128 Joint Cognitive Systems
effective by improving coagency, the coupling between man and machine, a
JCS view. Two main issues in the practical implementation of this view are:
(1) what humans should do relative to what machines should do, and (2)
whether humans or machines are in charge of the situation. The first refers to
the function allocation proper, and the second to the issue of responsibility.
The two issues are obviously linked, since any allocation of functions implies
a distribution of the responsibility. The responsibility issue quickly becomes
complicated when function allocation changes over time, either because
adaptation has been used as a deliberate feature of the system, or – more
often – because the system design is incomplete or ambiguous. Although
both issues have been most conspicuous in relation to the use of automation,
the problem of function allocation is fundamental to any kind of human-
machine interaction, from display design to adaptive interfaces.
One attempt to resolve the automation design dilemma, for such it must
be called, is to resort to human-centred automation. Yet this often raises more
problems than it solves, mainly because human-centredness is an attractive
but ill-defined concept. There are three typical interpretations:
• That automation design is based on human needs. This is known rarely to
be the case; design is usually due to technical needs.
• That people ‘in the loop’ are part of the development of the design (= user
participation in design).
• That design is based on predicted improvements to human cognition and
performance. This requires well-developed models and an understanding
of the nature of work.
The CSE alternative to human-centred design is that designers from the
beginning consider the impact of changes on the roles of people in the system
and on the structure of the larger socio-technical system. In other words, they
should confront the substitution myth and challenge it. This can be done by
adopting a JCS perspective and by describing the system in terms of goals
and the functions needed to achieve them, rather than in terms of components
and their capabilities. In other words, by adopting a top-down rather than a
bottom-up approach. In this way fixed descriptions of task sequences are
replaced by definitions of the functions that are needed at each point in time.
The JCS should at every point in time have sufficient capabilities to achieve
the goals while being able to anticipate and predict future events. The best
possible candidate for a top-down analysis principle is the goals-means
method, which will be described in the following.
Joint Cognitive Systems 129
FUNCTION ANALYSIS AND GOAL ACHIEVEMENT
The design of any human-machine system, and in particular the design of
complex systems, requires some way of determining how tasks and functions
shall be distributed or assigned. This in turn requires some way of identifying
which tasks and functions are needed for the system to fulfil its functions,
specifically to achieve its stated goals. A task or function allocation always
assumes an a priori description of tasks or functions, that can be assigned to
either humans or machines. In the case of a JCS, the need is to define the
functions and capabilities that are necessary for the JCS to maintain control
of what it does.
The normal way of approaching this is by means of a task analysis, but a
task analysis logically requires as a starting point a description of the system
as already decomposed into humans and machines with their component
functions. Whereas a task description can be useful for improvements (as
well as for many other purposes), system design requires a functional
analysis rather than a task analysis. Hierarchical task analysis exemplifies a
decomposed view, with functions allocated to humans and machines (Annett,
2003). Other approaches, such as CWA (Vicente, 1999) looks at demands
rather than existing tasks, but does so from the view of a hierarchical set of
concepts, which means that it becomes difficult to account for variable
couplings and dependencies among functions.
Goals and Means
The alternative to a hierarchical analysis is to use a goals-means analysis. The
goals-means principle was used as the main strategy in the General Problem
Solver (GPS; Newell & Simon, 1961), although here it was called the means-
ends analysis. (In the two expressions, ‘goals’ and ‘ends’ are basically
synonymous, but the order between goals/ends and means has been reversed.)
The basic principle of the means-ends analysis was that differences between
the current state and the goal state were used to propose operators (meaning
symbols that represent operations, rather than humans that work), which
would reduce the differences. In the GPS, the correspondence between
operators and differences was provided as knowledge in the system (known
as a Table of Connections). The means-ends principle was later used as the
basis for the GOMS method (Card et al., 1983).
The General Problem Solver was by no means the first use of goals-
means analysis in behavioural science. In the 1950s several psychologists had
become interested in the developments in the meta-technical sciences, in
what became the forerunner of cognitive psychology and cognitive science.
Many researchers at that time felt that cybernetics could be a useful basis for
describing human behaviour, as demonstrated by Boring’s (1946)
130 Joint Cognitive Systems
deliberations about mind and mechanism. A different approach was that
taken by Miller, Galanter and Pribram who in their influential book on Plans
and the Structure of Behavior declared their fundamental interest to be:
… to discover whether the cybernetic ideas have any relevance for
psychology ... There must be some way to phrase the new ideas so that
they can contribute to and profit from the science of behaviour that
psychologists have created. (Miller, Galanter & Pribram, 1960).
Simple Test-Operate-Test-Exit (TOTE)
Their proposal for how to phrase the new ideas was the Test-Operate-Test-
Exit (TOTE) framework. TOTE demonstrated how a goals-means
decomposition could be used as the basis of a functional model of behaviour,
and thus preceded Newell & Simon’s use of it for the structural analysis of
reasoning. The work on plans by Miller et al. clearly showed that cybernetic
ideas and concepts could be used by psychology and was one of the more
notable influences for the emerging cognitive psychology.
The example that Miller et al. used was the simple action of hammering a
nail into a piece of wood. This was described in terms of a plan for
hammering, which is shown graphically in Figure 6.4. The goal is that the
nail is flush with the surface of the piece of wood, while the means is to
hammer down the nail. The plan (or procedure) is to continue hammering
until the goal condition has been achieved. In this case the plan has the
following components: (1) test if the head of the nail is flush (Test), (2) if not
then hit the nail with the hammer (Operate), (3) if so then stop hammering
(Exit); otherwise repeat the plan.
Test: position
of nail
Operate
(forcefully hit
nail on head)
(Head is flush)
(Head sticks up)
Exit
Figure 6.4: TOTE for hammering a nail.
Joint Cognitive Systems 131
One of the more powerful features of this description is that the activity,
i.e., hammering, is the result of a plan or an intention, rather than the
response to a stimulus. The person obviously can see that the head of the nail
sticks up, but there is no simple way of explaining how this ‘stimulus’ leads
to the response of hitting the nail with the hammer. From a contemporary
point of view the description of TOTE is simple and obvious and may seem
to be no more than a basic conditional statement. Yet it is important to
consider what the level of common sense knowledge was when it was
developed in the late 1950s. At that time few people had any experience with
programming or flow chart descriptions, or even of thinking in terms of
algorithms.
The value of TOTE is, of course, not just that it can be used to describe
how a nail can be hammered into a piece of wood, but that it provides a
general framework for describing any kind of activity or behaviour. The
TOTE unit is functionally homomorphic to the feedback control loop, which
in the 1960s was commonly known among control engineers and
neurophysiologists, but not among psychologists. TOTE represents a
recursive principle of description, which can be applied to analyse an activity
in further detail. (That Aristotle’s description of reasoning about means,
mentioned in Chapter 3, was also recursive shows how fundamental this
principle is.)
The goals-means decomposition for TOTE can, of course, be continued as
long as needed. In the case of plans there is, however, a practical stop rule.
Once a means describes an action that can be considered as elementary for
the system in question, the decomposition need go no further. As simple as
this rule may seem, it has important consequences for system design in
general, such as the expectations to what the user of an artefact is able to do,
the amount of training and instruction that is required, the ‘intuitiveness’ of
the interface, etc. The advantage of a recursive analysis principle is that it
forces designers to consider aspects such as stop rules explicitly; in
hierarchical analyses the stop rule is implied by the depth of the hierarchy
and therefore easily escapes attention.
Recursive Goals-Means Descriptions
A more complex version of the goals-means principle makes a distinction
between three aspects called goals, functions (or activities), and particulars
(or mechanisms). In the language of functional analysis this corresponds to
asking the questions ‘why’, ‘what’, and ‘how’, as illustrated by Figure 6.5.
The goals provide the answer to the question of ‘why’, i.e., the purpose of the
system. The functions describe the activity as it appears and provides the
answer to the question of ‘what’, i.e., the observable behaviour of the system.
Finally, the particulars provide the answer to the ‘how’, i.e., by giving a
132 Joint Cognitive Systems
detailed description of the way in which the functions are achieved. Figure
6.5 also illustrates how the goals-means decomposition can be used
recursively. That which constitutes the means at one level, becomes the goals
at the next level down, and so on. Altogether this provides a method for
functional decomposition, which is essential to CSE because it logically
precedes the structural decomposition that is commonly used in systems
analysis. The goals-means analysis has been used extensively as the basis for
the engineering analysis method known as multi-level flow modelling
(MFM; cf. Lind & Larsen, 1995), for instance as a basis for constructing
correct procedures. For the purpose of illustrating the goals-means analysis in
CSE, it is nevertheless sufficient to stay with the basic goals-means
decomposition method, since the means always can be decomposed further
into functions and particulars.
WHY
WHAT
HOW
The purpose or
objective of the
activity
The details of how
it is done
(“mechanism”)
Goals
Means
WHY
WHAT
HOW
WHY
WHAT
HOW
The activity or
function itself
(“phenomenon”)
Figure 6.5: The goals-means decomposition.
The advantage of a recursive analysis principle is that the resulting
description is as simple as possible with regard to the concepts and relations
that are used. In contrast, an analysis based on a hierarchical principle must
refer to a pre-defined structure of a finite, but possibly large, set of concepts
organised at a fixed number of levels. Each level typically represents a
different mode of description, which means that the descriptive dimensions
can be radically different at the top and the bottom (Lind, 2003). A
hierarchical analysis furthermore has a pre-determined depth, corresponding
to the number of levels in the hierarchy, and usually tries to achieve the
maximum depth. Since a recursive analysis principle does not imply a pre-
defined depth or number of steps, it compels the analyst to be parsimonious
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.