dealing with complex decisions. Then
Section 4 gives a brief account of Dörner’s work and his findings. The consequent
implications arising from these two streams for the questions posed above, are
discussed in Section 5, leading to a synthesis which we term the Adaptive Stance, and
a number of conjectures and observations, together with an emerging research
agenda. Finally, Section 6 describes an experimental approach to improving
individual CxDM by education, reflection and mentored practice of the Adaptive
Stance, the early results achieved so far and the future work needed.
In summary, the thesis we present and discuss here is that it is possible to do
better. In particular, we claim that cultivating an Adaptive Stance offers a more
effective methodological framework for managing, creating, shaping and interacting
with complex systems and situations. While many important research questions
remain to be answered, the evidence accumulated so far warrants cautious optimism
and further investment in research.
1 While our current focus is on individual decision making in complex situations, we recognise that it is
essential to also address collaborative and organisational decision making for larger scale issues. However
our strategy is to first build a solid base of understanding at the individual level and then to apply and
extend it to the social, team and organisational domains.
2 The Nature of Complex Decisions
Although a case could be made for even the simplest human decision – vanilla or
chocolate ice-cream? – to be described as complex based on the complexity of the
underlying neural processes, this is not what we refer to as CxDM here. Rather, as
intimated in the introduction, we are interested in the decisions that an individual
decision maker (denoted DM hereon) makes in the process of attempting to influence
the development of a complex system or situation, precisely because they are so often
fraught with unintended consequences, and serious or catastrophic failures.
Systems are described as complex if they consist of many interdependent
components, whose interactions with each other and with the wider environment lead
to system properties and behaviours which are not simple aggregates of the
components’ properties and behaviours. In a similar way, problematic situations such
as epidemics, financial crises, insurgencies or the aftermath of natural disasters, can
also be described as complex, and may contain many complex systems as elements,
even if they are not systems in their own right.
We note that describing situations as problematic immediately implies a human
perspective to judge desirability of outcomes, and a goal for human decision making,
to solve or ameliorate the problem.
To probe more deeply into the nature of such decisions, we must first examine the
sources2 of complexity in the situation itself, and their impacts on a human DM.
Interdependence: What makes a system or situation complex is the network of
interdependencies between its constituent elements. Impact: the situation cannot
be successfully treated by dividing it into subproblems that can be handled
separately. Any attempt to do that creates more problems than it solves because
of the interactions between the partial solutions.
• Nonlinearities and Dynamics: In the real world, linearity is the exception not the
rule. Dynamical processes driving development of the situation often involve
many positive and negative feedbacks, thus amplifying and suppressing different
aspects of the situation, and resulting in highly non-linear dynamics. Impact:
linear extrapolation of current conditions can lead to serious errors.
Open boundaries: there is no natural boundary that completely isolates the
situation from external influences, there is always some traffic of information,
resources, and people, in and out of the situation. Impact: unknown influences
from outside the situation can bring about unexpected changes, and therefore the
DM cannot afford to focus only on what events inside some arbitrary boundary.
Multi-scalarity: Complex systems and situations exist at multiple scales, with
different agents, behaviours and properties at each, raising questions about how
interactions at one scale affect or are manifested in behaviour and properties at
other scales. This includes both emergence, the appearance of complex structure
and dynamics at larger scales as a result of smaller-scale phenomena, and its
converse, top-down causation, whereby events or properties at a larger scale can
alter what is happening at the smaller scales. In general, all the scales are
important, there is no single ‘right’ scale at which to act. Impact: the situation
cannot be successfully addressed at only one scale, mutually supporting actions
2 Note that the aspects listed are not an orthogonal set, there are many links between them.
need to be taken at every scale (eg individuals in local communities through to
national and international scales), and the interactions between the scales also
have to be understood and managed.
• Causal and Influence Networks: Interdependence of the elements also implies
that there are multiple interacting causal and influence pathways leading to, and
fanning out from, any event or property. Thus the consequences of any event or
property of the situation develop and propagate through many interacting
pathways, and therefore have impacts on many different aspects of the situation,
over different timescales. Similarly, analysis of how a particular property or
event came about, will find many cross-linked pathways that contributed to it. In
such a system therefore, one cannot expect simple causality (one cause – one
effect), or linear causal chains (each effect inexorably causing the next, like a
Rube Goldberg machine), to hold in general. Yet models based on such simple
causality and linear causal chains are attractive because they seem tractable,
unlike the confusing complexity of the real system. In fact, much of our cultural
conditioning is predicated on a naïve view of linear causal chains. Examples
include finding ‘the cause’ of an effect, or ‘the person’ to be held responsible for
something, or ‘the cure’ for a problem etc. This focus on singular or primary
causes makes it more difficult to intervene effectively in complex systems and
produce desired outcomes without attendant undesired ones – so-called ‘side-
effects’ or unintended consequences. A DM may be tempted to make a complex
situation appear simpler by restricting the scope of attention to a particular
pathway, but if the scope is widened to include other pathways, or if unexpected
side-effects that have propagated through those pathways are linked back and
suddenly manifested within the initially restricted scope, one is quickly reminded
that the ‘causal chain’ was just one of many pathways through a network. Such
systems are therefore inherently characterised not by linear causal chains, but by
networks of causal and influence relationships through which consequences
propagate and interact. Impact: the DM has a difficult cognitive task within
limited time and information resources, to develop sufficient understanding of
the situation to engage with it effectively, neither oversimplifying it, nor
becoming overwhelmed with unnecessary levels of detail.
• Emergence: Furthermore, such networks of interactions between contributing
factors can produce emergent behaviours which are not readily attributable or
intuitively anticipatible or comprehensible. Impact: there is significant potential
for unintended consequences and developments in diverse aspects of the
situation over various scopes and timescales, and hence unknown risks and
unrecognised opportunities.
Complex goals: There are generally multiple interdependent goals with respect to
the situation, both positive and negative, poorly framed, often conflicted, vague
or not explicitly stated. Moreover, stakeholders will often disagree on the
weights to place on the different goals. Impact: goalposts are likely to be moved
or to turn out to be unrealistic, goals that are achieved may turn out to have been
the wrong goals, and a DM is often left without sufficient guidance to interpret
the stated high-level goals and develop concrete goals for action.
Adaptation and innovation : Complex situations generally contain many adaptive
individuals and groups with complex relationships and shifting allegiances, and
new behaviours and features continually arise. Impact: approaches that worked
in the past may no longer work, interventions that frustrate the intents of some
agents will often simply stimulate them to find new ways to achieve them, and
opportunities created by vulnerabilities will be rapidly identified and exploited.
Opaqueness : many (perhaps most) important aspects of the situation are hidden,
whether through deception or simple unobservability. Impact: there is
considerable uncertainty as to how the events and properties that actually are
observable, are linked through causal and influence pathways, since many
intervening steps are hidden, and therefore many hypotheses about them are
possible. These cannot be easily distinguished based on the available evidence.
To make matters worse, these same difficult properties also characterise not only
the larger situations within which the situation of interest is embedded, but also the
larger situation within which the DM acts, for example a commander in a coalition
force operating within a whole-of-government and inter-agency context. Making
decisions in such environments is beset with uncertainties, confusion and
misunderstandings. As a result, decision heuristics which may be adequate in simpler
situations, can become dangerous cognitive traps. We now develop some important
and related observations that arise from the discussion of the sources of difficulty in
the complex situation and its environment.
Firstly, for an individual DM acting independently, the impact and relevance of
each of the above aspects needs to be assessed in the context of what the individual’s
goals are with respect to the situation of interest. In general, most of the inherent
complexity of a situation will not be relevant to the DM. For example, a simple goal
of repairing a road in a village does not require understanding of all the sources of
complexity in the village and its environment. But if the repair of the road is not the
ultimate goal, but rather an intermediate objective that has been adopted because it is
believed it will contribute to a higher level goal such as improving the local economy
and hence the welfare of the villagers, then more of the inherent complexity becomes
potentially relevant to the DM, in considering possible unintended effects and
consequences of repairing the road that could undermine the higher level goal.
Failing to do so might result in a repaired road actually detracting from the welfare of
the villagers, for example by facilitating the spread of a disease.
This introduces the important distinction between actual measures of success and
failure, and proxies for success and failure. Real success and failure relates to the
realisation (or not) of the individual DM’s own values, what intrinsic positive or
negative value is placed on various outcomes. Proxies on the other hand are not
valued for their own sake, but are pursued or avoided because the DM believes that
achieving or avoiding them will move the situation towards a real success outcome or
away from a real failure outcome, i.e. they are based on the DM’s conjectures about
the dynamics operating in the situation. In the above example, the repair of the road
could have been the measure of success for one DM who saw competent execution of
his tasks as his life’s work, but it would have been no more than a dubious proxy for
another DM who cared passionately about improving the lot of impoverished
villagers. Performing the same task in the same environment, the former does not
have a very challenging level of relevant complexity to deal with, while the latter
certainly does. Relevant complexity is in the eye of the beholder.
The concept of a proxy is recursively scalable. One can develop proxies for
proxies, i.e. more immediate outcomes which are sought or avoided because they are
believed to be steps towards or away from the proxies at the next scale up. A simple
test of whether a DM’s stated positive and negative goals correspond to real measures
of success and failure, or to proxies for them, is to ask the question : Why do you
want to achieve (or avert) this outcome ? If the answer is of the form in order to…
then it is a proxy and the rest of the answer identifies what it is a proxy for. This
process can be repeated until it terminates with an answer of the form because that’s
what I want, indicating it reflects the DM’s values. Note that except for this final
response, every other successive response identifying a proxy also surfaces the DM’s
underlying conjecture: the impact of achieving this outcome in the situation will be …
based on his conceptual model of the complex situation. This is very important to
elicit explicitly because all such conjectures carry significant risk of being flawed, but
once they are made explicit they can be subjected to an adaptive risk management
approach, by deliberately looking for disconfirning evidence and being prepared to
adapt the proxy as a result if the conjecture is refuted.
Thus, each repetition creates an additional scale in what might be seen as a house
of cards – several layers of conjectures built on conjectures, all dependent on flawed
and incomplete pictures of the reality. Nevertheless it is a very important construct to
understand since it captures the DM’s internal rationale for his actions and offers
useful insights into his decision processes. We have termed it the DM’s multi-scale
intent hierarchy [3], spanning high-level values-based success and failure objectives
for the situation, down to very low-level concrete objectives for action in the
situation. The more complex is the situation, the more intermediate scales are created
in the process of transforming the former into the latter. Evidently, in the above
example, the number of scales in the first DM’s intent hierarchy will be fewer than in
that of the second. This corresponds with the observation above, that the relevant
complexity is much lower for the first than for the second DM.
In summary, proxies are important because they provide a natural multi-scale way
of structuring the complex situation and surfacing the DM’s (often implicit)
conjectures about the situation’s dynamics, and in doing so, they also enable an
effective multi-scale adaptive approach to be taken to both learning about the
situation, and to influencing its development. There are important relationships
between inherent complexity in the situation, relevant complexity for the DM, and his
multi-scale intent hierarchy.
In principle it is straightforward to generalise this discussion beyond an individual
DM acting independently. In the military for example, where a top-down command
hierarchy provides the ‘higher intent’ context within which individuals make the
decisions they are empowered to make, that higher intent replaces the independent
individual’s personally chosen success and failure measures. It is possible therefore
to identify both an organisational multi-scale intent, and individual multi-scale intent
structures for each DM operating in the organisation. Ideally there will be enough
alignment between organisational and individual values, for the individual DMs in
the organisation to readily deconstruct and implement organisational intent, as well as
contributing to its development. In practice however there are many nuances to be
explored which are beyond the scope of the present paper.
We are now in a position to make some observations about individual decision-
making in a complex situation.
It should be clear by now that what we are calling CxDM is very challenging, and
differs from simple choices and from rational choice (optimising) decisions. It does
bear some relation to naturalistic decision-making [4], based on intuitive recognition
of dynamic patterns in the situation, and to expertise-based decisions [5] where the
DM has had sufficient experience with a class of complex situations of similar
structure, to have developed adequate conceptual models of their dynamics, and
strategies to cope with them. However our real interest is in tackling the hardest end
of the spectrum, where the DM is confronting novel one-off situations where there
has been no opportunity for the development of expertise, and for which intuitive
judgements carry unknown risks.
Since the purpose of making a decision is to produce desired outcomes, the
ultimate measure of its effectiveness has to be based on the outcomes actually
produced. But as discussed above, complex situations call for complex objectives
which span all the relevant dimensions and scales (including timescales) of the
situation, and the consequences of any one decision play out through many pathways
and interact with consequences of other decisions to influence many aspects of the
situation over various timescales. Therefore the effectiveness of complex decisions
cannot really be properly assessed until all the relevant outcomes have had sufficient
time to develop, nor can they be assessed independently of the other decisions made
that influence those same outcomes. It does not make sense therefore to talk about the
effectiveness of a single decision in a complex situation, but rather about the
effectiveness of the whole set of decisions made over a sufficiently extended time. In
other words, decision quality cannot be meaningfully defined or measured for single
decisions taken out of the context of all the related decisions being made as the
situation develops. In fact, because in general each decision made will close or open
up the possibility of particular further decisions, such a set is better described as a
network of decisions.
As we will discuss in the next section, dealing with complexity requires an
adaptive approach which is inherently cyclic and iterative, and which must permit
many cycles that ‘fail’ in some ways in order to learn and find what does work. Thus
a particular decision may not directly contribute to success, or may even lead to some
undesirable effects, but in the context of a decision network which constitutes an
adaptive approach, it may have played an important role because subsequent
decisions benefit from what was learned and start building towards successful
outcomes. So although it is only the overall outcomes from a network of decisions
that can be effectively judged, particular patterns of decision-making within the
decision network may be significant indicators of decision quality. In particular the
pattern corresponding to an Adaptive Stance, (defined and further discussed in
Section 5 below) will arguably be more successful than other possible CxDM
patterns such as a reactive stance (choosing actions in reaction to events), a
‘scattershot’ stance (random choices) or a procedural stance (following rules) .
3 Adaptation
Adaptation is a powerful natural process for dealing with complex problems and
situations. A conceptual framework for understanding and exploiting adaptation,
developed in earlier work [6], has been successfully applied to a number of defence
problems and has contributed to the development of the CxDM research reported in
this paper. We present here a brief overview of those elements of the conceptual
framework which are most relevant to the present work.
To adapt means literally to improve the fit of something (for example an
operational stratagem, and/or the system that will execute it) to something else (the
complex situation it interacts with). This immediately implies two things: firstly, it
must be possible for something to change – either in the stratagem / system, or in the
complex situation, otherwise there cannot be an improvement. Secondly, there must
be a way of assessing the result of such variations in terms of their impact on fit, or
the value function, so that improvement can be recognised. This enables selective
retention of success-enhancing variations and rejection of deleterious ones. A third
important aspect, is that adaptation is an ongoing cyclic process. There are always
more problems to be solved, more challenges to be dealt with, and moreover, the
environment itself is constantly changing, so that what brought success yesterday
may no longer do so today.
One can say therefore that the essence of adaptation is that it is (a) Value-based:
Variations are judged by their impact on what matters – success or failure, (b)
Grounded in reality: The judgment is based on trying things in the real world and
getting objective feedback, (c) Incremental: The system has to remain viable and
functioning throughout, and (d) Cyclic: It takes many iterative cycles to develop and
maintain the level of success needed in an environment which poses many complex
and continuously evolving challenges.
Together, these properties make it possible for an adaptive system to ‘grow’ an
effective solution to a complex problem without having to work it all out before
engaging. This is the secret of the power of adaptation in dealing with complexity,
since for problems that are complex enough, it is simply not possible to work out a
final solution in advance. Iteration with feedback permits starting with a partial
solution and improving it through learning over time. It is not difficult to see the
relevance of such an approach to CxDM.
The basic cycle of variation
Æ
interaction
Æ
feedback
Æ
selection
Æ
implementation of selection decision
Æ
repeat sounds simple and almost
algorithmic, but there are many devils in the details of how the cycle is implemented.
Therefore the effectiveness of any particular instance of adaptation is highly
dependent on the quality of its implementation [7] . In particular we note that the raw
material on which adaptation acts - the variations that are generated - determines the
possibility space that can be explored, and in the context of CxDM this is where the
DM’s creativity, intuition and experience play very important roles, whereas in
biological evolution, variation is to some extent random3 and therefore evolution is
slow and wasteful. Equally important is the quality of evaluation in the selection
3 This is an oversimplification which we do not have space to explore here, but see for example the
discussion of facilitated variation in Kirschner, M.W. and Gerhart, J.C. The Plausibility of Life Yale, 2005.
process, which again in the context of a human DM, depends on the quality of his
multi-scale intent framework as discussed in the previous section, while in evolution
there is no such judgment required since surviving the selection process is the
measure of fitness. But evolution is only one example of adaptation in the biological
world. Detailed study of the many forms adaptation can take, and how they operate as
successful strategies for surviving, even thriving, in a complex environment, has been
distilled into the elements of a conceptual framework [6], of which we select two
aspects which are particularly relevant to a human DM.
The first addresses what adaptation can achieve for the system. The framework
identifies continuous improvement in key system functions, plus four classes of
dynamic stress that need to be coped with:
• responsiveness to immediate threats and opportunities that arise;
• resilience to damage and shocks;
• flexibility in finding different ways to use its capabilities to cope with
environmental demands; and
• agility in shifting from one strategy to another as called for by significant
changes in conditions.
The second addresses how to improve success, via 5 nested levels of adaptation:
Level 1. Adaptive Action – the system uses its existing capabilities (to sense, to
process information and decide, and to act) to act adaptively in the world;
Level 2. Capability Improvement – the system applies adaptation to the capabilities
themselves, leading to improved sensing, information processing, deciding and
acting, and hence improves the success it can generate at Level 1;
Level 3. Learn to Adapt – the system applies adaptation to the adaptive processes
themselves i.e. to how potentially useful variations are generated, evaluated,
selected, and implemented, leading to improved outcomes at all levels;
Level 4. Defining Success – the system adapts the proxies for success and failure that
are internalised as seelction criteria in the adaptive processes for Levels 1-3, in
order to improve their alignment with ‘real’ success and failure; and
Level 5. Co-Adaptation – this level addresses interactions between adaptive processes
of interacting systems, eg by adapting allocation of roles and resources to them
in a System-of-Systems (SoS) context (thus altering the context of each
component system and the design of the overall SoS), and by choosing actions
with a better appreciation of their consequences through anticipating the likely
adaptive responses to them by other adaptive systems in the situation.
Note that Level 1 is the only level at which real world outcomes are produced.
The other levels basically act on the system itself and improve its ability to generate
success and avoid failure at Level 1, i.e. Levels 2-5 change different aspects of the
design of the system so that it is better equipped to succeed.
Applying this conceptual framework to the context of a DM in a complex
environment, suggests that all of these aspects offer the opportunity to achieve better
outcomes. The framework is scalable and can be applied in various ways, eg the
system can be thought of as the DM plus the systems he employs to plan and execute
his intervention. By utilising all five levels of adaptation the DM can not only
improve the quality of execution of his current functions, but also evolve his
functions to better fit the demands of the complex situation, and develop more
robustness in his own systems and in his strategy, to unexpected developments in all
four classes listed above. As another example, the system can also be thought of as
just the DM. Then application of the conceptual framework leads to human-centered
outcomes such as personal resilience and flexibility, the individual learning relevant
new capabilities, and improving the quality and effectiveness of his own learning.
All in all, application of the conceptual framework for adaptation to seeking to
influence complex situations implies that the DM needs to take an adaptive approach.
Broadly speaking, this means: clearly articulate his higher level values, deconstruct
them into higher level goals, then through as many intermediate levels of proxies as
necessary to arrive at an actionable level of detail, clarifying all the underlying
conjectures about the dynamics operating in the complex situation, and subjecting
them all to adaptive review. For each conjecture, this amounts to asking how would I
know if this conjecture was incorrect ? and how much would it matter if it was ? If it
does matter, i.e. it would lead to significant changes in the approach, then efforts
should be made to look for the earliest available indicators that the conjecture does
not hold. However since collecting and analysing information has costs, in practice
all such decisions need to be made in the light of a proper cost/risk/benefit analysis.
In addition to monitoring for evidence that the proxies need to be adapted, the
proxies themselves also need to be monitored. One of the reasons for developing
them in the first place is that because they generally have intrinsically smaller scope
and faster timescales than real success and failure measures, they can provide
feedback that is timely enough to assess the trajectory of the evolving situation and
guide adaptive action. Thus monitoring the time development of the proxies gives
feedback about the direct outcomes of the intervention, supporting adaptive
execution, while monitoring for evidence that underlying conjectures need to be
modified supports agility – adapting the approach to better fit the situation.
This account of the conceptual framework for adaptation is necessarily very
condensed. There is much more to successfully exploiting adaptation, but the
interested reader can find more detail in some of the referenced papers.
4 Microworld Experimentation with CxDM
The CxDM research program owes a very significant debt to Dörner’s pioneering
work [2] on the nature of human thinking when dealing with complex problems, and
how the seeds of later catastrophes are sown. Through many years of careful
laboratory experimentation with human subjects making complex decisions in
computer-based simulations with complex underlying dynamics – so-called micro-
worlds – he has identified the detailed decision making behaviours that differentiate
the small minority of individuals who succeed in achieving acceptable and
sustainable outcomes in the longer term, from the majority who fail to do so.
There are several important features of Dörner’s approach. The subjects are not
given concrete objectives but vague high-level goals, the microworlds are complex
enough to offer an enormous array of possible information to seek, actions to take, to
support a wide range of strategies, and to yield a wide range of outcome data across
many aspects of the microworld, thus reproducing several of the dilemmas of real
complex situations : what specific goals to choose ? what information to seek ? what
strategy to pursue ? The possibility space is too large for any exhaustive search
therefore optimisation approaches are out of the question.
Furthermore, the much-faster-than-realtime execution of the simulation means
that subjects are confronted with the unintended consequences of their earlier
decisions in a reasonable game timeframe, and have to decide how to deal with them.
This gives them the necessary opportunities to make the kinds of mistakes that real
people usually make in real complex situations. By contrast, terminating the game
too early, before the chickens have come home to roost, can give subjects a false
sense of their CxDM competence.
In addition to collecting information about the subjects’ decisions, and their actual
immediate and longer-term outcomes, Dörner’s experimentation protocol also
collects a rich dataset about their actions and decision behaviours, thus permitting
scientific exploration of the relationships between decision behaviours, immediate
decision outcomes and overall effectiveness in managing the complex situation.
His experimental results indicate that most subjects can score some quick wins
early in the game, but as unintended consequences develop and confront them, and
their attempts to deal with them create further problems, the performance of the
overwhelming majority (~90%) quickly deteriorate and their complex situations end
up in catastrophic or chronic failure. However the exciting corollary is that the
remaining ten percent are eventually able to find ways to stabilise their complex
situation. Their success is not due to chance because there are systematic differences
in the detailed CxDM behaviours which account for the differences in outcomes.
Moreover, the detailed CxDM behaviours of the unsuccessful majority reproduce
many well-documented findings4 about such cognitive traps as commitment bias, loss
aversion, confirmation bias, failure to concretise goals, collecting data but with little
analysis, difficulty in projecting cumulative and non-linear processes and in steering
processes with long latencies, and acting in a ‘fire and forget’ fashion. As a result
they have poor situation understanding, are likely to treat symptoms rather than
causal factors, to display methodism, repair-service behaviour, encapsulation and
both impulsive decisions and over-planning5. But their most significant failure is
meta-cognitive: little self-reflection and acceptance of responsibility.
The reasons for these and other cognitive failures have been explained as resulting
from our evolutionary heritage of biases, heuristics and cognitive predispositions [11]
which can still serve us well in the types of decision challenges that our ancestors had
to deal with, but which can lead DMs seriously astray in situations characterised by
high levels of complexity as discussed in Section 2 above. By contrast, the decision
making behaviours of the successful minority are able to counter these innate
tendencies by taking what amounts to an adaptive approach, confirming the
4 See for example analyses of human errors in other domains requiring complex decisions, such as
diagnostic errors in medicine [8], research into aviation safety [9], environmental protection [10].
5 Methodism refers to the behaviour of clinging to an approach that has worked in the past, without regard
for whether the conditions for its success still hold; a DM who simply fixes whatever immediate problem
he can find in the situation without trying to understand the bigger picture is displaying repair-service
behaviour; encapsulation is Dörner’s term for those DMs who become completely engrossed in a pet
project within the situation while ignoring unfolding catastrophes; overplanning produces plans that are
more and more detailed and elaborate and fragile, until the DM reaches a point of frustration, throws the
plan away and makes an impulsive decision unrelated to the earlier planning.
conceptual analyses of Sections 2 and 3. They develop a conceptual model of the
situation, and a stratagem based on causal factors, they seek to learn from unexpected
outcomes, and constantly challenge their own thinking and views. Most importantly,
they display a higher degree of ambiguity tolerance than the unsuccessful majority.
Dörner’s findings are particularly significant here because most of the literature has
concentrated on how CxDM fails, not on how it succeeds. His detailed catalogues of
the ‘good’ CxDM behaviours offers important insights into what is needed and begs
some obvious questions as to how teachable they might be.
It is not difficult to understand how the decision making behaviours associated
with the ‘poor’ DMs contribute to their lack of success, nor how those of the ‘good’
DMs enable them to develop sufficient conceptual and practical understanding to
manage and guide the situation to an acceptable regime. Indeed if the two lists of
behaviours are presented to an audience, everyone will readily identify which list
leads to successful outcomes and which leads to failure. Yet if those same individuals
are placed in the microworld hotseat, 90% of them will display the very behaviours
they just identified as likely to be unsuccessful. This implies that the actual DM
behaviours are not the result of conscious rational choice, but are driven to some
extent by unconscious processes.
This observation led Dörner to develop a theoretical model incorporating both
cognitive and neurophysiological processes to explain the observed data. In brief, the
model postulates two basic biological drives that are particularly relevant to CxDM,
the needs for certainty and for competence. Dörner pictures these metaphorically as
tanks which can be topped up by signals of certainty (one’s expectations being met)
and signals of competence (one’s actions producing desired outcomes), and drained
by their opposites – surprises and unsuccessful actions. The differences between
current levels and the tanks’ set points create powerful unconscious needs,
stimulating some behavioural tendencies and suppressing others, and impacting on
cognitive functions through stimulation of physiological stress. If both tank levels are
sufficient (but not full) the result is motivation to explore, reflect, seek information
and take aggressive action if necessary, but if the levels get too low the individual
becomes anxious and is instead driven to flee, look for reassurance from others, seek
only tank-boosting information that confirms his existing views, and deny or
marginalise any tank-draining contradictory information. The impacts on cognitive
functions reinforce these tendencies when the tanks are too low by reducing abilities
to concentrate, sustain a course of action, and recall relevant knowledge.
An individual whose tanks are low therefore finds it difficult to sustain the CxDM
behaviours associated with success, and is likely to act in ways that generate further
tank-draining signals, digging himself deeper into a vicious cycle of failure. We can
now understand the 90:10 ratio as the situation is not symmetric – the virtuous cycle
of success is more fragile because one’s actions are not the sole determinant of
success in a complex siutation, so even the best DMs will sometimes find their tanks
getting depleted, and therefore have difficulty sustaining the ‘good’ behaviours.
5 Synthesis: The Adaptive Stance & Research Questions
We have already remarked on the congruence between Dörner’s ‘good’ CxDM
behaviours and the implications for CxDM from CAS science. The Adaptive Stance
is an attempt to synthesise and operationalise the insights from both sources. It is both
an intellectual stance that creates the preconditions for being adaptive, and a
particular pattern of decision-making in complex situations. In brief its elements are:
• Ambiguity tolerance – the ability to resist the urge for closure and certainty;
• Openness to learning – in particular:
o accepting the possibility of being wrong, letting go of having to be right;
o resistance to loss aversion, commitment bias, falling in love with own ideas;
o accurate persistent awareness of one’s assumptions and hypotheses and
o simultaneous awareness of possible alternative ideas;
o continuously seeking ways to test them, being prepared to revise them;
o comparing one’s explicit predictions to actual outcomes as they occur, so as to
learn, improve future predictions, and assess one's own predictive ability; and
o doing the same with implicit predictions contained in every action or decision;
• An ingrained habit of thoughtful self-reflection about one’s beliefs, actions and
decisions, and the question: how would I know if I was wrong about this?
• Supporting others’ learning by appreciating that it is much more important for
them to admit error so as to be open to learning, than to feel that they always
have to be right (which would require them to either be risk-averse or in denial).
It is evident that displaying these qualities and behaviours would massively
increase the quantity and quality of learning by leveraging the opportunities inherent
in every action and decision, and would constitute an antidote to falling into some of
the ‘poor’ actor behaviours identified. However, it cannot guarantee success. As
noted in Section 3, the quality of options generated, and of the evaluations made, are
also important, and these both depend on the quality of the DM’s conceptual model
of the situation, of his stratagem and of his multi-scale intent framework. Thus we
add the following hypotheses to our synthesis:
There is a ‘right’ level of complexity for the DM’s conceptual model. Too much
detail means the DM cannot see the wood for the trees, wastes time and resources,
and suffers from data overload, however without enough detail, the DM risks missing
vital interactions and influences, falling into simplistic planning, and dealing with
symptoms rather than drivers.
The process of developing the conceptual model needs to be adaptive. If it is not,
it means the model is not based on real world evidence and the DM has succumbed to
confirmatory information collection. If it is adaptive, then the DM keeps learning
about the situation and tracking changes in it.
The conceptual model and the stratagem have to be co-evolved. If not, then the
DM is at risk of pursuing a poor stratagem, but if done well, then the stratagem and
system designs evolve to match the complex structure of the situation and the DM
can display agility in the face of complex dynamics (see [12] for fuller discussion).
The stratagem has to be implemented adaptively. If it is not, then the DM is
engaging in ‘fire and forget’ behaviour unlikely to produce success, but if done well,
then he can adapt execution in the light of how the situation develops. and display
resilience and responsiveness to unexpected opportunities and threats.
Self-reflection supports the ability to improve all above.
This synthesis prompts a number of research questions: to what extent can an
Adaptive Stance be cultivated or taught? how it should be done? does CxDM
effectiveness improve as a result? how durable and transferable are the gains? Further
questions raised include whether there are any psychometric indicators that could be
used to identify a subject’s potential for development and to customise the training
approach used; what other interventions might improve CxDM; how to set up a
CxDM competency framework to support adaptive implementation of CxDM training
and education; and how to extend this approach to distributed and team CxDM.
6 Experiments: Early Results
In collaboration with Professor Dörner and the Australian Army, the authors designed
and implemented an experimental approach to improving CxDM, using one of
Dörner’s microworlds: managing a chocolate factory in a complex environment
requiring subjects’ decisions about suppliers, product lines, design, quality control,
production planning, inventory management, capital and research investment, market
research, hiring, pay levels, and firing of personnel, staff amenities, advertising,
pricing structures etc. There are also competitors, and 30 market segments with
different demographics and preferences that changed in response to many factors.
Given the powerful role unconscious processes play in the theoretical model, we
knew that just imparting information would not be sufficient to change metacognitive
behaviour. Another consideration was the need for a scalable approach that might
eventually be implemented as part of Army training and education. Given the
numbers involved it could not therefore depend on extensive one-on-one instruction.
Our design therefore included initial and final baselines of individual play,
lectures, after action reviews, and several sessions of mentored play where the
subjects were paired and took turns being the player and being the mentor. Mentors
were asked to help their player become more conscious of their ‘good’ and ‘poor’
CxDM behaviours, and to encourage and support them in practising the Adaptive
Stance, but refrain from getting involved in the game itself. This turned out to be a
very successful innovation as the subjects reported both roles were effective in
enabling learning, but being a mentor and observing their peers’ CxDM efforts
afforded them even deeper insight and understanding. They also recorded
observations to augment data collected automatically through the game itself. In
addition, players filled out short surveys and outlined their expectations and strategies
in regular letters to their ‘shareholders’.
The experiments began with subjects being thrown into playing the game cold to
establish an initial baseline and to confront them with their actual (poor) versus
anticipated (good) performance, followed by an after action review, then seminars on
Dörner’s findings and theoretical model, and the Adaptive Stance. A second after
action review then yielded very different responses from the earlier one, as subjects
re-interpreted their experience in the light of the seminar information. The following
paired sessions gave them time to practise, observe and reflect, and feedback from the
following after action review was very positive. The final baseline of individual
(unmentored) play established that there was significant improvement in CxDM
behaviours and outcomes. The first two experiments conducted were exploratory to
develop the experiment protocol, but the third one, conducted for a week with 26
military officers, was large enough to afford a control group that were not given the
seminar material nor the mentoring, but had an equivalent amount of individual game
time. The control group also showed some improvement from initial to final baseline
as would be expected through individual learning about the game, but the
improvement of the study group was greater on every indicator. These results are
encouraging, but really represent no more than a toe in the water.
Much research remains to be done to produce practical cost-effective
interventions that deliver better CxDM, but there is much to be gained if we succeed.
Bibliography
[1] Grisogono, A.M. (2006) The Implications of Complex Adaptive Systems Theory for
Command and Control, CCRTS, San Diego, June 2006.
The Technical Cooperation Program, (2010), Joint Systems & Analysis, AG14 reports:
- CAS Concepts for Complex Operations DOC - AG14 - CxOps #2-2010
- Causal & Influence Networks in Complex Systems DOC – AG14 - C&IN #4-2010
[2] Dörner, D., (1997) The Logic of Failure: Recognizing and Avoiding Error in Complex
Situations. Perseus Books Group.
Dörner, D., (1980) On the Difficulties People Have in Dealing With Complexity.
Simulation Gaming, 1980. 11(1): p. 87-106.
Dörner, D. and Güss, C., A Psychological Analysis of Adolf Hitler’s Decision Making as
Commander in Chief Rev. Gen. Psych. 2011. 15(1), pp37–49
[3] Smith, E.A. Jr., Grisogono, A.M. and Clements, M.N. (2008) Cajole and Coordinate?
C2 in Whole-of: -Government, -Nation, and -Coalition Action , 13th ICCRTS Seattle US
[4] Klein, G., Naturalistic Decision Making. Human Factors: The Journal of the Human
Factors and Ergonomics Society, 2008. 50(3): p. 456-460.
Klein, G., et al., A recognition-primed decision (RPD) model of rapid decision making, in
Decision making in action: Models and methods. 1993, Ablex Publishing, US
[5] Salas, E. and G. Klein, Linking Expertise and Naturalistic Decision Making (Expertise:
Research and Applications). 2001: Psychology Press
[6] The Technical Cooperation Program, Joint Systems and Analysis, AG14 Synthesis
Report: A Conceptual Framework for Adaptation DOC – AG14 - CFA #3-2010
[7] Grisogono, A.-M., Success and Failure in Adaptation. Interjournal, 2006(1901)
[8] Croskerry, P., The importance of cognitive errors in diagnosis and strategies to minimize
them. Acad Med, 2003. 78(8): p. 775-780.
Redelmeier, D.A., The Cognitive Psychology of Missed Diagnoses. Annals of Internal
Medicine, 2005. 142(2): p. 115-120.
[10] Sterman, J., Learning from Evidence in a Complex World. Am J Public Health, 2006.
96(3): p. 505-514.
Moxnes, E., Not only the tragedy of the commons: misperceptions of feedback and
policies for sustainable development. System Dynamics Review, 2000. 16(4): p. 325-348.
[11] Arnott, D., A Taxonomy of Decision Biases. 1998
Arnott, D., Cognitive biases and decision support systems development: a design science
approach. Information Systems Journal, 2006. 16(1): p. 55-78.
Reason, J., Human error: models and management. BMJ, 2000. 320(7237): p. 768-770
[12] Grisogono, AM Design as a Strategy for Dealing with Complexity ICCS 2011