86 Joint Cognitive Systems
weight on the input and too little on the output; i.e., it considers only one side
of the coin, so to speak.
At the Right Time
Presenting the information at the right time is the final challenge.
Fortunately, this problem is a little simpler to solve than the previous ones
because timing can be defined in relation to the process state, which in many
cases is known or at least identifiable.
The question of timing can be seen as a question of presenting the
information neither too early, nor too late. One criterion could be that certain
conditions obtain that define the right time. For instance, in a collision
detection system for air traffic management, the indication of a possible
conflict can be defined relative to a given criterion (the separation distance).
For most industrial processes, alarms are shown only when given conditions
are present. These conditions, such as a high level in a tank, define when the
information is needed, at least from the process’ point of view. In practice
this may be either before or after it is needed by the operator.
The problem of whether the information is presented too late can be
solved in a similar manner, since the cessation of a process state or process
condition can be used as an indication. This, again, is a frequently applied
principle in presenting alarm information, where many alarms automatically
are cancelled by the system when the defining condition disappears.
The problem of the timing of information presentation is more difficult if
the criterion is the operator’s needs in the situation. In practice it is nearly
impossible to determine when an operator needs information of a certain type
(and in a certain form), since there is no way of determining the operator’s
mental state. There have nevertheless been several proposals for adaptive
interfaces and adaptive information presentation (e.g., Onken & Feraric,
1997), as well as more specific studies of the triggering conditions for such
systems (Vanderhaegen et al., 1994). An alternative to concentrating on the
operator is to use a combination of system state and operator state to
determine the timing issue, and furthermore to let the determination of the
operator’s state depend on measurable performance indicators rather than
inferred mental states (Alty, Khalil & Vianno, 2001).
How Should the Interaction Be designed?
The discussion above leads to the conclusions that the principle of designing
for simplicity is impractical. It has in a number of cases met with some
success, but only when it has been possible to map the conditions or
scenarios onto a limited number of categories or views. Furthermore, it must
be possible to make some kind of transformation or projection from the
Coping with Complexity 87
situations onto the categories or views. In general, the principle of designing
for simplicity requires that situations can be decomposed and described
structurally, and also that a sufficiently strong principle or theory of how
humans make use of that information is available. As we have shown
repeatedly in this book, there are reasons to be wary about the decomposition
principle, since it provides a conceptually simpler world at the cost of a
reduced match to reality.
DESIGNING FOR COMPLEXITY
The logical alternative to designing for simplicity is to design for complexity.
This may at first sound counterintuitive, since complexity was the initial
problem. Designing for complexity nevertheless makes sense because it
recognises that epistemological complexity cannot be reduced to an arbitrary
low level, which means that it can never be simplified so much that it can be
ignored. In consonance with the principles of CSE, the goal of ‘designing for
complexity’ is therefore to enable the JCS to retain control, rather than to
simplify the interaction or the interface.
Designing for complexity is also in good agreement with the Law of
Requisite Variety. Since the controller of a system must have at least as much
variety as the system to be controlled and since the requisite variety cannot be
reduced by interface design alone, the design goal must be to enhance or
increase the variety of the controller. This cannot be achieved by hiding
complexity, but possibly by making it visible, thereby allowing the JCS to
learn and increase its own variability. Designing for complexity reflects a
basic principle of CSE that the ability to maintain control depends on the
ability to understand what happens and to predict what will happen. Referring
to Figure 4.3 above, coping is facilitated if the JCS knows what has happened
and what happens, and if it can predict likely developments and thereby
anticipate the effect of possible actions (i.e., the classical principles of
observability and controllability).
Designing for complexity fully acknowledges that the work environment
of the operator is complex but draws the conclusion that this complexity is
necessary for effective control. Designing for complexity is not hostile to the
use of technology. On the contrary, the foundation for the principle is the
recognition that the joint cognitive system is a fundamental entity. The
difference is that designing for complexity aims to support general functions
of coping, rather than specific ways of acting in particular situations, hence
chooses generality over specificity. Some of the main principles that can be
applied to that are described next.
88 Joint Cognitive Systems
Support for Coping
One principle is that the design should support the natural human strategies
for coping, rather than enforce a particular strategy. A trivial example is to
consider the two extreme views of decisions making, rational decision
making and naturalistic decision making. Rational decision making is based
on a set of strong assumptions of what decision making is and what the
human abilities of being a rational decision maker are (Petersen & Beach,
1967). On the basis of that, it is possible to propose and build tools and
environments that support decision making – but only as it is described by
the model. For naturalistic decision making, and more generally for what are
called the ecological or ethnographic approaches to human performance, the
method would be to study what people actually do and then consider whether
it is possible to support that through design (e.g., Hutchins, 1995). This
approach is not atheoretical, but the theories are about what people do rather
than about the hypothetical ‘mechanisms’ behind. (An excellent example of
this approach is provided by Neisser (1982) – a decade before ‘cognition in
the wild’ became a catchphrase.)
The starting point, in other words, should be in an understanding of the
representative strategies for coping, with no initial assumptions about what
goes on in the operator’s mind. That understanding must obviously be
established separately for each domain and field of practice. It is
unreasonable to assume that there are strong domain-independent practices,
and that design can be based exclusively on those. Having said that, it is to be
expected that significant common features exist among domains, although
these may emanate from the work demands as likely as from inherent
psychological characteristics. In many cases there will be a need to control or
focus attention on important facets, for instance by applying a good
representation, cf. Gibson’s (1979) notion of perceived affordance. As
discussed above it is nevertheless far from easy to determine a priori what
will be significant and what will not. In general, since time is limited – and
time is perhaps the only true common feature across all situations – there will
be an advantage in reducing, filtering, and transforming information to avoid
obvious performance bottlenecks.
As a concrete illustration, at least three of the common strategies of
coping with input information overload – queuing, filtering, and cutting
categories – can be supported by interface design, and, in fact, often are,
although probably by coincidence rather than by design. Queuing is a feature
of VDU-based alarm systems, but may be used more systematically (Niwa &
Hollnagel, 2001). Filtering can be supported by categorising plant data and
measurements, for instance, according to urgency. And cutting categories can
be done by algorithmically mapping complex measurements onto a limited
set of more abstract functions (Corcoran et al., 1981).
Coping with Complexity 89
While designing for complexity neither can nor should advocate a single
principle or paradigm, it does provide a number of general principles, design
rules or heuristics that should be kept in mind. If these are followed, the
result should be a JCS that is better able to handle a wide range of situations.
The disadvantage is that the principles are high-level guidelines rather than
low-level design rules. They thus require a certain amount of interpretation as
well as some experience in system design. But this is only to be expected
since one cannot design a complex system or a complex interface without
knowing something about what lies behind. Notwithstanding many promises,
there are no simple step-by-step rules (like paint-by-numbers) that will allow
a person without sufficient experience to design a good system. There are
many sets of rules that can be used to check the outcome of a design and to
evaluate a system, but there are none rules that will produce it in the fist
place.
Time
An important principle in designing for complexity is to provide sufficient
time for the JCS to do its work. This is in most cases easier said than done,
since it is usually very difficult – if not impossible – to reduce the speed of a
process sufficiently. Physical processes and chemical reactions, for instance,
have their natural rate of development according to the laws of nature, and an
airplane must fly fast enough to create lift. Delaying or slowing down a
process is possible only in rare cases, such as stock market trading where
there may be a built-in ‘freeze’ if the volume of trading passes a given limit.
In cases where the speed of the process cannot itself be reduced, it may
sometimes be possible to buy some time by providing additional resources
for the control functions, i.e., increase the speed of the controller. In terms of
the cyclical model that means reducing the time needed for evaluation and
selection by introducing parallel processes, for instance by calling in
additional staff or in other ways lightening the task load.
Slowing down is in practice feasible only for processes where the user is
in direct control of the speed, such as driving a car. In this case there are no
physical limitations on the lowest possible speed, unlike, e.g., flying,
although the surrounding traffic may provide an obstacle. It is, indeed, a
common approach to pull up by the curb when orientation has been lost,
while driving in an unfamiliar environment – whether in a city or in the
countryside. Typically (in a city) the first reaction to losing orientation is to
slow down, to enable street names to be read and landmarks to be recognised
(depending on whether one is alone in the car or has a map-reading passenger
to assist). If this fails, the ultimate option is to stop completely and spend
enough time to re-establish orientation. Note that driving in an unfamiliar
environment also is a good example of how planning is interwoven with
90 Joint Cognitive Systems
actions, since the planning serves to identify the specific marks or waypoints
that guide the actual driving.
An interesting example of providing more time is the so-called 30-minute
rule that exists for nuclear power plants. According to this rule, the automatic
safety systems of the plant must be able to keep the reactor under control for
a period of 30 minutes, thereby giving the operators time to think. In the strict
interpretation of this rule, operators are not required – or even allowed – to
respond for the first 30 minutes. This rule does not slow down the process as
such, as the nuclear reaction continues at its own pace, but it does provide the
time needed to assess the situation and decide on a response. (There is
apparently no solid scientific or empirical reason for setting the limit to 30
minutes. In Japan, for instance, the corresponding rule calls for a 10-minute
respite.)
Another, but more indirect, way to provide sufficient time is to ensure
that the information presentation and the interface are as easy to use as
possible. An effective structuring of the information presentation will reduce
the demands to work. Conversely, a poor structuring of the information and
an inconsistent design of how the interface is controlled may lead to
unnecessary secondary tasks.
Predictability
Another way in which coping can be enhanced is by providing good
predictions or by supporting anticipation, either explicitly or implicitly.
Explicit support means that actual predictions are provided by some kind of
technology, raging from simple graphical extrapolations of trends (so-called
Janus displays, named after the Roman god of gates and doorways who had
two faces looking in opposite directions), over calculations of projected
developments such as in aviation and sailing (the position five minutes
hence), to faster than real-time simulations of not only the system itself but
also the environment. Well-known examples are the Traffic Alert & Collision
Avoidance System (TCAS) used in aviation, weather forecasts, warnings of
hurricanes and tsunamis, earthquake predictions (which usually have limited
success), market forecasts (also with limited success), predictions of
greenhouse effects, etc. On the fringe, the use of horoscopes and psychic
readings also illustrates the inexorable need for predictions, although the
accuracy may leave something to be desired.
Predictions are often partially supported by the way in which information
is presented. A simple example is when the level in a tank (or the value of a
stock or a currency) is shown graphically instead of digitally, i.e., by showing
present and past values rather than just the present value. Although display
design generally has been used to support designing for simplicity, many of
the established techniques can equally well be used to support designing for
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.