Course Handout - Situation(al) Awareness (SA)
in Effective Command and Control
Copyright Notice: This material was written and published in Wales by
Derek J. Smith (Chartered Engineer). It forms part of a multifile e-learning
resource, and subject only to acknowledging Derek J. Smith's rights under
international copyright law to be identified as author may be freely downloaded
and printed off in single complete copies solely for the purposes of private
study and/or review. Commercial exploitation rights are reserved. The remote
hyperlinks have been selected for the academic appropriacy of their contents;
they were free of offensive and litigious content when selected, and will be
periodically checked to have remained so. Copyright
© 2018, Derek J. Smith.
|
|
First published online 08:28 GMT 17th March 2003,
Copyright Derek J. Smith (Chartered Engineer). This
version [2.1 - links to graphics] dated 09:00 BST 14th June 2018
1 - The Historical
Background
"My experience
is what I agree to attend to" (James, 1890/1950, pI.402).
The nineteenth century
philosopher-psychologist William James once defined the difference between
sensation and perception as resting in the fact that the latter included
"the consciousness of farther facts" associated with the former (James,
1890/1950, pII.77). When you perceive something, James argued, you become aware
not just of the existence of an object in the external world - its shape, colour, height, size, and speed, say - but also, thanks to
your previous experience of the world, of that object's implications for you.
Is it, for example, good to eat, or likely to eat you, or useful as a tool, or
any one of a thousand other things; and what new behaviour
on your part (if any) is warranted by its presence, and how urgently? As a result,
your thoughts are punctuated every few moments by a new cycle of <object -
implication> processing, and this has
ramifications, in turn, for how we might explain the phenomenon of
consciousness. In the event, James proposed a "personal consciousness"
with five key characteristics, as follows .....
1. One's personal
consciousness contains all one's thoughts.
2. These thoughts
are ever-changing.
3. Within one's
personal consciousness, thought is "sensibly continuous" (pI.237). That
is to say, thought "feels" continuous even when we can point to
objective time gaps. "Consciousness," James concluded, "does not
appear to itself chopped up in bits" (pI.239).
4. One's personal
consciousness "deals with objects independent of itself; that is, it is
cognitive, or possesses the function of knowing" (pI.271).
5. One's personal
consciousness, James wrote, is "interested in some parts of these objects
to the exclusion of others, and welcomes or rejects - chooses from among
them - all the while" (pI.225; italics original). In other words, it
selectively attends.
It is James's fifth point which is
most relevant to the present paper, because the ability to choose between
inputs - that is to say, the mental faculty of "selective
attention" - is central to our momentary understanding of the world.
The objects it selects largely determine the subject matter of consciousness,
and their implications largely determine the quality of the resulting
experience. Moreover, it was in discussing selective attention that James
introduced the related notion of the "span of consciousness",
a concept which James defined as "the number of things we may attend
to" (pI.405), and saw as being severely limited .....
ASIDE: James described a number of
earlier studies which had tried to determine the maximum array size of briefly
visually presented items whose number could be known instantly by an observer
(that is to say, without the need for counting). He quotes data from Jevons
(1871) to the effect that errors were rare on displays of up to five items, and
then started to appear at six items (his subjects scored 120 out of a maximum
147 with six item arrays). Woodworth and Marquis (1949) call this limitation
the "span of apprehension", define it as "the amount of material
that an observer can grasp or 'apprehend' in a single act of perception"
(p416), and place the point of total breakdown of accurate perception at
"about 12" items. Past this point, observers can only "give an
estimate".
Both of James's new topic areas
were, of course, unremittingly cognitivist, meaning that neither received much
scientific attention during the behaviourist
interregnum. The study of selective attention then suddenly came back into
fashion in the wake of a series of new age experiments carried out in the early
1950s (eg. Cherry,
1953). The study of consciousness took a
little longer to become accepted, but is now the central topic of
interest and investigation within the whole of cognitive science. Indeed, there
are several distinct perspectives to its study, including
.....
the philosophical
perspective - this line of enquiry looks at what consciousness really is.
the computational
perspective - this line of enquiry considers how to build artificially conscious
machines.
the neuropsychological
perspective (macro) - this line of enquiry looks for the brain centres responsible for consciousness.
the neuropsychological
perspective (micro) - this line of enquiry looks at what is happening at
neuronal level (or smaller) to deliver consciousness.
the clinical
perspective - this line of enquiry looks at how consciousness-impairing diseases
can be better treated.
the safety perspective - this line of enquiry
looks at the limitations of consciousness behind the steering wheel (drivers,
pilots, etc.), in the control room (power stations, etc.), or in the boardroom
(disastrous decision making).
the forensic
perspective - this line of enquiry looks at how said limitations can be diagnosed
after a disaster .....
the design perspective - ..... and prevented from recurring by designing better systems.
the military
perspective - this line of enquiry looks at how consciousness can be broadened so
as to enhance individual or group battlefield performance.
The first of these perspectives has
been with us for several thousand years, and, ultimately, cannot be resolved in
isolation because it has no data, only opinion. The next three perspectives are
much younger, have more empirical data than they can yet interpret, and are
already individually massive supersciences. However,
it is the last four perspectives which interest us here, for as a topic cluster
they make up the science of "situation(al)
awareness (SA)" .....
2 - Early Work
"Found difficulty in
thinking of everything at once, and stalled too high in first attempt at
landing" (Kenneth Craik, after trying out the Link Trainer; undated
pre-1945; as cited in Sherwood, 1966, p115).
The idea that we have to struggle to
control our passing through the world, especially when "at the wheel"
in some way, is not new within psychology. During the 1930s, for example, Frederick
(later Sir Frederick) Charles Bartlett (1886-1969), director of the
experimental psychology laboratories at Cambridge University, carried out
research on behalf of the RAF, studying amongst other things ways of reducing
air accidents by better selection and training procedures (eg.
Bartlett, 1937). This type of work automatically
became more important during World War Two, and drew in a number of talented
young postgraduates who would go on, after the war, to transform psychology. At
Cambridge, this included Norman Mackworth and Kenneth Craik
(1914-1945), and across the Atlantic it included Alphonse Chapanis (1917-2002) (the "father of ergonomics"),
J.P.Guilford, Paul M. Fitts,
and George A. Miller. Craik joined Bartlett's laboratories in 1936 as a
doctoral student, and when war came became involved in researching the design
of cockpit simulators (Bartlett, 1946). He played an important part in the
development of the "Cambridge Cockpit", a model of good
aircraft design, and suggested many practical improvements to both controls and
instrumentation in recognition of the many physical, physiological, and perceptuo-motor limitations of the pilots being trained.
As for pilots' higher cognitive
functions, it made no difference whether you focused on divided attention,
selective attention, sustained attention, or decision making, the finding was
always the same, namely that the cognitive system was routinely severely
limited. Studies of divided attention found a cognitive system which could only
divide itself so many times, studies of selective attention found a system
which took time to select its target and was then easy to distract, studies of
sustained attention found that even the fittest and most motivated personnel
could only go so far in watching a screen or peering through a pair of
binoculars before they started seeing things (or, worse, not seeing things),
and studies of decision making found depressing levels of carelessness and poor
judgement just about everywhere.
ASIDE: For a recent study of divided
attention, see Wickens
(1997). The sustained attention studies were amongst the first to enter
civilian psychology after the war, as the psychology of "vigilance"
(eg. Mackworth, 1950). The study of vigilance as a psychological construct
has an abstract dated 1923 by no less an authority than the famous neurologist
Sir Henry Head. However, Head saw vigilance as being much like general
awareness, rather than the much narrower capacity for sustained perceptual
performance under conditions of monotony. Many examples of ultimately disastrous
decision making are given in our online resources in military, aerospace, railway, maritime, and IT
project management disasters.
Thus far, the study of cockpit
cognition had simply been an extension of the prewar science of "human
factors", which was itself an offshoot from the tradition of "scientific
management", the formal study of industrial and commercial processes
which had taken shape in the US more than half a century beforehand. The lead
author on this particular timeline was Frederick Winslow Taylor (1856-1915). It
was Taylor, for example, who had introduced "time and motion
study" methods into late nineteenth century industry, and who formalised what would nowadays be termed
"evidence-based management". After Taylor came Charles Samuel Myers
(1873-1946), who referred to the science of humankind at
work as "industrial psychology" (Myers, 1929). In 1912 Myers
founded the Cambridge psychology laboratories, and became its first Director
(thus Bartlett's predecessor). The new science's own
journal, Human Factors, was first published in the early 1930s. After
1945, the term "human engineering" became popular for a while
in the US (Roscoe, 1995/2003
online), and "applied psychology" in the UK. There were further
name changes and reorientations during the 1950s. Paul Fitts
led the movement towards "engineering psychology" in 1951 (Fitts, 1951), but as that decade unfolded the name "ergonomics",
the science of humankind at work (Greek: ergon = effort), grew more
popular. The journal Ergonomics was launched in 1957. Finally, in the
early 1960s, the concept of "cognitive engineering" was
introduced (Bonjer, 1962), to be followed in 1977 by
the term "cognitive ergonomics" (Newman, 1977).
ASIDE: For a fuller discussion of
the modern names and angles of approach, see Stanton's (1996) special edition
of The Psychologist or Lintern's (1999)
special edition of The International Journal of Aviation Psychology.
3 - Motor Skills
|
Much of the following section appeared in Smith (1996; Chapter 1). It is repeated here with minor amendments and supported with hyperlinks. |
The
psychological investigation of motor skills began late in the nineteenth
century as the processes of industrialisation began
to place a premium on the rapid training of machine operators. Bryan and Harter
(1897, 1899) investigated how quickly Morse code operators could learn the
necessary keying and decoding skills. They interviewed and observed both
experienced and trainee Western Union telegraph operators and plotted learning
curves of performance against practice, finding distinct periods where
continuing practice seemed to have no effect on ability. Because
these periods showed up as flat areas of the learning curve, Bryan and Harter
called them plateaus (optionally, plateaux).
The topic was further investigated by William Book in the early years of the
twentieth century. He studied the rate of acquisition of typewriting skills,
and concluded as follows:
"There is in
all [subjects] a rapid and continuous rise in the early stages of practice
followed by a slower and more gradual rise in the later stages. All show marked
daily variations [and] short periods of non-improvement." (Book, 1908, p158.)
In fact,
Book distinguished two types of non-improvement. The first - called
"breathing points" - lasted between six and eight days and were found
in all subjects, and the second - the plateaus first noted by Bryan and Harter
- lasted 17-33 days but were not found in all subjects. Other popular devices
for testing motor skills were soon devised, including .....
The Pursuit Rotor: This task requires
subjects to keep a stylus in contact with a target point mounted off-centre on
a turntable. (The test exists in both one- and two-handed versions.)
Linear Tracking: This task requires
subjects to keep a stylus in contact with a moving spot viewed through a slot.
Steering and Fine Control: This task requires
subjects to control a piece of machinery like a bicycle, car, or aeroplane (or, at least, a simulator thereof). (Note that
aircraft simulators and cars both involve simultaneous hand and foot activity.)
Mirror Drawing: This task requires
subjects to trace round a line pattern viewed in a mirror.
Upside Down Letter Drawing: This task requires
subjects to invert text letter-by-letter as quickly as they can working from
right to left and up the page (so that the response sheet becomes readable as
normal if simply rotated through 180o).
Track Tracing: This task requires
subjects to move a looped electrical contact along a bent wire without touching
it (a task now commonly used as a fete sideshow).
As to
why plateaus should exist, Bryan and Harter argued that learning a skilled act
involves setting up a hierarchy of habits. At the bottom of this
hierarchy are discrete low-level motor acts such as transcribing or
transmitting the individual Morse letters, then comes
the ability to send firstly syllables and eventually whole words as known
strings, and at the top of the hierarchy comes the ability to send firstly
phrases and eventually whole sentences as known strings. Singer (1968) calls
these first-, second-, and third-order habits respectively. Using
tennis skills as his example, he identifies the lowest order habits as the
basic strokes, the middle order habits as combinations of strokes and movements
on the move, and the highest order habits as tactical game plans and the like.
As to
why plateaus are important, the standard argument is that as learning takes
place the brain reorganises itself so that more and
more information can be processed at a time. It is just that it takes time and
lots of practice for this mental reorganisation to
take place. Plateaus occur whenever a lower-order habit approaches its absolute
maximum speed, but cannot yet be invoked sufficiently automatically for
concentration to be switched to developing the next higher-order habit.
It is
also possible to identify more than one category of motor task, each making
different demands on different types of memory, and each therefore following a
differently shaped learning curve. Miller (1969) lists four possible categories
of motor task, as follows:
Category I - Adjustive Tasks: This category of skill involves continuous perceptual-motor adjustment,
and includes such tasks as tracking, pursuit, and steering.
Category II - Response Selection Tasks:
This category of skill involves selection of appropriate responses from a
repertoire, and includes such tasks as typing, sight reading of music, and
sending Morse code.
Category III - Procedural Tasks: This
category of skill involves complex, step-by-step, skills such as piloting an
aircraft or assembling something from its components. (Thus
making it a skill of executing other skills.)
Category IV - Skilled Performance Tasks: This
category of skill involves complex, but nevertheless discrete, motor skills
such as throwing and catching, ball control, and athletic skills such as
vaulting and diving.
But the
count could well be higher even than that. Following a series of factor analytic
studies going back to 1954, Fleishman (1972) identified nine indepently variable physical components of human motor
ability and eleven independently variable mental factors. He called the
former physical proficiency factors, and the latter psychomotor
ability factors. Here is what they seem to measure:
Physical Proficiency Factors: 1
- Extent Flexibility: This is the ability to
flex-stretch the trunk or limbs to their limits in their various planes of
movement. 2 - Dynamic Flexibility: This is the ability to make repeated
rapid flex-stretchings of the trunk and limbs. 3 -
Explosive Strength: This is the ability to do work in a burst of effort.
This ability would be seen in the shuttle run test. 4 - Static Strength:
This is the ability to exert force over a brief period. This ability would be
seen in weightlifting. 5 - Dynamic Strength: This is the ability to
exert force repeatedly or continually, as when doing pull-ups. 6 - Trunk
Strength: This is the ability of the trunk muscles, especially the abdominal
muscles, to exert force repeatedly or continually, as when doing leg-lifts. 7
- Gross Body Coordination: This is the ability to coordinate the activity
of different parts of the body. 8 - Gross Body Equilibrium: This is the
ability to maintain balance without visual input, as when standing on one leg
with your eyes shut. 9 - Stamina: This is the ability to maintain
maximum effort over a period of time.
Psychomotor Ability Factors: 1
- Control Precision: This is the ability to
fine control rapid muscle adjustments. 2 - Multilimb
Coordination: This is the ability to coordinate a number of limbs
simultaneously. 3 - Response Orientation: This is the ability to select
the right movement from a repertoire of several possible movements at speed. 4
- Reaction Time: This is the ability to respond quickly to a stimulus. 5
- Speed of Arm Movement: This is the ability to execute a rapid ballistic
arm movement. 6 - Rate Control: This is the ability to make continuous
anticipatory motor adjustments in a tracking task. 7 - Manual Dexterity:
This is the ability to perform hand-arm manipulations of fairly large objects
at speed. 8 - Finger Dexterity: This is the ability to perform finger
manipulations of fairly small objects at speed. 9 - Arm-Hand Steadiness:
This is the ability to make precise arm-hand positioning movements (as opposed
to powerful or rapid ones). 10 - Wrist-Finger Speed: This is the ability
to perform wrist-finger manipulations at speed. 11 - Aiming: This is the
ability to position probes and pointers accurately at speed, or to dot in very
small circles and suchlike.
There
is, of course, no memory involvement whatsoever with the physical proficiency
factors. These - as their name suggests - are purely skeleto-muscular
in derivation, and any improvement with practice is entirely due to
physiological factors (muscle strength, joint mobility, glucose metabolism,
etc). The psychomotor factors, on the other hand, are mediated by the central
nervous system, and when they show improvement with practice it has to be due -
in part at least - to memory-like processes resulting from that practice.
But the
situation is highly complex. For one thing, Fleishman describes the physical
factor of gross body coordination in remarkably similar terms to the psychomotor
factor of multilimb coordination, and
yet the former is scoring a skeleto-muscular ability
whilst the latter is scoring some aspect of nervous system organisation
and efficiency. For another, there is evidence that the psychomotor factors are
in fact largely separated from the "higher order" mental functions
such as awareness and will. It is not uncommon, for example, to find that
brain-injured subjects can learn new motor skills without being able to remember
any of the practice sessions! And last but not least, the relative contribution
of Fleishman's 11 psychomotor factors actually shifts as performance improves
with practice. Thus when first learning a particular skill, the reaction time,
speed, and dexterity factors effectively play no part at all: like Craik in his
simulator, you are too busy getting things right to worry about doing them
quickly. However, by the time the skill is well established, these same factors
are contributing about 25% of the variance (Fleishman and Hempel, 1955).
The most
probable reason for all this confusion is that the intact brain works by
linking up separate systems, known as modules, each of which can be
separately trained, and each of which can be separately damaged. And you become
"skilled" when each module knows its own job and can talk fluently to
its neighbour. Unfortunately, psychology has not yet
figured out exactly how many modules there are, nor what each
of them is responsible for, nor (consequently) what information needs to
pass between them.
4 - Definitions of
SA
As far as we can establish, the
concept of SA was born within the specialist and rather secretive world of
military ergonomics and air accident investigation in the mid-1970s, as an
attempt to explain the large number of individual variables known to affect the
cognitive performance of military and civilian aircrew. However, it did not go
public in the civilian literature until about ten years later [the earliest PsycINFO reference when we checked 26th February 2003 was
Spiker, Rogers, and Cicinelli (1986)]. It is best
introduced by looking at some of the attempts to define it
.....
SA is the ability
to "maintain the 'big picture' and think ahead" (Dennehy
and Deighton, 1997, pI.284).
"SA
is defined as the 'operational space' within which personal and environmental
factors affect performance" (Dennehy and
Deighton, 1997, pI.287).
"[SA is] the
degree of accuracy by which one's perception of his current environment mirrors
reality" (US Navy website).
"[SA is a] cognitive
state or process associated with the assessment of multiple environmental cues
in a dynamic situation" (Isaac, 1997, pI.185).
"[SA] can be
thought of as an internalised mental model of the current
state of the flight environment. This integrated picture forms the central organising feature from which all decision making and
action takes place. A vast portion of the aircrew's job is involved in
developing SA and keeping it up to date in a rapidly changing environment."
(Endsley, 1999, p257; emphasis
ours.)
"[SA is] the
accessibility of a comprehensive and coherent situation representation which is
continuously being updated in accordance with the results of recurrent
situation assessments" (Sarter and Woods, 1991,
p52).
But the definition we like best
(because it is closest to James's conceptualisation
of consciousness) is .....
"[SA is]
the perception of the elements in the environment within a volume of time and
space, the comprehension of their meaning, and the projection of their status
in the near future" (Endsley, 1988, cited in
Endsley, 1999, p258; emphasis ours).
SA is thus substantially more than
the simple perception of data. Instead, it is an index of how a multitude of separate elements
interact, and of how that interaction then affects what each element is likely
to do next; it is William James's object implications all over again.
And why does all this matter? Well above all because .....
"..... problems with SA were found to be
the leading causal factors in a review of military aviation mishaps [citation],
and in a study of accidents among major aircarriers, 88%
of those involving human error could be attributed to problems with situation
awareness" (Endsley, 1999, p258; emphasis
added).
5 - SA as a
Multidimensional Construct
The fact that SA is not a single
ability creates serious problems for human factors researchers, because it
means they have no single straightforward measure of the phenomenon they are
most interested in. Indeed, when Dennehy and Deighton
(1997) reviewed the literature from the period 1979 to 1992, they identified no
less than 28 different variables which had been used as measures of SA.
Suspecting that these variables could profitably be clustered, Dennehy and Deighton set out to find the main SA subscales.
After discounting some obvious synonyms, they reduced the list of factors to 22
variables, and they then arranged for 231 pilots and instructors to rate these
22 variables on a seven-point scale of how necessary each was to maintaining an
effective level of SA. A principal components
analysis was then carried out on the raw subjective ratings, and this
revealed five main subscales (the authors use the descriptor "meta-categories"),
as follows .....
SA Meta-Category 1
- Pilot Knowledge: The first principal component was a composite measure
of knowledge, per se.
SA Meta-Category 2
- Anticipation and Understanding of Future Events: The second principal
component was a composite measure of an ability cluster for anticipating and
understanding future events.
SA Meta-Category 3
- Capacity to Manage Stress, Effort, and Commitment: The third principal
component was a composite measure of an ability cluster for managing stress and
maintaining effort and commitment.
SA Meta-Category 4
- Capacity to Perceive, Attend, Assess, and Assimilate Information: The fourth principal
component was a composite measure of an ability cluster for obtaining and
processing information from the world itself, both directly, and from the
cockpit dials and gauges available.
SA Meta-Category 5
- Overall Awareness: The fifth principal component was a composite measure
of general awareness, similar to the "g-factor" popular among some
intelligence theorists.
In a further attempt to understand
this clustering, and to help defend against criticisms of the subjective
ratings methodology, Dennehy and Deighton adopted the
Endler-Mischel "Interactionist Theory"
(Endler, 1973; Mischel,
1973). This theory holds that it is impossible "to understand the role of
an operator in isolation from the context within which she/he operates" (Dennehy and Deighton, p285). The result is a "P-E
Fit" model, that is to say, one which tries to "fit" the
person into his/her environment. However, the overlap between the person and
the environment can never be complete, meaning that four distinct theoretical
domains need to be taken into account, as follows .....
"Capacities" - P-Factors, Not
Interacting: These are matters of fact relating to the person and defining what they
are objectively capable of. This might include their baseline memory span or
reaction time, for example. Dennehy and Deighton deal
with this cluster of factors under the title "objective person"
factors.
"Abilities" - P-Factors, Interacting
with E-Factors: These are the environment-facing aspects of the person, such as task
awareness and the ability to anticipate system response in a given environment.
Dennehy and Deighton deal with this cluster of
factors under the title "subjective person" factors.
"Roles of Work" - E-Factors,
Interacting with P-Factors: These are the person-facing aspects of the
environment, such as the tasks demanded of a system operator, or the objectives
of their mission. Dennehy and Deighton deal with this
cluster of factors under the title "subjective environment"
factors.
"Real World" - E-Factors, Not
Interacting: These are matters of fact relating to the environment, such as the
objective nature of the equipment, the nature of the weather, etc. Dennehy and Deighton deal with this cluster of factors
under the title "objective environment" factors.
Dennehy and Deighton then characterised SA as the "operational space"
provided by the two interacting domains. The thrust of this argument is summarised diagrammatically in Figure 1.
|
Figure 1 - The Interactionist ("P-E Fit")
Model of Situational Awareness:
Here is Dennehy and Deighton's "interactionist
framework". It shows two main categories of factors, namely personal (or
"P") factors (left oval) and environmental (or "E")
factors (right oval). These separate sets of factors have been aligned so
that they only partly overlap, giving a central zone (deeper yellow) where
the P- and E-factors "interact" (hence the tag
"interactionist"), plus two non-interacting zones on the flanks
where objective measurement is possible. Specimen factors for each sub-cluster
are shown from left to right across the belly of the diagram. The
operational space which makes up SA is the zone of overlap between the two
ovals. The inner dotted ovals show areas where the subjective and
objective elements further interrelate. If this diagram fails to load automatically,
it may be accessed separately at |
|
Enhanced from a black-and-white original in Dennehy and Deighton (1997; Figure 1). This version Copyright © 2003, Derek J. Smith. |
Endsley (1995, 1999) divides the known
factors up in a different way, separating them into (a) perceiving, (b)
understanding, and (c) prediction factors, and naming the resulting categories
as follows .....
"Level 1
SA": This level of SA contains the factors which affect the efficiency of
the perceptual process.
"Level 2
SA": This level of SA contains the factors which promote an accurate current
understanding of the world.
"Level 3
SA": This level of SA contains the factors which allow for accurate
projections of future states or events.
However, Endsley
then points out that it is simultaneously possible to divide up a pilot's
declarative knowledge base (that is to say, Dennehy
and Deighton's first meta-category) according to the type of knowledge
involved, as follows .....
Geographical SA: This knowledge
domain stores facts about the physical world, for example, the heights of
mountains, the nature of the terrain, the locations of alternative airfields,
the direction of the prevailing wind, etc.
Spatial/Temporal
SA:
This knowledge domain stores facts about the four-dimensional physical world
(the three spatial dimensions, plus time), for example, aircraft position and
speed, time into mission, etc.
System SA: This knowledge
domain stores facts about system states, for example, engine, control, or
instrument malfunctions, fuel status, etc.
Environmental SA: This knowledge
domain stores facts about weather and visibility, together with such things as
restricted (keep out) or prescribed (keep in) airspace.
Tactical SA: This knowledge
domain stores facts about the identity and capabilities of all other units in
the vicinity (and, if military, that will include their combat intention as
well).
Endsley brought all these concepts together
into what he presents as an all-embracing theoretical model of SA, as now
reproduced in Figure 2.
|
Figure 2 - Endsley's (1995, 1999) Model of Situational Awareness: Here is Endsley's version of the interaction between person and environment. The green-highlighted box represents the broad spread of the pilot's higher cognitive functions, and incorporates all three levels of SA (light blue panel, centre), plus his/her behavioural selection and execution capacities (see the "decision making" and "performance of actions" boxes). This higher cognitive resource is seen as being influenced from two directions, namely (1) the person (see the "individual factors" caption, bottom left), and (2) the environment (see the "task/system factors" caption, top left). The yellow-highlighted box (bottom right) represents more basic cognitive resources, and this includes the three main categories of knowledge, namely (a) procedural knowledge (the "information processing mechanisms" box), (b) declarative knowledge (the "long term memory stores" box), and (c) skill (the "automaticity" box) [this cross-mapping is ours, and approximate]. Note how matters of complex understanding ("goals and objectives") and misunderstanding ("preconceptions" and "expectations") have been given a box of their own (bottom left). This is not without justification, given the number of real life incidents in which preconceptions and false expectations have remained fatally uncorrected. Note also the four red "DANGER ZONE!!" panels. These mark points where the system is particularly vulnerable to error, and will be referred back to by subsequent discussion. If this diagram fails to load automatically,
it may be accessed separately at |
|
Enhanced from a black-and-white original in Endsley (1999; Figure 11.1) After Endsley (1995). Danger zoning overlays ours. This version Copyright © 2003, Derek J. Smith. |
6 - When SA Fails
Endsley (1995, 1999) believes that the
factors which most predict and limit SA performance are attention and working
memory (WM) capacity, two of the most important functional elements within the
"Information Processing Mechanisms" box in Figure 2. Endsley regards the limits of attention and WM as being
largely constitutional, set ultimately by the fundamental layout of our nervous
systems, and thus at best slow to improve with training. This takes us to the
first of the danger zones overlaid onto Figure 2, namely "OVERLOAD
ERRORS", and
this is how Endsley assesses the risk so far as
attention is concerned .....
"Direct attention is needed for perceiving
and processing the environment to form SA, for selecting actions and executing
responses. In the complex and dynamic aviation environment, information
overload, task complexity, and multiple tasks can quickly exceed the aircrew's
limited attention capacity. Because the supply of attention is limited, more
attention to some information may mean a loss of SA on other elements. The
resulting lack of SA can result in poor decisions leading to human error."
(Endsley, 1999, pp260-261; emphasis
added.)
SA can also be adversely affected by
such factors as poor communication, fatigue, stress, both overhigh
and overlow workload, inappropriate expectation, and
even the feeling of being pressured to "press on regardless". Endsley (1999) summarises a
number of earlier studies to provide an epidemiological view of SA errors, by
level and subtype. This is summarised diagrammatically
in Figure 3.
|
Figure 3 - The Epidemiology of Disasters: Here is a breakdown of US air accident statistics from Jones and Endsley (1995), broken down by SA Level and specific cause. Failures at SA Level 1 constituted 80.2% of the total, and fell into the five subtypes shown. The worst single culprit (at 37.2%) is information which is available to the pilot or crew, but which is not attended to for some reason (eg. due to distraction or excess workload). Failures at SA Level 2 include using "the wrong mental model" to interpret perceptual data, and this factor thus includes the sort of expectancy errors seen in the USS Vincennes [details] and Kegworth disasters [details]. If this diagram fails to load automatically,
it may be accessed separately at |
|
Enhanced from a black-and-white original in Endsley (1999; Figure 11.2). After Jones and Endsley (1995). This version Copyright © 2003, Derek J. Smith. |
However, Sarter
and Woods (1991) warn that there is no such thing as the most important factor.
This is their argument .....
"Attempts to define the critical contents
or components of situation awareness in general suffer from the fact that,
given the dynamic environment of the flight deck, the relevance of data and
events depends on their context, [and] will therefore vary within and between
flights" (p47).
Hitchcock (1999) adds that despite the
clarity of the data presented in Figure 3 we should nevertheless interpret the
phrase "pilot error" wisely, even where there has been a demonstrably
incorrect control action. This is because most such errors are in fact
"mistakes waiting to happen" (Hitchcock, 1999, p311); it is just that
the pilot in question was simply the person in the hot seat when things finally
went wrong. Mistakes waiting to happen enter the system via the "DESIGN AND
MANAGEMENT ERRORS" and "SELECTION
AND TRAINING ERRORS" danger zones shown in Figure 2.
7 - SA and Workload
Jensen (1997) is typical of recent
work in this area. She warns that there is no satisfactory single measure of
workload, especially for cockpit or process control tasks, and, like Dennehy and Deighton, has been forced to rely on subjective
ratings. Working at the Farnborough DERA (= Defence
Evaluation and Research Agency), she has combined self-report data with frame-by-frame video analysis
and instrumentation data, and finds that if input workload, central processing,
and output workload are monitored separately, they are found to respond
differently to different types of event. In a simulated helicopter-helicopter
combat, she found that central processing surges when tactical decision making
- the command element of command and control - is called for.
Another way for workload to diminish
performance is for it to impair communication. Two of Huber's (1982/2003 online)
laws of organisational communication are worth
quoting on this point .....
"The probability that a
message will be transmitted from a unit is inversely related to the workload of
the unit" (p143; Proposition R2).
"The probability or extent of message
modification [later defined as the "undesired reduction of message
complexity"] is positively related to the extent of the sender's work
overload" (p148; Proposition M6).
8 - SA and
Situation Assessment
Prince and Salas (1997) have studied
how SA is formed in the first place, and then subsequently kept up-to-date.
They argue that if SA is a complex mental state, maintained by a complex of
cognitive processes, then problems with the updating processes are likely to be
just as critical as problems with the state itself. Prince and Salas call the process(es) of continuous updating
"situation assessment", and characterise
the relationship between awareness and assessment as follows:
"Within the context of normal flight,
situation assessment must occur continuously to ensure situation awareness. In
addition, [] when a decision event arises, situation assessment must serve the
decision process." (Prince and Salas, 1997, p291).
In their subsequent discussion of
the awareness events, they cite a paper by Adams, Tenney,
and Pew (1995), which emphasised the role of top-down
influences in directing exploration of the environment. The two things go hand
in hand .....
"..... the crew must have a good
understanding of their present situation so they can assess the stimulus event for
relevance, urgency, procedural, and goal related implications" (Prince and
Salas, 1997, p293).
And in their discussion of the
central decision making, Prince and Salas draw on the work of Lipshitz (1993), Klein (1993), and Noble (1993). Lipshitz had compared various competing theories of "naturalistic
decision making", and found that situation assessment was a major
factor in most of them. He defined the process as a "sizing up and
construction of a mental picture of the situation" (Lipshitz,
1993, p132). Klein had discussed "recognition primed decision
making", where "experience allows the decision maker to identify
a single good option for action" (p292), and where assessment of the
situation is critical to deciding which action from among many counts as
"good". And Noble had introduced the term "reference
problems" to describe memories for problems previously solved,
proposing that they were indexed within the memory network on three major
dimensions, namely (a) objective, (b) action, and (c) environment. Thus, one
might choose the objective of "escape" rather than
"surrender", the action of "run to the left" rather than
"run to the right", and the environment of "jungle" rather
than "desert".
9 - SA, Risk Space,
and Time to Contact
Another insight into situation
assessment comes from considering humankind's evolutionary past. We evolved in
a world full of predators, where being surprised would often mean being eaten
and where only the fittest survived to breed. As a species, we responded to
this challenge by going for cognitive processing power and teamwork rather than
sharper teeth and longer claws. The behaviours which
helped you make it through the day included bonding and social altruism, and
the cognitive capacities included predictive feedback systems, motor schemata,
the powers of abstraction and symbolisation, learning
from experience, effective problem solving, plus one which is closely related
to what we now call SA, namely "time to contact (TC)"
processing. Smith, Lewin, and Hancock (1997) summarise
the problem thus: "The detection and avoidance of collisions in a
congested space is a problem that all independently mobile agents must solve
quickly and efficiently" (pI.229). They then suggest that the single most
valid factor in predicting safe navigation is ultimately an organism's estimate
of TC. They studied this variable in commercial airliner
pilots, and found that conflicts (ie. potential
collisions) were consistently detected with no less than three minutes to spare
as they entered a subjective "risk space". This indicates that
experienced pilots code all items entering their risk space on a TC
basis. There is however little in the literature on what the upper limit to
this sort of target tracking is. As to the peripheral mechanisms involved, the
work on "optic flow" fields may be implicated (see, for
example, Koenderink, 1986, and Van de Grind, 1988, if
interested).
10 - SA and the
Mental Model
The modern concept of the "mental
model" has been very heavily influenced by the work of Philip
Johnson-Laird, Stuart Professor of Psychology at Princeton University.
Johnson-Laird wrote his first major paper on the subject in 1981
(Johnson-Laird, 1981), and the monograph "Mental Models" in 1983
(Johnson-Laird, 1983). Nevertheless, the term goes back at least to the 1940s,
for it appears in Craik's wartime papers, thus .....
"[While] as a rule mere objects evoke, in
themselves, little response, anything which in the light of our previous
knowledge puzzles us or defeats our insight or suggests new possibilities may
evoke a very definite response. We want then to find some scheme in which our
various experiences combine to produce patterns in us [..... which .....] can
become developed to the point where their relevance to one another becomes
functional; in consciousness this is indicated by ideas 'striking' us []. We
then have a mental model of a possible event in the external world
....." (Craik, undated pre-1945, in Sherwood, 1966, p72;
emphasis added.)
Similarly, nearly half a century
later Endsley (1999) used the term "internalised mental model" in his definition of SA
(see Section 2), and in explaining the influence of memory on current control behaviour. This is how he states the relationship:
"During active decision making, a pilot's
perceptions of the current state of the system may be matched to related
schemata in memory that depict prototypical situations or states of the system
model. These prototypical situations provide situation classification and
understanding and a projection of what is likely to happen in the future (Level
3 SA)." (Endsley, 1999, p263.)
But are Johnson-Laird's mental
models and SA's "internalised mental
models" one and the same thing? Well it appears not. The problem is that
there are several qualitatively different types of memory, all interacting to
add content to consciousness. Johnson-Laird and his followers have largely
studied verbal reasoning and the mental model they have modelled is different
from the sort of three-dimensional spatial intelligence measured by SA
researchers. Helge Helbing, of the Technische Universität, Berlin, has studied the latter process in air
traffic controllers (Helbing, 1997). The basic problem, Helbing asserts, is
that while we know a lot about the mental structures which are activated during
text comprehension, "we know relatively little about the cognitive
representation of spatial information" (pI.177). Helbing goes on to cite
earlier work by Whitfield and Jackson (1982), who had argued that what is most
important to controllers is the "picture" they form and maintain.
Helbing summarised Whitfield and Jackson's approach
in the diagram now reproduced in Figure 4.
|
Figure 4 - The Mental Picture: Here is a 2S-O-2R variant of the classical S-R
diagram [reminder],
in which one of the main functions of central processing is characterised as
maintaining a "picture" of what is going on in the world. Note
how this diagram is structurally very similar to the core layout of the
"transcoding models" commonly used within modern psycholinguistics
[details]. The central understanding of the world - the
"picture" shown below - is referred to therein as the "Semantic
System" [see, for example, the Ellis and Young (1988) transcoding diagram]. If this diagram fails to load automatically,
it may be accessed separately at |
|
Enhanced from a black-and-white original in Helbing (1997; Figure 1). After Whitfield and Jackson (1982). This version Copyright © 2003, Derek J. Smith. |
The work of Dr Dieter Wallach, of
the Universität des Saarlandes,
Saarbrücken is also relevant here. Wallach (1997)
argues that cognitive engineering - applications of cognitive psychology within
the engineering sciences - frequently involves problems of complex problem
solving, and that it is commonly assumed that active interaction with a system
is necessary to develop the mental model of that system required by its
controller if s/he is to control it. But again different memory types allow
different levels of performance. Wallach carefully distinguished between knowing
how something works (which is deep conceptual knowledge) and knowing how
to work something (which is far shallower practical knowledge), and
stressed that the latter is sufficient only until things start to go wrong.
Wallach's research therefore investigated the effects of different types of
training activity upon the development of control room skills. The test
environment was a computer simulation of the control room of a coal-fired power
station, and the subjects were 40 engineering students divided into two groups.
the first group - a system control group - was
allowed to explore, interact with, form and test hypotheses about the system.
The second group - a monitoring group - was only allowed to observe a
trained operator at work: that is to say, it was given no "hands-on"
experience. A battery of tests was then used to assess system knowledge
(declarative knowledge) and control performance (procedural knowledge). Results
indicated a significant superiority of the monitoring group, in sharp
contrast to the prevailing assumptions. There was also a strong correlation
between knowledge and performance in the system control group, but none
in the monitoring group.
11 - Shared SA
(SSA)
More problems are likely to arise
whenever SA becomes a team responsibility. This is because teams can all too
easily add to their members' cognitive workload rather than reducing it. This
is because any one individual's mental model has to include other team member
nodes in his/her definition of the situation. A co-pilot, for example, needs a
mental model in which the pilot's mental model is represented.
ASIDE: This is remarkably similar
concept to the "theory of mind" literature which has accumulated in
the last quarter century. The co-pilot is modelling a model [ie. meta-modelling] and researchers are modelling that
meta-model [ie. meta-meta-modelling!!].
Yet again, the idea is a good one
but the science is not yet in place to develop it. Nofi
(2000/2003
online) warns that "the concept of 'shared situational awareness'
[.....] is elusive and ill-defined, and does not lend
itself easily to traditional scientific evaluation" (p1). Nevertheless, he
manages to distill a dozen or so competing definitions into the following .....
"..... it
develops by a process of integrating the mission-essential overlapping portions
of the situational awareness of individual team members - thus developing a group
dynamic mental model" (ibid, pp1-2).
"Crew Resource Management"
(CRM) [previously Cockpit
Resource Management] was introduced in the early 1990s to help overcome the
limitations of individual understanding by broadening the concept of command
and control to bring all flightdeck and cabin crew
(and, on occasions, the passengers themselves) into the loop. [See our Transportation Disasters -
Aerospace Database, under the headings Sioux City and Kegworth, for these are the disasters which most
directly prompted this initiative.] And if CRM is as good as it has been
claimed to be, then it follows that there should be measurable improvements in
SSA as a result. This issue is still under investigation, and there is little
available data at present.
Nor can the definition of "the
team" even be restricted to those on the aircraft. Hill (2003 online)
explains how traditional flight training fails to explain how to tell the
important clues from the unimportant, given that flying exposes pilots to
constant activity centred around
their radio interaction with Air Traffic Control (ATC). He points to a
particular weakness in training procedures, to the effect that pilots are
insufficiently aware of how controllers do their job. This leaves them unable
to see behind the words emanating from their earpieces,
and thus less able to spot dangers in advance. Hill supports this argument with
material taken from the radio interactions immediately prior to the 1991 Los Angeles
International (LAX) collision between a US Air 737 and a Skywest
Metroliner [details]. They established that the Skywest crew failed to note ongoing radio exchanges between
the tower and the 737, even though they clearly stated that the 737 was intending
to land imminently on the same runway they were about to take off from. They
had, moreover, asked to join the runway from an unusual taxiway, and had been
warned to wait for other ground traffic to cross in front of them. The end
result was that none of the three separate control teams (that is to say, the
737 crew on their final approach, the Metroliner
preparing to take off, and the controllers) had a comprehensive enough picture
of the real world, only their confused assumptions. The 737 struck the Metroliner a few moments after touching down, and 34 people
lost their lives [picture].
Hill's conclusion is that you must take your time and deliberately go as far beyond
the minimum standards set by others. And yes, if that means putting a lot of
effort into being your brother's keeper all the time, then so be it, because
that is how the world works, and it is a lot better than getting fried.
12 - SA and
Automation
Endsley (1999) warns that it is all too
easy for SA to be reduced by task automation. This is because it takes
the human "out of the loop" (p267); worse - it renders them less
capable of recovering effective manual control should the automation fail. This
factor was clearly implicated in the 1987 Detroit MD-80 air disaster [details]. It is not all
bad news, however, because automation of cockpit information flow, especially
in the direction of greater integration and more efficient man-machine
information transfer, can actually increase SA (although again you have to
design in recovery pathways should the automation fail. Isaac (1997) adds
that you also have to design so as not to disrupt visual search (because the
visual iconic system relies on a memory type which is very sensitive to
interference from near-simultaneous inputs) and minimises
unnecessary delay (because short term memory is sensitive to decay).
13 - SA Outside the World of Aerospace
UNDER CONSTRUCTION
14 - Training
Implications of the SA Concept
Sarter and Woods (1991) warned that SA
cannot be sustained over time unless pilots and crew are sensitive to minor
changes in the massive volume of data presented to them by their instruments.
They pointed to the fact that pilot training regularly simulates such
"instant-onset failures" as engine failures, but devotes far less
effort to training at detecting gradual deterioration of one data index amongst
many (the same slow needle-creep which regularly causes motorists to run out of
fuel or boil dry). Sarter and Woods refer to this
type of event as "subtle problems".
Other researchers have studied
process control industries such as power generation, petrochemicals, and steel
manufacture. For example, Kluwe (1997) looked at the
effects of different training methods on acquiring an industrial control
performance and contrasted the two basic philosophies of training, namely learning
by being told and learning by solving problem. The commonly
claimed advantages of the latter, he claimed, were often more assumed that
empirically demonstrated. The test environment was a computer simulation of an
asphalt mixing plant, and 42 subjects took part. Initially, all 42 subjects
were given equal experience with the system's training manual. Then they were
split into two groups. Group 1 was the learning by being told group, and
group 2 was the learning by solving problems group. Both groups were
then given different experience for 6 days, and then tested. Results showed
that the levels of achieved system knowledge did not differ, nor did control
performance "on a global level". However, at a finer level of
analysis it emerged that Group 1 was inferior when (a) subsequently having to
acquire additional control skills (i.e., they were not as good at
extending their knowledge), and (b) being required to modify their control
strategy (i.e., they were not as flexible). This has considerable impact
upon any training regime intended for operators in changing control
environments.
15 - References
See the
Master References List
[Home]