Development of skill, reduction of workload

Suggestions about the mechanisms ('types of skill') underlying different types of behaviour.  Focussing particularly on :

- perceptual-motor skills,

- familiar cognitive skills,

- problem solving skills,

- flexibility of behaviour.

Note : I use the word 'skill' when talking about the development of ability from beginner to expert, not to label a particular type of task processing.

This paper is very complex, as is the behaviour it is trying to describe.   So I have made many small revisions to (hopefully !) make it easier to understand.  In this case I have not used [ ] to distinguish revisions and additions from original text. 

This paper was written as a book chapter which had, in part, the aim of integrating other material in a book. The book / volume referred to throughout is Bainbridge and Ruiz Quintanilla (eds.) Developing Skills with Information Technology. Wiley, 1989.


1. Introduction.

2. Examples of cognitive activity.

2.1. Following simple written instructions.

2.2. Parallels with perceptual-motor skill.

2.3. Following a more complex instruction.

3. A simple schema for types of skill.

3.1. The main structures in the schema : frames structuring activity, levels of specificity; types of knowledge of the environment; working memory.

3.2. Types of skill : perceptual-motor skills; familiar cognitive skills; problem solving skills using frames of activity + knowledge bases.

4. Flexibility of skill.

4.1. Types of skill not represented in the schema.

4.2. A mechanism for flexible skill : 'means' as a frame/pattern of sub-goals; skills of behaviour choice - alternative methods, meta-knowledge about methods.

5. Additional implications for workload.

6. Conclusion.


Development of skill, reduction of workload

Lisanne Bainbridge

Department of Psychology, University College London

In Bainbridge, L. and Ruiz Quintanilla, S.A. (eds.), Developing Skills with Information Technology, Wiley, 1989, pp. 87-116.

1. Introduction

Equipment based on information technology often provides flexible facilities which can be used for many purposes. This flexibility puts the emphasis on the user’s cognitive skills. Classical training methods teach standard pre-specified sequences of behaviour, but these can now be programmed, so what the user is now asked to do involves problem-solving and understanding. To help people acquire these sorts of skills we need a change of emphasis in our understanding of skill, putting the focus on adaptability and cognitive aspects rather than simply on 'automatic' behaviour.

The simplest general definition of 'skill' is that it is the efficient use of appropriate behaviour which develops with experience. A skilled person does a task effectively and with minimum effort. In this way, the notion of workload is intrinsically linked with that of skill. Beyond this simple definition, we will see that the notion of skill is very complex.

The aim of this chapter is to provide a simple framework for understanding the nature of cognitive skill and flexible behaviour. We will start by describing some simple examples of cognitive activity, as they illustrate some of the constituents of cognitive skill and its development. We will then consider a simple schema for the relation between different types of skill, and will discuss a possible basis for the flexibility of skill. We will mention how workload changes with the development of skill, and also the control of, or response to, workload as an aspect of skill. The practical implications for training will be outlined in another chapter (Bainbridge 1989).

2. Examples Of Cognitive Activity

To illustrate cognitive behaviour, we will look at two examples, one simple and one more complex, of what is involved in following written instructions. This sort of example has been chosen because it is often assumed that following instructions is a 'mindless' activity in which behaviour is fully pre-specified, while actually following instructions for the first time uses knowledge and can involve problem-solving. These examples will illustrate cognitive mechanisms in operation in a simple situation, and illustrate the development of cognitive skill.

2.1. Following simple written instructions

Although an instruction may look like a fully specified account of what should be done, for a novice it is often a statement of a goal to be achieved, rather than a command whose meaning is obvious.

Suppose someone has bought a new hi-fi. The instruction book contains the message: 'Turn antistatic knob to obtain the best sound quality'. How does someone go about following this instruction? As they have never used this equipment before, they have to find the required control and devise a method for changing its position. They then have to try this plan out, listen to whether it has the expected effect, and if not think of something else to do. A suggested schema for this is shown in Figure 1. The main sequence of activity is shown down the left : the main goals are met by doing subsidiary activities which, in the diagram, are linked to them by stepped arrows. The knowledge bases referred to during understanding the instructions are in boxes on the right (there would be additional knowledge bases referred to about how to identify that the effect is changing successfully).

Figure 1 : Sequence of activity in following an instruction for the first time.
(KB = knowledge base, m/c = machine.)

To describe this processing in more detail : the person has to work out and then use a plan of action. First there is a need to interpret the sentence, by referring to knowledge bases about the likely meanings of the words. A user with appropriate prior knowledge can then make some best guesses about three things: that the object of interest has a certain shape and label and will be in a particular area, that a particular hand posture and movement will be needed, and that the changes of sound quality will be of a certain type.

The user makes a plan of action based on these best guesses, and tries it out. The first problem might be that the user cannot find a knob labelled 'antistatic', so might look for knobs with labels with related meanings, or for non-knobs with this label. If the control turns out not be a knob at all, they will have to change the type of movement planned. If the sound does not apparently change in quality when this control is turned, then they also need to try something else or go back to the instruction book. Once they have found a plan that has the expected sort of effect, they can go on using it until they find an optimum sound. Plans that at first have the effect of reducing the distance to the goal, but finally do not actually get to it, are the ones that are the most difficult to relinquish or revise.

In this simple sort of situation, if most goes as expected : someone learns very quickly what to do, as well as specific knowledge about the components of the device and how to act towards it to get particular effects. Thus the cognitive processes needed after experience are simpler, as suggested in Figure 2. Finally, the actions will be done from memory, perhaps without much conscious attention.

Figure 2. Activity in following an operating instruction when the user has some knowledge about the device.

What general points can be extracted from this about the nature and development of skill? Even a very simple activity can involve problem-solving during the first attempts. The 'simplicity' of an activity lies not necessarily in how simple it is at first but in how simple and undemanding the cognitive processes involved can become after a little experience.

After practice, the user no longer needs to devise a plan of action, or to test and revise it. Figure 3 suggests a schema for revising working methods on the basis of experience. After practice, the person knows which methods have which effects, so chooses an appropriate action first time; only the part of Figure 3 above the line may be needed. That is, once the person has situation-specific knowledge, they can do the task in a different way, different cognitive processes can be used and the goal is reached more quickly and more accurately. For both reasons, the person has less workload.

Figure 3. Developing skill by acquiring relevant knowledge.
Once the effect of an action is known, it is possible (in a stable environment) to choose an effective action without revising either the execution of the action or the action knowledge base.

A human factors/ ergonomist has three approaches to optimising this transformation of task thinking from complex to simple, by the design of : interfaces, operating manuals, and training schemes.

1. A basic ergonomic principle is that information should be compatible with the task, as well as easy to see. In this case instructions and interface have to be compatible (for example 'knobs' are knobs), so that the prior assumptions that people bring to interpreting the instructions do not have to be revised.

2. The instructions need to be both goal-centred and laid out in such a way that it is easy to pick out the key notions and the sequence of activity.

3. Task analysis for training should identify (see also Section 2 of this volume) what the user needs in a knowledge base (the relevant repertoire of actions and information about machine properties), without which it will be difficult to develop device-specific skills. 

It is these prior knowledge bases which college-based training schemes attempt to identify and develop (for example see Nijhof and Mulder, Chapter 6 of this volume, and Diaper and Johnson, 1989).

2.2. Parallels with perceptual-motor skill

At the most general level, the development of cognitive skills and perceptual-motor skills can be very similar.

The classic example of perceptual-motor skill is given by Miller, Galanter and Pribram (1960, Figs. 2-5). Movements change from being monitored visually to being monitored kinaesthetically; that is there is a change in the 'resources' used as skill develops. The sub-units of behaviour also build up into a nested hierarchy (though this notion is inadequate for representing adaptable skills, as we will see later).

Figure 4 shows what can happen (unconsciously) during the development of control skill. At first the person does not know the best time lag and control gain to use in choosing a control action. However, after experience, the effects of actions may be known so well that the first action chosen is effective, so it is no longer necessary to revise the way actions are chosen. Simultaneously this means, in a stable environment at least, that there is no need to check the result of the action, because an action has been chosen that is known to have the required result. The workload is therefore reduced in two ways (Bainbridge, 1978). Figure 4 is similar in principle to Figure 3.

Figure 4. Developing control skill (From Bainbridge, 1978).
lower loop : controlling the device.

upper loop : learning what actions have what effects.

2.3. Following a more complex instruction

We will add to the concepts that we need to account for cognitive skill by considering a more complex example of following instructions.

Suppose an operator following a fault management procedure in a nuclear power plant reads the instruction: 'Close initial steam valve (RAXXSOO3) of SG in question'. What do they need to do to respond adequately to this instruction? (SG = steam generator).

Norros and Sammatti (1986) have shown that operators need to use intelligence when following the procedures for operating a complex plant. They cannot simply follow instructions but have to think for themselves. For example, the valve might be stuck open, or there might be two faults and the procedure for the other fault might say that this valve should be kept open. In either case, the action specified by the procedure is not available, so the operator must think of an alternative way of achieving the aim which would have been achieved by closing the valve.

Figure 5 suggests a simplified representation of the processes involved. This is a complex diagram, but it has the same general layout as Figure 1. The main sequence of activity is shown down the left; the knowledge bases referred to are in large boxes down the right.

The small squares make an additional feature explicit, the working memory used. 

Squares with a diagonal represent items that are already known from previous activity.  

An empty square indicates an item that is not known, so there needs to be subsidiary activity to find it - referring to a knowledge base, or information seeking and action making.

Figure 5. The cognitive activity and knowledge bases which may be used in following part of an industrial process operating procedure.
(SG =steam generator).

There are several general features of this representation.

The top left of Figure 5 indicates the main goal of the activity. In this case it is to get the plant into a safe stable state. In many industrial processes the written procedures do not say what the purpose of an activity is, so it can be difficult for the operator to work out whether a procedural step is appropriate in a particular incident.

The general framework for following a step in a procedure is shown down the left of Figure 5. The sequence is:

1. Devise a plan, that is (sequence of) action(s) to be made.

2. Check whether action: is available? will have the required effect?

(Operators do check the applicability of procedures in this way; see Pew, Miller and Feeher, 1981, on the Prairie Island incident.)

3. If the action is not appropriate, find another action that will have the required effect.

4. Execute the action(s).

5. Monitor the effect. (Even highly experienced operators must monitor in a fault situation, because the behaviour of the process cannot necessarily be predicted from how it has responded to this sort of action in the past.)

6. Assess whether the goal has been met and revise the plan if not.

We will now describe these activities in more detail:

1. The first step is to understand the instruction and make a plan (as in Figure 1). The plan is built up from information in two knowledge bases. 

One is a knowledge base (KB) about the physical structure of the plant, including the positions of valves, either the actual location of this valve on the plant (or of its associated control on the interface) or general knowledge about likely locations of the valve/control. The other KB is the operators' knowledge of their own actions, how to reach to and move the valve, or the social or communication skills involved in asking the shop-floor operator to do so.

The plan of action built up may itself consist of a hierarchy of subframeworks, each of which is a set of subgoals which are themselves met by subsidiary methods. For example, the framework of hand movements, at the lowest level in an action to change a control position, might be "aim, acquire, move, release", and each of these might involve subsidiary perceptual motor skills. 

This description uses the word 'frame / framework' as a label for a group of activities which meet a goal.  In this goal-means representation, there is a frame of slots, the 'goals', to be filled by carrying out sub-frames (sub-goals), the 'means'. (See more about this in Section 4 on flexibility.)

The plan will be built up, not necessarily in the sequence in which it is executed, in working memory. Working memory contains not only information about the actual state of the external world but also information about future activities which are being considered (Bainbridge, 1974a, 1975).

2. The second step is to check the appropriateness of the action planned

One aspect of this is to check whether the action is available. This would probably be done while the plan is being developed, if working memory contains information about the recent status of the valve.

The second aspect is to predict the effect of the action. This effect could be modelled in working memory (in the strict sense of the term 'mental modelling' as originally used by Craik, 1948). The effect would be predicted by referring to knowledge about process behaviour, in particular how it responds (output) to specific control actions (input). The prediction could be made on the basis of specific knowledge about this particular valve, or general knowledge about the effects of valves, or from analogies with other devices that affect flow, such as taps.

The predicted effect is then assessed against the required state.

3. If the action is not available or not suitable, the operator needs to think of another one. They could imagine (in working memory again) the trajectory of process states needed to reach the required state and then search for an action that has this effect. This search would use a knowledge base of information about process dynamics which is organised in the following direction: required process state - to - action that causes it. Note that this cannot be done simply by working backwards through the cause to effect (or input to output) knowledge base used in step 2, as human knowledge bases are not automatically reversible.

Note that the whole framework includes two inverse types of activity. One is to take a given action and predict and assess its effects (steps 1 and 2). The other is to formulate a goal state and devise a plan for reaching it (step 3). The latter would classically be called problem-solving, but actually does not always involve the development of new working methods. We will discuss this further in the next section where problem-solving skills are discussed.

3. A Simple Schema For Types Of Skill

We will now generalise the mechanisms, which were suggested in these two examples, to give a simple schema or model for the processes and knowledge underlying skill.

The aim of modelling is to provide a framework for thinking about a particular category of problems. The simple schema discussed in this section (Figure 6 below) suggests a structure for understanding the relation between different types of skill. As we will see in the next section, several important types of cognitive skill are not represented in this schema, and we need to know some operating details about parts of the schema before we can draw inferences about its practical implications, for example for training.

Figure 6. A schema for the main mechanisms underlying skill (see text).
(Responses : aR = attention response, cR = cognition response, pR = physical response).

Note :

- perceptual-motor skills bypass/ don’t need conscious attention, i.e. use of working memory,

- familiar cognitive skills bypass/ don’t need problem solving about what to do.

We will outline the main structures in this schema before discussing the types of skill it summarises.

3.1. The main structures in the schema

The schema includes:

1. Knowledge of one's own activities, and a general problem-solving framework.

2. Knowledge of the structure and behaviour of the environment.

3. Working memory.

These have all been used in the Section 2 examples of 'following simple and complex instructions'. We need to make some more general points about each of these before discussing types of skill.

3.1.1. Frameworks of activity, 'levels' of specificity

We have suggested that there is a standard pattern of activity in following operating procedures (as in Figures 1 and 5), a 'frame', 'script' or 'strategy' for this general category of task. The 'framework' used here both behaves like a series of instructions and exists like a simultaneous pattern.

Such a framework for a cognitive activity could be at any one of (at least) three 'levels' of difficulty of cognitive skill (using the word 'skill' in the sense of being able to draw on existing frameworks for activity, which make it easier to respond to given task requirements). Note that these three 'levels' are not distinct categories but points along a continuum. The three are:

1. Familiar specific situations. The framework is a standard method for dealing with a frequently recurring specific situation - which is sufficiently stable for the same method to be used each time, though it may vary in detail, for example choosing the furnace to alter when allocating electric power between steel-melting furnaces (Bainbridge, 1974a, fig. 2).

2. Familiar general tasks. The framework for a familiar general category of task, such as following a procedure, can be applied to unfamiliar specific situations in this category. This general framework can be skilled, because the operator can have frequently practised the general category of task. Figures 1 and 5 show a general framework for understanding, assessing and revising instructions.

3. Unfamiliar tasks. More unfamiliar tasks can be dealt with by more general problem-solving and planning skills. An example is the emergency management planning described by Samurçay and Rogalski (1988). This general problem-solving framework will be discussed further in the next subsection when problem-solving skills are dealt with.

In a well-known specific situation, three frameworks for activity could be available: a method specific to this task, a general method for this category of task, or a general problem-solving method. How is the working method chosen? Two general aspects of the way in which these three are invoked are usually proposed. 

The first is that any situation is dealt with by passing through the three 'levels' (for example Rouse, 1983) : 

a situation-specific method is used if available,

a task-specific method is only used if a situation-specific method is not available,

and a general method is only used if a task-specific method is not available. 

The second general assumption is that, once someone has an extensive vocabulary of situation-specific or task-specific methods, then they are likely to try to apply one of these rather than to use a general method, and this can lead to restricted reactions to problem situations (for example see Leplat, Chapter 2 of this volume).

3.1.2. Types of knowledge of the environment

The cognitive activities in Figure 5 referred to several types of knowledge base:

1. Plant/ environment knowledge: about its physical structure, and functional, structural and other constraints on operation, in the directions input-to-output, or output-to-input.

2. Operator knowledge: sub-frameworks for achieving subgoals.

Each of these types of knowledge was referred to by a different part of the activity represented in Figure 5.

Developing a cognitive skill does not only consist of developing methods for doing tasks and of acquiring the necessary reference information. It also involves developing appropriate structures and categories for the knowledge, and developing rapid access from a framework of activity to the part of the data base that it refers to.

It is useful to make another distinction between two types of background information/ reference data:

- standard data about what may occur in a task situation, for example the possible states a steel furnace can be in,

- knowledge of the structure and behaviour of the task environment, on the basis of which one can understand events, explain the reasons for one's behaviour or think out new working methods.

If the task situation is very stable, this latter general knowledge about the properties of the environment, which may formerly have been used in developing the working method, may no longer be actively required for generating new behaviour methods, although it may still be accessible for giving explanations. If the environment is so stable that new methods do not need to be developed, then this information may become progressively less accessible (see Leplat, Chapter 2 of this volume).

Knowledge bases which describe the properties of the environment will not be discussed further in this chapter. For more information see Bainbridge (1988).

3.1.3. Working memory

The final important mechanism suggested in the diagram is working memory. The contents of this working memory are not, as in laboratory studies of working memory, an exact replica of external data. Instead this is a working space in which data structures are built up and then provide a context for choosing appropriate behaviour (Bainbridge, 1974a, 1975; Johnson-Laird, 1983). In complex industrial tasks, and other complex tasks such as air-traffic control, the working memory contains information which is the result of thinking about the task: about required, actual and future states, with associated plans of action and their evaluation.

Probably the concept of working memory that best fits the data on complex behaviour is to consider it as a 'blackboard' (Rumelhart, 1977), in the sense that the information is available in parallel and built up and assessed in parallel, rather than in a sequence of separate steps as implied by Figure 5. However, note that it is a 'structured' blackboard, structured by the framework of the task; that is the 'framework' for the sequence of activity subgoals could also provide the structure underlying the contents of working memory (Bainbridge, 1975).

3.2. Types of skill

It is usual to talk about 'skill' as more 'automated' behaviour (Leplat, Chapter 2 of this volume). However, it can be confusing if one does not distinguish between the different ways in which behaviour can become more automatic, which are classically called:

- perceptual-motor skill,

- cognitive skill,

- problem-solving skill.

These different types of skill can be characterised as using different subsets of the mechanisms in the schema in Figure 6 above. The schema emphasises both the notion of skill as a change in type of processing and the centrality of working memory in the organisation of behaviour. 

A little more about the three types of skill (yet more detail later) :

1. Perceptual-motor skill. A skilled person can react appropriately to the environment without using conscious attention, that is this type of skill just uses senses and muscles and bypasses working memory.

2. Familiar cognitive skill. In familiar situations for which methods and knowledge are already available, a skilled person can do the task without first having to work out a method for doing it, that is this type of skill bypasses the need for problem-solving.

Most cognitive behaviour is concerned with building up working memory, i.e. with actively searching for the information needed to meet goals, rather than with passive reaction to inputs. A framework for activity may involve directing attention to the environment (aR), and accessing or processing knowledge (cR), as well as acting on the environment (pR).

3. Problem-solving skill. When someone does need to devise a new method, or acquire or reconfigure knowledge, these processes too can be skilled in the sense that they can draw on existing frameworks and environmental knowledge, and become easier and more effective after practice.

These three general types of behaviour are not necessarily easy to distinguish in practice. Cognitive behaviour is dynamic. In any task, any one of its substeps might be done in any of these ways. The three mechanisms could be active in different concentrations depending on the familiarity and stability of the environment. This flexibility will be discussed further in the next section. Despite this interdependence, the three categories will be discussed in more detail separately.

3.2.1. Perceptual-motor skill

Within the context of this book, we are concerned with the perceptual-motor skills which are involved in primarily cognitive tasks, such as interpreting and using a complex interface, rather than with the full range of perceptual-motor skills in tasks which can involve :

- exertion of physical effort, in which good ergonomic design and skill development have the effect of minimising and optimising the application of human muscular effort.

- skills of physical coordination which seem to be impossible to describe in words, such as swimming, bike riding, or flying a helicopter.

When using a complex conventional interface, an operator with the relevant perceptual-motor skills can automatically look to the required display, interpret the meaning of a display, move to the required control, etc. without conscious attention to this subtask. This is the domain of much of classical interface human factors/ ergonomics, in which one of the primary principles is to maximise the possibility of using an interface without conscious attention. 

For example, if the value of a process variable is presented on an analogue instrument, deviations from target can be detected by automatic pattern recognition. If the same value is presented as a digital number, finding the deviation from target involves a sequence such as:

1. Locate and read the actual value.

2. Locate and read or remember the target value.

3. Do the calculation to compare (1, 2).

4. Do the calculation to compare (3) with the tolerance limit.

That is, to interpret a digital display requires a framework of activity and considerable use of working memory, which may well disrupt the thinking about the primary task. Therefore an analogue display is often preferred.  (Though there are some tasks in which doing an explicit calculation is part of building up structured working memory, e.g. an air-traffic controller comparing aircraft speeds or flight levels.)

Figure 7. Ways of responding to the environment which bypass working memory - see text below for explanation.

Skilled = learned.

Responses : aR = attention, cR - cognitive, pR = physical.

Figure 7 suggests the three main routes by which we can respond to the environment in ways that bypass working memory/ conscious attention. This bypass is sometimes called 'implicit' processing. In more classic psychological language, Figure 7 shows three types of 'attention mechanism'. The first has a genetic basis. The others depend on learning; they can only be done automatically after considerable practice.

1. Orienting response (top of Fig.7). This is the override mechanism which is built into the nervous system, by which particularly strong (salient) signals attract attention whatever the current task activity. This is an important danger response, which can taken advantage of in the design of alarm and attention getting displays.

2. Active search (Skilled 1 in Fig.7). People involved in goal-directed activity usually actively search for the information they need, rather than passively reacting to information as it arrives. Initiating this search for the required information (by directing attention in the environment, aR, or by accessing stored knowledge, cR) may be an activity/ subgoal within a framework for achieving a higher goal, or it may be in a framework for dividing attention between timesharing tasks.

The reaction to the information attended to may be automatic, using :

(a) A highly practised physical response (pR). This can lead into an 'automatic' sequence of behaviour, if the response results in a signal from the environment which also has an automatic overlearned response, and so on.

(b) An automatic cognitive response (cR). The meaning of the signal is automatically brought into working memory without having to consider or search for it, for example in pattern-based fault diagnosis (Shepherd et al. 1977).

3. Serendipity ('making happy discoveries by accident') (Skilled 2 in Fig.7). This is the type of behaviour in which someone notices and responds to something in the environment that is relevant to a goal which may be at the back of their mind but is not currently active. For example, when someone notices information that explains a previous event which they have not had time to think about, or when they do something on the way from one part of the main task to another (for example Beishon, 1969). (This is the type of shopping supermarkets are designed to encourage.) The automatic cognitive response in this case may be to bring the relevant framework for goal- related activity into working memory.

'Serendipity' typically happens when a person is somewhat disengaged from their main goal, either because they are searching for information or because they are between activities in the main sequence, for example walking from one place to another.

Which way someone chooses, of the two ways of responding to the environment which involve learning (active search or serendipity), might depend on personality type or on the task situation (for example work or leisure) as well as on opportunity.

3.2.2. Familiar cognitive skills

The second main group of skills involves evoking frameworks of cognitive activity which have been practised frequently so there is no need to work out what to do. A skilled person has a library of task-related frameworks, whose availability depends on their frequency of use. Only a subset of the mechanisms in Figure 6 are used, as suggested in Figure 8. While classic interface ergonomics aims to bypass working memory, cognitive ergonomics aims to bypass problem-solving.

Figure 8. The sub-set of mechanisms involved in familiar cognitive skills.

3.2.3. Problem-solving skills

Table I below gives a preliminary classification of types of cognitive skill according to which relevant knowledge bases are available. Problem-solving is required when either activity frameworks or reference knowledge are not available.

Table 1. Types of more complex cognitive behaviour, classified according to :
(a) knowledge available :

column 1 : activity frameworks known/not known (+/-)

columns 2-3 : relevant knowledge base known/not known (+/-)
(b) use of working memory - column 4.

see text for explanation.

Problem-solving can, of course, itself be skilled, if general working methods and ways of structuring information are already available and practised. 
The classification in Table 1 will need to be extended. There are several important types of cognitive process that are not represented here or in the general schema in Figure 6. Some of these will be mentioned briefly in the next section.

The main dimensions of Table 1 are concerned with whether activity frameworks are or are not already available, and whether knowledge about the properties of the environment is or is not available. This defines several main types of task-processing:

1. Goal-activity frameworks + / Environment knowledge + / content +

This is the category for the familiar cognitive skills dealt with in the previous section. 

In the table (column 2) these are distinguished from scheduling/ planning tasks (in the simple sense).  In those, an overall method for deciding on a sequence of events is available but a new sequence has to be devised each time (for example Shackel and Klein, 1976). 

This distinction has been made because, in tasks such as steel-furnace control (Bainbridge, 1974a) the structure built up in working memory is concerned with the actual state of the environment while, in scheduling, the structure in working memory is a model of alternative proposed states of the environment. (Most complex tasks build up both 'real' and hypothetical structures in different proportions.) The standard human factors/ ergonomics method for minimising working memory load in scheduling tasks is to provide an 'external memory' aid for trying out the scheduling alternatives, e.g. Shackel and Klein op cit.

2. Goal-activity frameworks + / Environmental knowledge frames +,  content -

(column 3 in Table) In this case, task-specific activity frameworks are available, but situation-specific reference information is not. This could describe tasks that involve a general area of professional expertise, such as accountancy, in which someone knows what to do and what information they need to know and how to find it. This is work in which the general methods include frames for information search and structuring, which build up a situation-specific data base for decision-making. Other examples are maintenance technicians (Rasmussen and Jensen, 1974) or operators using complex procedures as in Figure 5.

3. Goal-activity frameworks - / Environment knowledge +

This is the situation in which domain knowledge is available, but not frameworks for appropriate activity, for example when working in a familiar domain with a new goal. An example would be dealing with a fault situation in an industrial process, when the usual method of stabilising the plant is not available so the operator has to think of a new plant configuration. Reference information is available, and low-level goal-activity frameworks, but they need to be combined in a new task-specific strategy at a higher level.

It seems that this type of task may be difficult because the availability of domain-specific information may only be apparent. Actually a new goal may require recognising new features of the domain and restructuring existing knowledge. This type of activity therefore does also involve information acquisition and structuring skills, while it seems that existing knowledge bases which have been used frequently become rather inflexible in structure. Restructuring seems to be more difficult, the more familiar (and therefore automatically accessed) the domain specific information. One therefore needs practice in using old knowledge in new ways.

4. Goal-activity frameworks - / Environmental knowledge -

This is what is considered as the true problem-solving situation, in which neither methods nor domain-specific information is available. An example would be learning computer programming for the first time (Hoc, 1987). Norros (Chapter 17 of this volume) describes learning to use unfamiliar complex equipment as 'research'.

The problem-solving task situation is particularly difficult because the person has not only to devise working methods and to acquire the task-relevant data but also to devise task-appropriate structures for it, for example: How do different parts of the equipment fit together, physically or functionally? What are the task-related ways of categorising the information? All the structures must be built up at the same time, often starting from scratch. When a casual user approaches most IT equipment for the first time, they are faced with an unknown machine and poorly indexed reference literature. The instruction books typically do not directly answer the questions: How do I find out how the machine works? How do I find out how to do task x? The user is not sure where to start or where a given piece of information fits in, and may be using inappropriate analogies to understand the equipment (Eason, Chapter 12 of this volume; Hoc, 1987).

Figure 9. A possible general framework for problem solving.

box = storage point for result of activity linked by kinked arrow

Parentheses/brackets - these steps must be done, but not necessarily in a specific sequence.

A general problem-solving framework for devising a new working method might consist of the steps shown in Figure 9. (The figure does not include the processes involved in organising knowledge structures in a task-efficient way.) In this general problem-solving framework, note that :

(a) There may be few constraints on the sequence in which these steps are done.

(b) The steps may be done in parallel rather than in series.

(c) Any step 'later' in the list may require that a step 'earlier' in the list is revised (for example Hayes-Roth and Hayes-Roth, 1979).

(d) Each step could be done by any type of skill.

(e) If either automatic perceptual-motor skill or a more specific cognitive framework is evoked at or within any step, this may bypass the need for some of the other steps.

It is interesting to consider the possibility of training to develop general, rather than task-specific, cognitive skills, for example by practising : 

- developing plans (in the complex sense of devising new working methods),

- acquiring and structuring information for new tasks.

4. Flexibility of Skill

There are two ways in which we will expand this discussion of skill: by mentioning types of skill that are not represented in the simple schema in Figure 6, and by discussing in more detail the nature of behaviour flexibility.

4.1. Types of skill not represented in the schema

There are at least three important aspects of cognitive processes in working situations, for which there is no mechanism in the schema in Figure 6. These are areas that are in need of major research effort, as we understand little about them. Without them, this and any other schema is inadequate.

1. Prototype reasoning. People actually typically reason, not by thinking through a general causal chain but by thinking of an example situation or a past specific situation, and using that as a basis for considering action in the present context (for example Shepherd, Chapter 10 of this volume; Bainbridge, 1981; Rumelhart and Norman, 1981).

2. Timesharing tasks. Most real-life working situations involve dividing attention between multiple responsibilities (for example Beishon, 1969; Page, Heyden and Liere, 1983). How do people decide how to divide their time between several tasks? What makes this more or less difficult? Leplat (Chapter 2 of this volume) summarises some of the points that have been made about this aspect of skill.

3. Unspecified goals. The representation of skill in this chapter is entirely goal-oriented, but how might one account for behaviour in which there is a general goal, such as 'safety', but the dimensions and criteria defining an actual goal state cannot clearly be specified beforehand ? (Bainbridge, 1981; Norman, 1986; Samurçay and Rogalski, 1988). When can the goal state be recognised once it has been achieved, when it is not clearly defined beforehand?

Despite their importance, we will not discuss these aspects of cognitive behaviour. We have also not discussed the more predominantly perceptual-motor skill aspects of the lower triangle in the schema in Figure 6, nor the cognitive mechanisms underlying optimum mappings between different cognitive representations (for example Barnard, 1987).

Any final taxonomy of types of cognitive process should have implications not only for training. It should also be linked to :

- recommendations for interface and job aid design, 

- knowledge elicitation techniques (Bainbridge, 1986), 

- implications for error mechanisms and their prevention.

4.2. A mechanism for flexible skill

A major (although it may superficially appear small) change in emphasis in the basic concept of skill is needed to deal with training and other support for IT tasks.

It has frequently been held, at least since Bryan and Harter (1899), that skill develops by automating subunits of behaviour. When these subunits no longer require attention, then attention can be used for integrating groups of smaller units into larger units. By learning the arbitrary associations between a concatenation of lower-level items, the higher-level unit itself becomes automatic, and so on. 

This leads to two notions, that skill has an underlying hierarchical tree structure, in which both control and conscious attention are at the highest level, and that the purpose of task analysis for training is to identify this pre-specifiable hierarchy.

That account does not contain sufficiently rich concepts to deal with IT tasks. In particular, IT equipment usually offers flexible functions which can be used for many purposes, so it is not possible to give a pre-specified account of the behaviour needed to use it. 

Training schemes and instruction manuals frequently concentrate on individual functions, demonstrating that a given action will have a particular effect. Trainees and manual users then find it difficult to use the equipment, because they have been given information which is both inadequate and the inverse of what they need. The account of the equipment that they should be given needs to be goal-directed at least two levels. (This is an oversimplification; for more discussion of task analysis see Section 2 of this volume.) The users need help with how the functions available can lead to their own goals, which is not necessarily obvious (Norman calls this a 'gulf', 1986). At the lower level, they need the information in the following form: given that you want to achieve this function, this is what to do. This is the opposite to the order in which the information is frequently presented, and human reasoning chains are not automatically reversible.

Compare :

(what manuals typically say)  : if do this - get this result.

(what a problem solver needs) : if want this result - do this.

These difficulties with a simple model of skill imply that we need a model of skill in which the link between a goal and the means of achieving it is a key feature. As an example, Figure 10 represents the main behaviour in Figure 5 (which shows the processes involved in understanding an operating procedure) in a way that focuses on the goals-means relations. We can suggest that the key features of a more flexible account of skill are all aspects of this goal-means link.

Figure 10. The general 'hierarchical' framework of goals and sub-goals underlying the behaviour described in Figure 5.

box = storage point for result of activity (goal) 

kinked arrow = link to method(s) of meeting goal

4.2.1. The 'means' as a 'Gestalt' of subgoals

A method of achieving a goal can itself be a set of subgoals, which in turn refer to further subgoals, until they reach the sub-cerebral level. However, the set of subgoals that achieves a higher goal is not necessarily an arbitrary concatenation of items, but a group of items that together 'make sense' in some overall way. They provide an organising structure for the activity, a sort of 'information processing Gestalt' in which the whole is greater than the sum of the parts.

Although there are higher and lower levels of this structure of subgoals, the structure of 'levels' can be more flexible than a hierarchical tree. Goals can be met by various means, and means can be used for reaching various goals. So the behaviour organisation may be a heterarchy or network, rather than a hierarchy with unique links. In this case the specific 'level' of a particular unit of behaviour cannot be identified without reference to its use in a particular task situation (Bainbridge, 1975).

4.2.2. The skill of behaviour choice

Choosing behaviour appropriate to a given task context may become automatic in very stable tasks (Leplat, chapter 2 of this volume). However behaviour choice is usually flexible, adapted to the particular task context. To amplify, we can suggest that there are four aspects of the skill in making this 'choice' :

1. Alternative 'means'. Alternative frameworks of activity that can meet the goal.

2. Knowledge about the properties of each method. From experience, a person can learn the general properties of each of these frameworks for activity, for example (a) the processing resources it needs, (b) the outcome of the activity. This could be either in general terms such as the level of accuracy achieved and the time taken or (in highly practised and particularly in perceptual-motor skills) the exact output which will be obtained for a given input, on the basis of which the person could work open-loop, without checking for feedback. 

Each method will have this information stored with it, as in Figure 11 (Bainbridge, 1975).

This meta-knowledge about methods can be used to choose between them depending on what is best in the context.

Figure 11. Choice between alternative methods of reaching a goal.
Each method has stored with it information about its general properties and requirements, which is used in choosing the most appropriate method at a particular time (from Bainbridge 1975).

Even general problem-solving methods could have general information associated with them, such as the likelihood of success. This could, for example, make this 'goal to choice between methods' link point the location of the mechanism for 'success' and 'failure' personality types. Several aspects of 'cognitive style', such as working for speed or accuracy, might be located in the choice at this point.

3. Skilled choice between methods. The ability to make a rapid optimum choice between alternative methods, by comparing the known properties of a method with the requirements of the task context, seems to be a skill that can be learned. Sperandio (1972) presents data from air-traffic controllers which suggest that choosing the strategy as a function of task demands, to control the level of workload subjectively experienced, may be done more effectively by more experienced people. 

There may be two aspects of this skill : 

- comparing the learned properties of a method with the requirements of the task context, 

- configuring one's information processing resources to match those required by the activity, according to the level of effort the person is willing to commit to it (Hockey, 1984, 1986).

4. Flexibility in the use of type of skill. There could be flexibility at this goal-means choice point, not only through choosing between available task-specific alternatives but also because it is possible to access other frame works, for example searching for methods that are not usually used in this context but that have some of the appropriate properties, or accessing general problem-solving methods when no suitable specific frameworks for activity are available. Familiarity with finding and using 'unusual' methods can be an important aspect of problem-solving skill.

Access to general problem-solving methods could happen at any point in a framework of subgoals and at any level in the task organisation, if a usual method either has not been developed or is not available in this specific context. Inversely, when using a general problem-solving framework, such as in Figure 9, it may not be necessary to devise a method for any step for which a method is already available. Therefore real behaviour can be a complex mixture of new and existing working methods, according to the details of a particular situation. 

During the development of skill, problem-solving attention may be predominantly given to increasingly 'higher' levels of behaviour organisation as skill develops. Actually such attention may be given to any level below this (see Leplat, Chapter 2 of this volume). 

Once the skill is acquired, conscious or 'controlled' attention could be allocated to any level of task organisation at which there is a problem, while other levels are simultaneously being dealt with 'automatically'. (This condition exists unless the task context for this unit of behaviour is so stable that all task decisions are made in the same way each time, so that internal flexibility is lost. See Leplat, Chapter 2 of this volume, for some implications of this for training.)

5. Planning a sequence of activity, allocating attention between different tasks. This could also make use of the general information about working methods that are stored with them, as indicated in Figure 11.

5. Additional Implications For Workload

The relation between increasing automatisation of skill and reducing workload is well known (Leplat, Chapter 2 of this volume; Bainbridge, 1978). However, this relation is by no means simple. This chapter suggests some extensions of this idea.

The three general categories of skill have been distinguished because the mechanism of reduction in workload takes a different form in each. 

Increasing perceptual-motor skill leads to lower working memory/ attentional monitoring load.

Increasing cognitive skill means that frameworks for activity are already developed, with appropriate frames for, and content of, the information they need to refer to, so there is lower problem-solving load. 

Problem-solving skills may develop when people have experience with, and learn the properties of, the general problem-solving methods, which reduces workload by increasing effectiveness and confidence. We might argue that processing capacity is limited to a  single-channel during problem-solving (in contrast with other types of skill in which impressive amounts of parallel processing can be done), because all the cognitive processing mechanisms need to be held available during problem-solving in case they are needed.

Although it may be possible in simple laboratory tasks to investigate the relation between a task and the load that it imposes on processing resources, in any real task the flexible way in which working methods are chosen in relation to the particular task context means that there can be moment-to-moment changes in the processing resources used, and short-term changes in the effort needed.

If there are alternative methods which are equally effective in meeting the task goals but which require different amounts of cognitive processing and use different resources, then one would not expect to find a monotonic relation between the task demands achieved and either a general measure of the workload needed to meet them or a specific measure of the use of a particular 'cognitive processing resource' (see Leplat, Chapter 5 of this volume; Sperandio, 1972; Bainbridge, 1974b). Therefore, as with other aspects of ergonomic design, it is much easier for an ergonomist to predict ordinal changes in workload ('if' this task feature changes in this way 'then' workload will increase/ decrease) than to make specific numerical workload predictions about the outcome of the interactions of all these many factors.

6. Conclusion

Although many topics have been touched on in this chapter, the main focus has been to outline the general mechanisms underlying skill. After outlining two examples of cognitive behaviour, we suggested that five basic mechanisms are involved :

1. A knowledge base of information about the environment, which is referred to when thinking out what to do to achieve particular goals. For someone who is skilled, the information is organised in task-relevant groupings, and easily accessed.

2. A knowledge base of standard 'frameworks' for activity. Within these the goal-method link is an important decision point. A skilled person is flexible in their working method, and chooses methods that are appropriate at a particular time, by comparing information about the properties of each method with the context.

3. Working memory, in which information about the working context and the results of task decisions is built up into a structure shaped by the current framework for activity.

4. A general problem-solving framework, which makes use of information in any knowledge base and is evoked when an appropriate existing method is not available. In familiar cognitive skills, this is not necessary.

5. 'Automatic' responses to the environment. These perceptual-motor skills do not use working memory in thinking about the task.

Classic human factors/ ergonomic methods are concerned with optimising perceptual-motor skills by making the choice of response 'obvious' (and by reducing physical workload). Classic training methods are appropriate for structuring the extended practice needed to develop these skills.

Cognitive ergonomics and training are concerned with minimising the amount of 'problem-solving' that a new user needs to do before they can develop a standard way of doing a task.

Most of the chapters in this book are concerned with the flexible problem-solving behaviour required from the users of much IT equipment. Training for these tasks needs to develop the knowledge bases of relevant information, by giving practice in using them in typical task situations. The many training methods for this, which have been mentioned throughout this book, are summarised in another chapter, Bainbridge 1989.


Many of the ideas presented in this chapter were developed during a two-month visit to the Laboratoire de Psychologie du Travail de l'Ecole Pratique des Hautes Etudes, Paris, supported by the French Ministère de l'Education Nationale. The author would like to thank Professor Jacques Leplat and his colleagues for their hospitality and for many interesting seminars and discussions.

The author would also like to thank Antonio Ruiz Quintanilla and Andrew Shepherd for their constructive comments on an earlier manuscript.



Bainbridge, L. (1974a) Analysis of verbal protocols from a process control task. In E. Edwards and F. P. Lees (eds.), The Human Operator in Process Control. London: Taylor and Francis, pp. 1-58.

Bainbridge, L. (1974b) Problems in the assessment of mental load. Le Travail Humain, 37, 279-302.

Bainbridge, L. (1975) The representation of working storage, and its use in the organisation of behaviour. In W. T. Singleton and P. Spurgeon (eds.), Measurement of Human Resources. London: Taylor and Francis, pp.165-83.

Bainbridge, L. (1978) Forgotten alternatives in skill and workload. Ergonomics, 21, 169-85.

Bainbridge, L. (1981) Mathematical equations or processing routines? In J. Rasmussen and W. B. Rouse (eds.), Human Detection and Diagnosis of System Failures. New York: Plenum Press, pp.259-286.

Bainbridge, L. (1986) Asking questions and accessing knowledge. Future Computing Systems, 1, 143-9.

Bainbridge, L. (1988) Types of representation. In L. P. Goodstein, H. B. Anderson and S. E. Olsen (eds.), Mental Models, Tasks and Errors. London: Taylor and Francis, pp. 7O-91.

Bainbridge, L. (1989) Cognitive processes and training methods : a summary. In L. Bainbridge and A. Ruiz Quintanilla (eds.), Developing Skills with Information Technology. Wiley, pp.177-192.

Barnard, P. (1987) Cognitive resources and the learning of human-computer dialogues. In J. M. Carroll (ed.), Interfacing Thought: Cognitive Aspects of Human-Computer Interaction. Cambridge, Mass.: MIT Press.

Beishon, R. J. (1969) An analysis and simulation of an operator's behaviour in controlling continuous baking ovens. Reprinted in E. Edwards and F. P. Lees (eds.), The Human Operator in Process Control. London: Taylor and Francis, pp.79-90.

Bryan, W. L., and Harter, N. (1899) Studies on the telegraphic language. The acquisition of a hierarchy of habits. Psychological Review, 6, 345-75.

Craik, K. J. W. (1948) The Nature of Explanation. Cambridge: Cambridge University Press, 123 pp.

Diaper, D., and Johnson, P. (1989) Task analysis for knowledge descriptions: theory and application in training. In J. Long and A. Whitfield (eds.), Cognitive Ergonomics and Human-Computer interaction. Cambridge: Cambridge University Press.

Hayes-Roth, B., and Hayes-Roth, F. (1979) A cognitive model of planning. Cognitive Science, 3, 275-310.

Hoc, J-M. (1987) Psychologie Cognitive de la Planification. Grenoble: Presses Universitaires de Grenoble, 197 pp.

Hockey, G, M. (1984) Varieties of attentional state: the effects of environment. In R. Parasuraman and D. R. Davies (eds.), Varieties of Attention. London: Academic Press, pp. 449-83.

Hockey, G. M. (1986) A state control theory of adaptation and individual differences in stress management. In G. M. Hockey, A. W. K. Gaillard and M. Coles (eds.), Energetics and Human Information Processing. Dordrecht: Martinus Nijhof, pp.284-98.

Johnson-Laird, P. N. (1983) Mental Models. Cambridge: Cambridge University Press, 513 pp.

Miller, G. A., Galanter, E., and Pribram, K. H. (1960) Plans and the Structure of Behaviour. New York: Holt, Rinehart and Winston, 226 pp.

Norman, D. A. (1986) Cognitive engineering. In D. A. Norman and S. W. Draper (eds.), User Centred System Design. Hillsdale, N.J.: Erlbaum, pp.31-61.

Norros, L., and Sammatti, P. (1986) Nuclear power plant operators errors during simulator training. Research Report 446. Technical Research Centre of Finland, Espoo, Finland.

Page, S. J., Heyden, W., and Liere, B. (1983) Verhalten des Kernkraftverk-Warten personals. Phase I: Pilotuntersuchung. Institut fur Unfallforschung, Technischer Uberwachungs-Verein, Koln.

Pew, R. W., Miller, D.C., and Feeher, C. E. (1981) Evaluation of proposed control room improvements through analysis of critical operator decisions. Research Project 891, NP-1982, Electric Power Research Institute, Palo Alto, Calif.

Rasmussen, J., and Jensen, Aa. (1974) Mental procedures in real life tasks: a case study of electronic trouble shooting. Ergonomics, 17, 293-307.

Rouse, W. B. (1983) Models of human problem solving: detection, diagnosis, and compensation for system failure. Automatica, 19, 613-25.

Rumelhart, D. E. (1977) Towards an interactive model of reading. In S. Dornic (ed.), Attention and Performance Vol. VI. Hillsdale, N.J.: Lawrence Erlbaum, pp.573-603.

Rumelhart, D. E., and Norman, D. A. (1981) Analogical processes in learning. In J. R. Anderson (ed.), Cognitive Skills and Their Acquisition. Hillsdale, N.J.: Lawrence Erlbaum, pp.335-59.

Samurçay, R., and Rogalski, J. (1988) Analysis of operator's cognitive activities in learning and using a method for decision making in public safety. In J. Patrick and K. D. Duncan (eds.), Training, Human Decision Making and Control. Amsterdam: North-Holland, pp.133 52.

Shackel, B., and Klein, L. (1976) ESSO London airport refuelling control centre redesign - an ergonomics case study. Applied Ergonomics, 7, 37-45.

Shepherd, A., Marshall, E. C., Turner, A., and Duncan, K. D. (1977) Control panel diagnosis: a comparison of three training methods. Ergonomics, 20, 347-61.

Sperandio, J. C. (1972) Charge de travail et regulation des processus opératoires. Le Travail Humain, 35, 85-98.

Access to other papers via the Home page

©2021 Lisanne Bainbridge



Popular posts from this blog

Ironies of Automation

Types of skill, and Rasmussen's SRK schema

Complex Processes Review : References