Forgotten Alternatives in Skill and Workload

A beginner at a task may not know what is the best action to make, nor what size of action has what effect. Their actions are not always good and they have to do much revision and error correction.  So they may have more mental and physical workload than an expert doing the same task.

This paper discusses the changes in knowledge, behaviour, and behavioural choice that take place with learning, and the effect of these changes on mental workload.

Note : I use the word ’skill’ to indicate the change from beginner to expert, not as a label for one type of cognitive processing.

Topics covered :

1. Introduction

2. A simple control task

2.1. Using process knowledge : size and timing of actions; whether to check result of action; whether to make an action.

2.2. Qualitatively different methods and an operator's knowledge of their own behaviour.

2.3. Subsidiary decisions.

2.4. Background decision criteria.

2.5. Learning : learning about process behaviour; criteria and learning; learning adequate methods. 

3. Workload

3.1. Summary of relations between workload and experience.

3.2. High Task Demands.

References for this paper.

Forgotten Alternatives in Skill and Workload

Lisanne Bainbridge

Department of Psychology, University of Reading


Ergonomics, 1978, 21, 169-185. 

Simultaneously published as : 

Possibilités oubliées en matière d'habilété et de charge de travail. Le Travail Humain, 1977, 40, 203-224.

1. Introduction

Even in the most simple situation it is usually possible to do a task in different ways, so there must be some mental process, not necessarily conscious, which determines the how, when and if of a person's behaviour. When we study such 'decisions' between alternatives we find that they require several types of knowledge. Knowledge of the present and potential behaviours of the external world, and of their importance and cost, is basic to choosing appropriate actions. To choose a strategy which matches not only the present external task requirements but also their own mental and physical state, a person needs similar knowledge about their own mental and physical actions.

In this paper we will analyse this knowledge and decision-making explicitly, but will also discuss how it is not explicit but implicit in the judgements and 'feel' which are more frequently used. We will suggest that with increased experience the number and complexity of task 'decisions' is reduced so the amount of mental work needed to achieve a given task performance is reduced, and that the development of skill lies in this change in knowledge and decisions. We will also suggest that these decisions about what to do, which generate the sequence of behaviour, are one locus of the breakdown of behaviour under high task demands. Under normal circumstances a person tries to balance task demands and mental demands on their behaviour. Increased task demands lead to changes in behaviour priorities which the person is not necessarily experienced in reacting to.

2. A Simple Control Task

We will illustrate these generalisations with examples taken mainly from process control, from the task of controlling a single dynamic part of the external world: for example, the temperature of an oven or a water heating boiler which has no thermostat. The oven or boiler is the process. (The word 'process' will also be used in this paper to refer to mental information processing, I hope that each meaning will be clear from the context.) The examples will therefore illustrate decision making and skill in this simple control type of task only. To analyse fully the decision-making underlying information processing tasks, complex multi-variable tasks, or the overall planning of behaviour sequences, would introduce a large domain of new problems: but I hope that the general principles and inferences would not differ from those presented here.

Consider the task of changing the temperature of an oven or boiler from one steady value to another. Many studies have shown (for examples see Edwards and Lees, 1974) that when an inexperienced process operator tries to make such a change the temperature, the controlled variable, oscillates around the target value and the operator makes many changes in the power supply to try to get the temperature under control. When the task is done by an experienced operator the controlled variable moves smoothly to the target without oscillations, and the operator makes only a few actions. Most of these actions are made before there is much change in the controlled variable, so they must be made in anticipation rather than by feedback from the results of earlier actions. Correct size and timing of anticipatory actions is critical if they are to be successful, so they require an accurate knowledge of the relations between control actions and process behaviour. If an operator has this knowledge then minimum effort is needed on their part to achieve smooth efficient behaviour of the process, and this is what is commonly accepted as skill [the change from beginner to expert].

In controlling, the operator has to 'choose' whether to make an action, the size and timing of the action, whether to check the result of the action, and how qualitatively to make this check. The first three 'decisions' are interlinked by referring to common background knowledge of the process behaviour.

In discussing these choices we will make an explicit analysis of the necessary processes. The use of words like 'choice' and 'decision' is not meant to imply however that an operator necessarily deals with the alternatives in a conscious and explicit manner. Frequently they use judgements which may not be available to introspection or verbal report, and in which the nature and use of task knowledge are implicit. Control tasks can be done in a wide range of ways, from an operator who works by 'feel' and is unable to give an account of what they do independently of doing it (most of us drive a car in this way) to the commissioning engineer who needs to know explicitly the reasons underlying the process behaviour.

2.1. Using Process Knowledge

2.1.1. Size and timing of actions

Someone beginning to learn to control a process might be faced with a display showing the value of the variable they are to control, and a control knob, movements of which have an effect on the value of the controlled variable, perhaps after a time lag. The task is to change the controlled variable from one value to another, but there is no way in which (without outside help) a beginner can select the correct control knob setting from the range of alternatives. Even 'turning the knob clockwise will increase the controlled variable value' is an untested hypothesis. They must pick a knob setting to try; they then have high uncertainty about the results of this action, and will want to check it immediately.

Feedback showing that behaviour was unsuccessful is used in two main ways. It indicates the need for a further action to reach the target value. It also indicates that this previous action was not the best one to use in the circumstances. To improve their control an operator may need to change their assumption about the general rule connecting the directions of movement of display and control. They then have to adjust the size of their control changes until they uses the correct alignment between the present behaviour of the process and the action which will bring this to the required value. They may learn to control by' feel', so that in a given situation they 'know' what actions will give the desired result but their knowledge of the process behaviour is not independent of their own behaviour, only implicit in their responses to the process. Alternatively, their knowledge of process behaviour may develop into an explicit and distinct form, so that they can think about process behaviour independently of making actions in relation to it.  They can then predict process behaviour and either take preventive action now or prepare actions in advance (see fuller discussion of learning in Section 2.5).

As an experienced operator knows the correct alignment of process state and appropriate action, the need for choice between alternative actions no longer exists for them. They also have less uncertainty about the effect of their actions, so have less need to check on the process behaviour. As their first actions are correct they have less need for further actions to reach the target. As their actions are successful they also have no need to learn, i.e. to adjust their own behaviour. If they can predict process behaviour they can anticipate the need for action and so can select their action, not at the time when it is required, but beforehand in a period of low workload. When the need for this action arises they can then react immediately, without going through the decision processes and with less uncertainty. They consequently use much less mental and physical activity than an inexperienced operator, so have a lower workload to reach a given performance goal.  And their process knowledge is primary in this reduction of workload.

In looking at skilled behaviour one can easily forget how many alternatives exist for the inexperienced person, and how many decisions no longer need to be made by the experienced. The behaviour of a trainee is by no means a subset of the behaviour of the experienced but is different in kind, more complex in some ways and less complex in others. Feedback is necessary to become skilled, but actually to be skilled is to reach a stage at which feedback is minimally necessary. The skilled worker has sufficient knowledge of the properties of the environment and their own behaviour to work without checking, to a considerable extent by 'open loop'.

2.1.2. Whether to check result of action

An operator can obtain their knowledge of the current state of the process either by checking (sampling) the process, or by inferring from the information obtained in their last sample. Whenever an operator is not looking at the process they are in a betting situation about their knowledge of the process state. They must tradeoff the need for a sample against the effort required to take this sample, and decide (not necessarily consciously) which is more important. These utilities and costs may change as skill develops. In addition, more general weighting factors are involved, such as the operator's general attitude to care in their work (see fuller discussion in Section 2.4).

The need for a check will depend on their uncertainty about the process state, which will depend on their knowledge of process behaviour (assume for this discussion that taking a sample depends on uncertainty, although this is an oversimplification). A trainee has no knowledge about the effects of their action, so has a high need to check what is happening. With increased experience, their uncertainty, and therefore the need to sample, decreases (Crossman et al. 1964). The importance of this uncertainty interacts with the tolerances on acceptable process output : for example, if the operator knows only that the value is within a given range, but this range is within tolerance, then this is acceptable.

2.1.3. Whether to make an action

Having made an action and checked its effect, the operator must decide whether another corrective action is necessary. This is also a betting situation, as the operator must tradeoff the need for action from the control task point of view, on the one hand, against the effort they must make to choose and implement the action on the other.

In formal terms, the need for action can be assessed by measuring the difference between present state and target, and evaluating whether this difference is acceptable or unacceptable (assuming the simplest case where control actions are made in response to control error alone). By a simple definition, the process state is acceptable if it is within the range: target ± tolerance (threshold). The evidence suggests however that an experienced operator uses finer categories of evaluation, as they behave differently when the process output is near the boundaries of the target region than when it is central. Crossman et al. (1964, Table 2) have shown that they check the process state less often when it is central.

The control-state assessment is also usually made not by calculation, but by a different mechanism. In absolute judgement an existing value is compared with the subjective expected average and range of values, to give broad categories of assessment, e.g. very low, low, average, etc. In such judgements, measurement and evaluation are not distinct processes, as the measures are made directly in relative terms. Evidence suggests a similar method is used in process control, only here judgements are relative to the control target and the categories relate to action, e.g. decrease required, above, on target, below, increase required (Bainbridge 1971, Table 4). The control state may therefore not be evaluated independently of its implications for action. The operator learns the divisions between these overlapping categories by interaction with the process. They may, for example, categorise a process state as needing no action. When future process behaviour shows that an action should have been made, they must revise their categorisation and so they learn the alignment of process state with action category. This could be done by judging dial pointer angles, for example, without ever knowing the actual variable values corresponding to the divisions between categories. The criteria for evaluation, which are an aspect of their knowledge of the process behaviour, are therefore not necessarily explicit and available to conscious report, but are implicit in the process of judgement. As absolute judgement can only be made with any validity after experience, this is another sub-task which is more difficult for an inexperienced operator.

2.1.4. General comments

As we have seen, choosing an action, checking the action, and deciding to make a further action, are all interrelated by the use of process knowledge. We can suggest that, if control is done by 'feel', these three types of decision are made in a combined way, in which knowledge and the criteria of 'choice' are implicit in the judgement categories. The control-state evaluation implies both the need for action and the size of the action. For an experienced operator the size of action used corresponds to the control state with minimal uncertainty about its effect, while an inexperienced operator must realign their evaluations on the basis of new feedback information, and this learning must be continued until uncertainty is minimal.

In a formal analysis of the control decisions, knowledge of the process behaviour and knowledge of evaluation criteria for decision making are distinct and independent. When an operator controls by 'feel', however, the evaluation criteria by which the process state is assigned to a particular response category are based on process knowledge.

This discussion has centred on a simple control task, in which the target is a stationary value and the operator learns the behaviour of the process being controlled. In order to follow a varying target (tracking) an operator must also learn about the regular or probabilistic behaviour of the environment, and use this together with their knowledge of process dynamics (see for example Poulton (1957), Young (1969), Pew (1974 a)). We know very little about the integration and coordination involved in controlling a complex process with many interacting variables. The operator must learn about these interactions, including the redundancies of information, and how to sample efficiently these multi-variable interrelated changes in time.

2.2. Qualitatively Different Methods and the Operator's Knowledge of their Own Behaviour

Most of the alternative behaviours discussed so far have differed quantitatively, for example different sizes of timing of actions. In Section 2.1.3 we saw that the operator could assess the control state either by calculation or by judgement; we therefore also need to discuss the process of choice between qualitatively different methods of doing the same task. Again, this type of 'decision' involves knowledge and costs. The operator needs to 'know' the alternative methods, the dimensions of choice between them, their values on these dimensions, and the state of the working context which determines what is the most appropriate behaviour. The inexperienced operator does not necessarily have these types of knowledge, but must acquire them by doing the task.

The Table below shows an example comparing calculation and absolute judgement as alternative working methods.

Calculation uses accurate numerical values and arithmetic. Absolute judgement is different in nature and the results are quoted in imprecise relative categories - these categories represent a subjective scale of expected values which can only be developed by experience with a particular task, by feedback about the validity of previous judgements. To choose which behaviour which is most appropriate in a given context it is necessary to assess the potential properties of the behaviours [first 2 columns below] against what is required by the context [3rd column]. 

Some example dimensions on which this matching might be done are given in Table I. On one side [first 2 columns] are the utilities and costs of the alternative behaviours, in terms of their potential contribution to task demands, such as time taken and accuracy achieved, and their potential costs to the operator in terms of mental effort. These must be compared with the actual demands and facilities in the environment at the particular moment, i.e. the current task demands of time and accuracy, and the currently available mental processes and capacities [3rd column]. 

This example compares two strategies at the level of information processing. Alternative behaviours with manual components would also require information about the operator's potential and currently available physical capacities. Again it is not intended to imply that these comparisons are explicit, or that this knowledge is available to conscious access. Introspection suggests that sometimes what to do "comes straight into one's head", while on other occasions several alternatives flash through one's mind; it is not clear what determines which occurs. This is the sort of experience about which it is very difficult to obtain adequate reports. If one is not conscious of the alternatives this does not necessarily imply that alternatives are not considered. An operator could 'choose' their behaviour by some multi-variable judgement process, with the 'knowledge' about alternative behaviours implicit in the evaluation. Again, introspection suggests that one matches each possible behaviour in turn against all the criteria which are integrated together in some pattern, rather than comparing all the behaviours against each criterion in turn, but all this is pure speculation. (One can also speculate that this patterning of the context is an important aspect of the integration of complex multidimensional activities.) [And different people have different cognitive styles.]

Obviously studies of human decision-making are relevant here, but they will not be discussed in this paper as the evidence available so far about this type of choice in working contexts is inadequate to add to the debate about actual mechanisms of decision making. It is relevant however to note that decisions are not necessarily made by maximising utilities, but can be made by 'satisficing', by finding some method which is sufficient rather than optimal (see example in Section 2.5.3).

For an experienced operator working in an unvarying environment, any necessity for decision-making or matching of alternative behaviours will disappear and the task may always be done in the same way. However the 'decision' about what to do will always be potentially relevant, and alternative behaviours can reappear if the environment changes. For example, an operator may usually turn a valve by hand. If feedback indicates it is too hot or too cold they may kick it or use a crowbar. The 'choice' of a solution to the problem of the changed environment may be easy and appear to be automatic. Alternatively, as both the behaviour and the process of choice are unpractised it may require thought and search, as the operator is inexperienced in this new situation. This effect of unfamiliar circumstances may be basic to the effects of increased workload (see Section 3). 

In comparison, there are some tasks in which the task environment is continuously changing, for example the varying number of aircraft to be handled by an air-traffic controller. Sperandio (1972) found that air-traffic controllers changed their behaviour according to the number of aircraft. In such a task the controller continually practises, and so is skilled at, both the ensemble of alternative behaviours and the process of changing between them.

An inexperienced operator does not know the potential abilities and costs of alternative behaviours. They can only discover these properties by experiencing the behaviours (properties such as speed, accuracy and the mental facilities required may also change with practice). An experienced operator does 'know' these parameters, so their behavioural 'choice' is made by anticipation of the results. For example, de Groot (1965) found that experienced chess players explore the same number of possible moves as inexperienced players, the difference is that all the moves explored by experienced players are potentially good ones. Choice between alternative behaviours is therefore another task decision in which the inexperienced person is acting by feedback, while the experienced person is working open loop. Flavell (1976) makes a related point in a different context when he discusses children learning what he calls 'meta-communication', or knowledge about the general characteristics of listeners, which is used in ongoing monitoring of a conversation to ensure one is communicating the message, and 'meta-memory', the difficulty of different memorising strategies for storage and retrieval.

It is useful to know the potential contribution to the task of some behaviour, e.g. the speed and accuracy obtainable, only if one can also assess, or knows from experience, the dimensions of the present 'task demands' context, e.g. time and accuracy required, so that the two can be compared. Similarly it is also necessary to assess how much working storage capacity and which mental operations are currently available in the person’s head, i.e. the present mental context, for comparison with data on the potential mental effort of carrying out the behaviour. These properties of the external and internal environment in which the operator is working may vary from moment to moment; for example, working storage usage varies with details of the task, or the types of mental operation available may vary with fatigue. Again, an inexperienced operator must learn how to assess these tasks, the physical and mental values which together form the context for behaviour choice. Skill consists of using the most efficient and minimal behaviour, both inside the head and outside. Adaptability, which is a more important aspect of complex skills than speed and accuracy, may come from 'skill' in making these higher level decisions, in reacting to variations in the ongoing task context.

In Section 2.1 we discussed an operator's knowledge of process behaviour, which they use in choosing control actions. The analysis in this section implies that an operator also needs knowledge about their own behaviours, mental and physical, if they are to choose behaviour which matches the task demands. We have also assumed that the operator's choice of behaviour is made, not only to control the state of the external world, but also to maintain their own physical and mental state (see Section 3.2). This assumption implies that they need knowledge about further dimensions of their own behaviours, and also an awareness of their own present state. It also implies that, in addition to criteria for acceptable task performance, they have criteria for their own acceptable mental and physical state (see Section 2.4). An operator's knowledge of the potential process behaviour is basic to all their decisions about control of the process, and the present state of the external world determines the most appropriate response to it, but the operator's knowledge of their own potential behaviour and their own present state are basic to the implementation of all their control decisions.

2.3. Subsidiary Decisions

We have so far discussed four types of decision involved in a simple control task. All these decisions, and the related physical and mental behaviour, involve subsidiary decisions and behaviours.

For example, to calculate an error in control state involves three operations, finding the present value, finding the target value, and calculating the difference. Each of these subsidiary operations could itself be carried out by a number of alternative methods. The operator could find the target value from a display, from memory, by asking another operator, by looking up the task instructions, or in some tasks by calculation from other variables. Which of these alternatives they use will depend on comparing the present task and mental contexts with their knowledge about the alternative behaviours. Again, the inexperienced operator will have to 'discover' all these sub-factors, and doing so may greatly alter the way in which the task is done.

To implement a given control change requires subsidiary decisions about how much to move the control knob to achieve this effect, and by what physical movement to achieve this movement of the knob. Each of these is a subsidiary control task which itself requires decisions about size and timing : whether to check, by what method to make the check, and whether further action is required. Of course, in many control tasks the operator's behaviour is not this analytic; one can suggest that when an operator controls by 'feel' they control by direct mapping from the process behaviour to the physical movement of turning the knob, without intermediate levels of awareness.

In other skilled tasks, subsidiary goals may have to be dealt with, concerned with the details of the ongoing environment but having nothing to do with the main task. For example, while one is driving to work, one has to avoid traffic at a roundabout. The overall plan for reaching the main goal, in this case the route to use, can usually be decided beforehand, but the behaviour in relation to traffic situations cannot. The result is that the actual behaviour is a function of the overall goal interacting with details of the local conditions in which one is trying to achieve this goal.

We can suggest that all the points made in previous sections about knowledge, choice and skill also apply at all these subsidiary levels of behaviour organisation, so that experienced knowledge of the potential behaviour of oneself and the dynamic environment can greatly increase the efficiency, or reduce the effort, of carrying out these subsidiary behaviours. For example, motor skill involves learning to make movements without visual checking, under kinaesthetic monitoring; these movements can then be made simultaneously with other tasks or with levels of this task which do require visual check. It is about 60 years since Lashley suggested that movement control involves a predictive model of muscle behaviour used open-loop (see Pew 1974a) so that an operator has foreknowledge of the results of their own physical behaviour. This can include knowledge of the sensory consequences. The general point can be made that pre-knowledge of the results of behaviour, whether it is in a nuclear power station or a big toe, leads to more efficient behaviour of the thing being controlled, and minimum effort in actions and checks to achieve this behaviour. This can have the further powerful effect that close integration between levels of behavioural organisation becomes possible. 'Higher' levels may not need to check the functioning of lower levels, but can proceed on the assumption that certain actions will be carried out and certain information will be available at certain times.

2.4. Background Decision Criteria

We have suggested that the potential utility and costs of a given behaviour, for both the process and the operator, are compared with the actual requirements and available effort at the time of making the decision. These are short-term criteria.  There are also long-term criteria : the overall task performance goals set by the management, and the general attitudes of an operator to their work. We need give only some brief examples of the types of factor affecting work attitudes which have been widely studied in other contexts (cf. Warr and Wall 1975, March and Simon 1958, Ch. 3).

The extent to which an operator bothers with their work is not simply determined by the physical and mental costs of the effort involved in doing the task, or the threat of being sacked, but also with their evaluation of the work as satisfying their needs. An individual's preference for particular types of behaviour, for risk, or for doing a job well, may affect the methods of working as well as the parameters of performance. (It is this, for example, which makes one suspicious about generalising from the results of laboratory experiments using university student subjects. Students are supposed to think, and it is an interesting and appropriate thing to do in a laboratory, but this is not necessarily representative of the behaviour of people with different motivations working in a situation which continues for many years.)

Work is also one dimension of one's membership of a social group. Not only the type of work but also the way one does one's work can affect group membership. This can easily lead to behaviour which is counterproductive or even apparently irrational. In a study of controlling the power supply to a simulation of five electric-arc steel melting furnaces, we found one operator who continued to consider more and more refined dimensions of action choice, without ever making an action, although the control state was becoming worse and worse. We concluded that it was very important to this person both to be absolutely right before committing himself to an action, and also to show the observers that he had a complete knowledge of the determinants of the process's behaviour. This leads to a weighting of behaviour criteria which seems irrational when considered in relation to the control task. This happened with a simulated task, perhaps it would not happen on the shop-floor. 

However, Cuny (1976, personal communication) has given an example of difficulties which can arise on the shop-floor when several operators are controlling different parts of one process. The operators he studied had more knowledge about their work than they used in doing the job. Their job had been defined as covering certain tasks on the process. The overall process output would have been better if they had gone outside this framework and made other actions which they had relevant information about. However these actions had been defined as the responsibility of other operators.

Long-term criteria about the amount and type of effort the operator is willing to put into the task are not permanent, but themselves develop in interaction with the task. The match between what an operator wants from their work and what they actually attain affects both job satisfaction and level of aspiration (see further discussions in Section 2.5.2.). A large mismatch between what is obtained and what is wanted in any task will lead to lack of commitment to meeting the task demand, which appears in the classic job dissatisfaction symptoms of high levels of absenteeism, high labour turnover, poor work, and accidents. These background decision criteria therefore form a major context of the task decisions. One would expect that the process state would provide the context of control task decisions, and in Section 2.2 we discussed the place of the operator’s present mental state in decision-making. We can also see that the overall costs and gains of the work for the operator could be an all pervasive determinant of behaviour, which can enter as a relevant criterion affecting behavioural choice at all levels of task organisation.

2.5. Learning

We have discussed how an operator's knowledge of the process and of their own behaviour changes with experience, and how this can have major effects on the way the task is done and on the workload involved. We will now look directly at this process of knowledge acquisition, and how it interacts with the task performance achieved. Only a few aspects of learning, which are relevant to the present discussion, will be covered. Learning to control a multivariable process with time lags and random disturbances is very complex, and real process control skills can continue to develop over many years.

2.5.1. Learning about process behaviour

As we have already discussed, if the process does not behave correctly after some action by the operator, this indicates not only that the process output needs further correction, but also that the operator's methods of judging the process state and choosing the action need correcting. Thus the process behaviour is used as information in two nested feedbacks loops, both in controlling the process [lower loop in diagram below] and as a basis for refining the operator's knowledge and decisions about controlling the process [upper loop].

Any learning requires some sort of memory for the previous action or judgement and the context in which it was made. At a minimum the operator should remember the action or judgement, note whether the result was wrong, and if so use a different action next time, so that there is some exploration, however unplanned, of the range of possibilities. At best an operator would have some notion of the expected result of their chosen action. They would compare the actual result with the expected result, and make use of the difference in refining their action choice and expectations, see above figure and Young (1969, Fig. 20). Cooke's (1965) student operators gave verbal reports while they were learning to control a simple process. These reports showed that they did suggest hypotheses about the possible future behaviour of the process and how it might be controlled. Young op. cit. quotes a large number of possible models for the mechanism by which an operator adapts to particular process dynamics.

To learn, as well as to control the process, the operator must notice errors in process performance; but noticing errors can itself be a skill which requires learning (Leplat and Pailhous 1974). In some cases the discriminations can involve perceptual skills which must be learned; judging the colour and surface texture of molten steel for example. In all such cases an operator can only come to make finer discriminations if they are also given equally finely distinguishable feedback about the adequacy of their performance.

All these comments imply that the inexperienced operator must perform many types of mental task and decision, in order to learn the task, which are at a different level to the process-related task decision themselves and are no longer necessary once the task has been learned. Once control performance has been developed to the point where an action makes process output reach the target within acceptable limits, then there is no further need to learn.

In fact it seems that after they have reached this level of performance an operator uses their expectations as a reliable representation of the state of the external world. If a mismatch between their expectations and the actual behaviour occurs, then they assume there is something wrong with the machine, rather than with their expectations. If the instrumentation has a history of being unreliable, then they may assume that the instrument is faulty rather than the process, and this can be a major reason why operators appear to ignore unusual process states which eventually lead to breakdown.

2.5.2. Criteria and learning

In talking about controlling a process, we have said that the extent to which an operator bothers to correct the process behaviour depends on the extent to which they care about the adequacy of its performance. Learning is affected by similar criteria. The operator will only learn, that is make the effort to correct their own behaviour, if they care about the difference between their present level of performance and what could be achieved. (Learning has an added advantage, as better task performance requires less effort.) Two nested types of evaluation criteria are therefore involved in learning. As an example, in assessing the process state, the operator has to learn the appropriate types of category for evaluation, e.g. action needed now, or action may be needed in future. They must also learn how to align these categories accurately against the control state. We are concerned here with how accurately an operator bothers to make their evaluations and action choices.

We have previously discussed knowledge of the task and knowledge of the criteria for decisions as if they are independent, although when an operator does their task by 'feel' they are not distinct (see Section 2.1.4). We can now see that the relationship between the two is complex. The care taken in discriminating levels of process error affects learning, which affects process knowledge, while in turn this process knowledge determines the performance achieved which, interacting with the performance which the operator should achieve, affects level of aspiration, which affects the care taken. Care and knowledge are therefore linked in a circle, with the result that, if the relation between aspirations and achievement is good, performance will improve: while if the relation is poor, performance may deteriorate. (See Lewin et al. (1944), March and Simon (1958, Chap. 3), and Welford (1976, pp.126-131) for further discussions of the feedback relation between level of aspiration and performance.) This has important implications for training. If the task is initially too difficult, an operator may come to think that the performance goals given to them are impossible and will work to their own goals, accepting a lower level of achievement.

2.5.3. Learning adequate methods

The aspects of learning discussed so far have been concerned mainly with parameter adjustments, making quantitative changes in behaviour. The operator also has to learn the general methods for doing the task. 

Vermersch (1976) has analysed the changes in behaviour of trainee technicians learning to adjust an oscilloscope. He suggests that an inexperienced worker starts by using some method which is suggested to them by the environment in combination with what task knowledge they have. If this gives inadequate results they must try other methods, which occur to them by trial and error, or from further knowledge they have, or are suggested by other people. Vermersch emphasises that a trainee is not necessarily actively searching for a good way of doing the task, but simply that the necessity to succeed at the task forces them to try other things until they find some sufficient behaviour, and that they are not necessarily conscious of the changes in their behaviour or of their nature. This finding would presumably depend upon the particular group of people studied, as there are some people for whom understanding and elegance of behaviour are possible aims.

However unwillingly, the trainee is in a problem-solving situation and must search, perhaps unconsciously, for methods which will achieve at least a minimal level of success. Some mechanism for selecting alternatives and evaluating results must also underlie this problem solving activity. It will be more complex than in the simple case of adjusting the parameters of control behaviour. Again, the trainee has to try several behaviours which are concerned with developing their skill, in addition to the behaviour required directly for doing the task.

It is evident that even a simple task requires a large amount of knowledge of different types, about alignments, criteria and methods, if the task 'decisions' are to be efficient. 

Beishon (1969) has listed the types of knowledge or "body of associated facts" needed in a complex task, the knowledge needed by an operator controlling a continuous baking oven : baking temperature profiles along the oven for each type of cake baked, and their tolerances, burner pattern profiles which will achieve these temperature profiles, methods of getting from one profile to another, methods of correcting faults in the product, and expectancies about cake types and the times when they may arrive for baking.

In many shop-floor situations, operators are left to discover these types of information by themselves, so it is hardly surprising if even an experienced operator's behaviour is not optimal. They may not have the necessary background information to bootstrap themselves up into having an adequate knowledge of the plant without help, as it has to occur to them to look for certain things. Nor is it surprising that people can be upset and worried when doing a complex task for the first time; when one analyses what is involved it is more surprising that they manage to do it at all.

3. Workload

We have already presented considerable support for the notion that an inexperienced operator has to do very much more mental work than an experienced one to respond to the same level of task demands. We will first summarise this evidence, and then discuss how behavioural decisions might underlay the breakdown in behaviour under high task demands. Note that we are using the term 'workload' to refer to mental work, and the word 'demands' to refer to the task requirements, and are not using a separate term for the levels of task demand which lead to performance breakdown.

3.1. Summary of Relations between Workload and Experience
We have suggested that increasing experience may have several effects on the number and type of decisions which must be made to choose an appropriate behaviour. An experienced operator may use a particular behaviour automatically, by habit, because the alternatives have been assessed against decision criteria in the past, and circumstances have not changed so a decision would be redundant. They have learned the properties of the process behaviour and of their own behaviour in relation to it, so they can 'choose' between alternative behaviours without exploration, by fore-knowledge of results. They do not have to carry out the behaviours to discover whether they are appropriate, so they do not spend time and effort using methods which are inadequate. They control the process with low error and low uncertainty, they also could have better control over the amount of effort they expend themself. As all subsidiary behaviours also improve in the same way, all the properties of their behaviours become more efficient.

For an inexperienced operator all behaviours are more complex and uncertain, and therefore slower. A large proportion of their behaviour is concerned with how to do the task, with decisions about their own behaviour such as learning and exploration, rather than dealing directly with the task demands. An experienced operator does not have to make decisions at these learning levels, so they not only have less work, but a higher proportion of their work is directly task related. When their task performance is measured they will therefore have a larger 'task capacity', but this may occur only because more of the mental work which they do in a given time is directly task effective.

The notion that less decision making and attention are required after practice at a task is a longstanding one, e.g. James (1890, Chap. 4). It has also been tested experimentally several times, using the argument that as less attention is required so there will be less disruption by or of a 'secondary task'. Bahrick and Shelley (1958), Adams and Chambers (1962) and Dimond (1966) have all shown that either primary or secondary task performance was better when stimulus sequences were more regular, supporting the notion that more predictable tasks are easier. In studies by Bahrick et al. (1954) and Trumbo et al. (1967) this effect appeared only after practice. The latter two experiments involved larger numbers of alternatives, so one can suggest that extended practice is needed only if it is more difficult to notice the task regularities. 

An experiment which does not fit into this pattern is described by Pew (1974 b). This was a continuous tracking task, while the above experiments all used key-presses or 'tracking' between discrete levels. Part of the track to be followed was repeated in each trial. While performance on the repeated track was better than on the changing track, performance on a secondary task was not better. One might extend the argument that more practice is needed.

Spelke et al. (1976) have shown, in a different task, that extensive practice leads to a release from 'capacity' limitations. 

One could also make the post hoc suggestion that practice may show improved performance while learning is consolidated at one level of task organisation, but more extended practice is required to change the type of task organisation. [Bryan & Harter (1889) found long ago that learning Morse code consisted of a sequence of alternating phases : learning phases during which performance improved, and plateaux when there was no observable change in behaviour, when they inferred a different type of learning was happening.]  For example, in Section 2.5.1 it was suggested that a learner has two types of task until learning is complete, after which they have only one.

3.2. High Task Demands

In Section 2.2 we discussed how an operator may choose their behaviour by considering both the task requirements and their own abilities and willingness in combination. They compare the task requirements with what can be achieved using available behaviours, and they compare the mental and physical operations required for these behaviours with their available capacities. We assume that an operator tries to maintain some rate of their own mental and physical work as well as to control the process to meet the task criteria, and that they try to adapt their behaviour to balance these two aims (see discussion in Bainbridge 1974 b).

Obviously changes in task demands should affect behavioural choice. In fact, if someone does not respond differentially to such changes they are considered irresponsible or depressed. When task demands increase, the important question is whether the operator can use a different working method which will give increased task performance for the same amount of mental work (Sperandio 1972). If no such strategy is available then mental work must increase and/or performance must deteriorate. As an experienced worker can perform a given task with lower mental workload than a trainee, they will be less immediately susceptible to the effects of increased task demands. If increased task demands can be dealt with only by a change in behaviour, then the mechanism of behavioural choice is central in the operator's reaction to these demands. We will discuss some ways in which high task demands may have a disruptive effect on the normally smooth operation of decision making processes, and so lead to performance breakdown.

Most points made in the literature about the effects of high task demands on performance are intuitively reasonable, rather than fully tested (but then so is most of this discussion!). The effects of 'stressors' external to the task have been studied more fully. It is usually suggested that there are changes in speed-accuracy tradeoffs, omissions, and incorrect overall uses of sequences, timing, and working storage. According to the above notion, these changes occur when it is not possible to find a working method which both matches the task demands and is possible within the available mental capacity. Leplat (1974) and Hacker (this symposium) talk of disequilibrium in behaviour when the operator has no solution to a task demands situation.

High task demands will have a high priority and will drive behavioural choices. For example, in a steel-making furnace study, when the control state was urgent, the operator immediately made an action without considering other possibilities first (Bainbridge 1974 a, Figure 3). He therefore returned to a feedback strategy, when under pressure, while he used a predictive strategy at other times. Prediction may appear a luxury when action is urgent, but actually a feedback strategy is more difficult, partly because the result of any check of the process state is unexpected and so more difficult to react to, and also because the action decisions must be made under time pressure rather than beforehand in periods of low workload.

Feedback is the strategy used by an inexperienced operator, and one can suggest that many aspects of performance breakdown under high task demands occur because an operator is in a situation in which they are returned partially to being inexperienced. For example, as task demands have high priority the operator is working in a changed cost context. They cannot use the behaviour relevant to the stable situation, but are returned to the necessity of making a 'decision' about their behaviour. Much experience of working in stable conditions may lead to faster responses, but adaptability to a changing environment is lost. Making this decision about what to do may not simply add to their workload at a time when they are under pressure, but could also cause problems because this decision-making itself requires practice if it is to be done well. (It would be interesting to know whether more or less difficulty arises in changed circumstances when a task is done by 'feel' or when it is done by relatively explicit and formal procedures.) An operator may know other working methods which are appropriate, but which they have not used recently. If they do not, they must devise another method : this leaves them in a situation of high uncertainty in which they must explore and experiment, when there is no time for this extra behaviour. A large proportion of their behaviour will not be directly task-demands oriented, and so their behaviour will show an apparent lowering of task capacity.

In well practised behaviour, changes in efficiency have occurred at all levels of behaviour organisation, with complex interrelations between levels, and lower levels being independent of 'conscious' monitoring. We can speculate that if the operator is returned to devising new methods and an active choice of behaviour, this can disrupt the parts of the behaviour which are organised autonomously. These 'lower' parts of the behaviour will be integrated, in that they expect certain cues at certain times. If these cues do not arrive at these times, because of different timing or patterning of 'higher level' behaviour, this can have a rapidly destructive effect on performance (try giving your legs explicit instructions about what to do while walking upstairs). This will add to an operator's uncertainty about the properties and adequacy of their own behaviour.

An operator might choose several methods of simplifying the problem when they are not able to do all the necessary tasks. They might attempt to continue as normal, find this impossible and so omit whole sections of the task at random, or they might concentrate on high priority aspects and forget others, as in 'tunnel vision'. Given a fuller understanding of the behavioural choice mechanisms, one might be able to predict what types of breakdown reaction could occur; assuming that each individual gives their main weighting to some particular type of parameter in the behaviour decisions.

One simple practical implication of this analysis, if behaviour breaks down under high task demands to a version of inexperienced behaviour, is that an operator should have practice of working at different task demand levels, so that they become skilled with different methods and with the process of adapting from one method to another. Such practice occurs naturally in the task of air-traffic control, and is given in special simulator sessions for airline pilots and nuclear reactor operators.

Though airline pilots and reactor operators are given not high task demands training but incident training. Dealing with machine malfunction is a problem related to high task demands. An operator does not know which parts of the process are behaving in a normal way, so they cannot use their normal assumptions and anticipations but must again work by exploration and feedback, with high uncertainty. Experience with emergencies can have a positive or negative effect on behaviour. The operator knows the possibilities and risks of the situation, but also their previous experience of success or failure will affect their subjective assessment of their own abilities and so will affect their judgements about whether their available behaviours match the task demands.

This discussion suggests that breakdown in performance under high demands could occur as a natural function of normal decision-making mechanisms, rather than through an actual deterioration in capacities. We have not included here any discussion of ways in which real changes in basic capacity (rather than task capacity) can occur, for instance through changes in physiological state. As their physiological state is also part of the mental context of their decision making, one could suggest that an operator will also try to choose their behaviours so that their physiological state is maintained. If under high demands the task requirements are driving the behavioural choices, and task and personal criteria for the decisions cannot both be kept in balance, then the person’s 'mental equilibrium' may not be maintained. In addition, an inability to produce behaviour which matches their own criteria of what should be done might lead to emotional states such as worry or fear. There are therefore several potential routes by which changes in behaviour under high task demands could also lead to changes in physiological state.

The author would like to thank Professor Jacques Leplat and his colleagues at the Laboratoire de Psychologie du Travail, Paris, for many interesting and valuable discussions during the initial preparation of this paper.

- - -


ADAMS, J. A., and CHAMBERS, R. W., 1962, Response to simultaneous stimulation of two sense modalities. Journal of Experimental Psychology, 63, 198-206.

BAHRICK, H.P., NOBLE, M., and FITTS, P. M., 1954, Extra-task performance as a measure of learning a primary task. Journal of Experimental Psychology, 48, 298-302.

BAHRICK, H. P., and SHELLY, C., 1958, Time sharing as an index of automatisation. Journal of Experimental Psychology, 56, 288-293.

BAINBRIDGE, L., 1971, The influence of display type on decision-making strategy. In Displays, Conference Publication No. 80 (London : INSTITUTION OF ELECTRICAL ENGINEER5, pp. 209-215).

BAINBRIDGE, L., 1974 a, Analysis of verbal protocols from a process control task. In The Human Operator in Process Control (Edited by E. Edwards and F. P. Lees) (London: TAYLOR and FRANCIS). Pp. 146-158.

BAINBRIDGE, L., 1974 b, Problems in the assessment of mental load. Le Travail Humain, 37, 279‹302.

BEISHON, R. J., 1969, An analysis and simulation of an operator's behaviour in controlling continuous baking ovens. In The Simulation of Human Behaviour (Edited by A. de Brisson) (Paris : DUNOD). Pp. 329-343. Reprinted in The Human Operator in Process Control edited by E.Edwards and F. P. Lees) (London: TAYLOR and FRANCIS). Pp. 79-90.

COOKE, J. E., 1965, Human Decisions in the control of a slow response system. D.Phil. Thesis, University of Oxford.

CROSSMAN, E. R. F. W., COOKE, J. E., and BEISHON, R. J., 1964, Visual attention and the sampling of displayed information in process control. Human Factors in Technology Research Group, University of California at Berkeley. HFT 64-11-7. Reprinted in The Human Operator in Process Control (Edited by E. Edwards and F. P. Lees) (London: TAYLOR and FRANCIS). Pp. 25-5O.

DE GROOT, A. D., 1965, Thought and Choice in Chess (The Hague: MOUTON).

DIMOND, S. J., 1966, Facilitation of performance through the use of the timing system. Journal of Experimental Psychology, 71, 181‹183.

EDWARDS, E., and LEES, F. P., 1974, (editors) The Human Operator in Process Control (London : TAYLOR and FRANCIS).

FLAVELL, J. H., 1976, The development of metacommunication. Paper given at the Symposium on Language and Cognition, 21st International Congress of psychology, Paris.

JAMES, W., 1890, The Principles of Psychology (New York: HENRY HOLT).

LEPLAT, J., 1974, Analyse du travail et genese des conduites. International Review of Applied Psychology, 25, 3-14.

LEPLAT, J., and PAILHOUS, J., 1974, Quelques remarques sur 1'origine des erreurs. Bulletin de Psychologie, xxvii, 312, 729‹736.

LEWIN, K., DEMBO, J., FESTINGER, L., and SEARS, P. S., 1944, Level of aspiration. In Personality and the Behaviour Disorders (Edited by J. McV. Hunt) (New York: RONALD). Volume 1, Pp. 333‹378.

MARCH, J. G., and SIMON, H. A., 1958, Organizations (New York: JOHN WILEY).

PEW, R. W., 1974 a, Human Perceptual-Motor Performance. In Human information Processing : Tutorials in Performance and Cognition (Edited by B. H. Kantowitz) (New York: ERLBAUM). Pp. 1-39.

PEW, R. W., 1974 b, Levels of Analysis in Motor Control. Brain Research, 71, 393-400.

POULTON, E. C., 1957, On prediction in skilled movements. Psychological Bulletin, 54, 467 478.

SPELKE, E., HIRST, H., and NEISSER, U., 1976, Skills of divided attention. Cognition, 4, 215-230.

SPERANDIO, J. C., 1972, Charge de Travail et regulation des processus operatoires. Le Travail Humain, 35, 85-98.

TRUMBO, D., and SWINK, J., 1967, Secondary task interference in the performance of tracking tasks. Journal of Experimental Psychology, 73, 232-240,

VERMERSCH, P., 1976, Une approche de la regulation de l'action chez l'adulte. Thèse du 3me Cycle, Laboratoire de psychologie du Travail, Paris.

WARR, P., and WALL, T., 1975, Work and Well Being (Harmondsworth: PENGUIN).

WELFORD, A. T., 1976, Skilled Performance: Perceptual and Motor Skills (Glenview: SCOTT, FORESMAN).

YOUNG, L. R., 1969, On adaptive manual control. Ergonomics, 12, 635-674.

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