Mental models in cognitive skill : the example of industrial process operation

This paper was originally written for a book on mental models, so starts with a section reviewing the different ways in which the term is used.  The editors also have interesting chapters in the book, reviewing and integrating the different attitudes and aims about 'mental models’ of the writers included.

As usual I use the word ’skill’ to indicate expertise, not a specific type of cognitive processing.

This paper includes a brief summary of a proposed cognitive processing mechanism. 

There is more information about how and why it was derived in the introductory notes and Section 2 of Bainbridge, 1975.

In this paper the basic processing component is called a 'module', in other papers it may be called an 'element'. There is no obvious good word for it.

As an example, this paper takes a small unit of behaviour, and attempts to illustrate how many different types of knowledge are used in doing it, and how they are all interrelated and interdependent.  Each is called a 'mental model' by different authors !

There are many small additions to the original text, and one larger one, all marked by [square brackets], which hopefully make this easier to understand.

Figures : the only versions I have on file are poor quality .gifs.  So these are photos from the book - wonky but readable.

Tables 1 and 2 were reversed by the editors in the book, they put data on the right and inferences on left.  I think the data comes first, the analysts’ inferences about underlying cognitive tasks second.  Either way, someone other than the person doing the task is imposing some structure on reported mental events.

Topics :

1. Introduction.

1.1. The domain of mental models.  The term 'mental model' is used by different writers with different meanings.  So instead I use the terms : knowledge base, working storage, meta-knowledge.

1.2. The approach taken in this paper.  Different writers have different needs of a model, which leads to much unnecessary debate.

2. An example of industrial process operation, from a nuclear power plant incident.

3. The cognitive modules in process operation.  How cognitive processing modules have been derived, what the modules do and how they make use of knowledge bases and working storage.

4. Summary of knowledge bases and working storage, types of knowledge base.

5. Goals-means structure of the task. A goal-oriented processing mechanism, meta-knowledge used in choosing between alternative methods of achieving a cognitive goal.

6. Processing and skill.  Skill as : 

- readily accessed knowledge bases, 

- working storage structured and cross-referenced relative to task goals.

7. Implications for theory and practice, some notes.


Mental models in cognitive skill : the example of industrial process operation

Lisanne Bainbridge

Department of Psychology, University College London

In Rogers, Y., Rutherford, A. and Bibby, P. (eds.) Models in the Mind. Academic Press, London, 1992, pp. 119-143.

1. Introduction

The term 'mental model' is widely but often vaguely used. For example, industrial process operators are said to have a 'mental model' of the process they control (steel works, oil refinery, etc.). They use this in understanding how the process works and in deciding how to operate it. But authors are rarely specific about what these mental models are like. In the book edited by Gentner and Stevens (1983) all the authors use a 'mental model' as an explanatory concept, but not one of them defines what they mean by the term. The aim of this paper is to be more explicit about the complex notion of a mental model, and its place in cognitive skill.

Firstly, what is a 'mental model' in general ? Early experimental psychology was almost exclusively concerned with tasks in which there was only one response to each stimulus. Cognitive psychology is concerned with understanding tasks in which a stimulus is processed in some way before a response is chosen; the brain of the person doing the task contributes something which is not in the original stimulus. For example, there is nothing explicit in 'example' and 'aelmpxe' which indicates that one is a word and the other is not, but the brain recognises this and processes the two strings of letters in different ways. In many tasks, knowledge about structures or cause-effect relations which underlay what can be observed play a central part in cognitive processes. When doing the task makes use of knowledge about the state(s) of a potentially changeable world, these structures of knowledge may be called 'mental models'.

People studying mental models have a large range of interests, and this affects what they consider should be in a model. This introduction outlines the issues that someone studying mental models might be concerned with.


What are the main dimensions of the topic of 'mental models', the issues that investigators are concerned with? I suggest there are five main themes:

- the part played by the 'mental model' in cognitive processing,

- the types of knowledge in the mental model,

- the ways in which the knowledge is implemented,

- the factors affecting what is in any one specific mental model,

- the factors affecting what is in an investigator's model of a task doer's mental model.

These points will be explained in a little more detail in this section. The paper will then concentrate on users'/ process operators' mental models. For a wider ranging discussion of the term 'mental model', see Wilson and Rutherford (1989).

1.1.1. The contribution of mental models in cognitive processing

There has been a great deal of confusion because the term 'mental model' has been used to refer to two different types of contribution to cognitive processing. If the term 'mental model' refers in general to ways in which knowledge is used in cognitive processing, then this paper can also include a third type.

a. Knowledge of the permanent or potential characteristics of some part of the external world.

This sort of knowledge is used in inferring what is happening in some part of the external world which cannot be observed directly, in predicting what is going to happen next, or in explaining or choosing what to do. This sort of knowledge has always been called a 'mental model' in the process control literature (e.g. Edwards and Lees, 1974), and by Gentner and Stevens (1983). This sort of knowledge may also be called long-term memory, or the knowledge base.

b. Temporary inferred knowledge about the present or predicted state of the external world.

In complex tasks, cognitive processes build up a structure of inference about what the available evidence 'means'. This structure has been called a 'mental picture' in the process control literature, e.g. Kraagt and Landeweerd (1974), but because 'mental picture' was often interpreted as implying a mental image this term has not been used recently.

Johnson-Laird (1983) calls this temporary structure of knowledge a 'mental model', which has caused confusion in process operator research. This knowledge has also been called short-term memory, operational memory, working memory, or the blackboard.

c. Knowledge of the properties and outcomes (permanent or potential characteristics) of the user's own behaviour.

This will be explained in Section 5.

This paper will distinguish between these three 'mental model’ contributions to cognitive processing by calling them 'knowledge base', 'working storage' and 'meta-knowledge' (another term with multiple meanings) respectively.

1.1.2. Types of knowledge
Someone doing a complex task may use many types of knowledge : about physical structure, causes and effects, geographical position, visual appearance, goals and constraints, operating procedures, associations between appearance and meaning, etc., see Section 4.

1.1.3. Implementation of knowledge
Knowledge may be embodied in various forms, with various levels of accuracy.

1. Mechanisms for embodying knowledge.

Many underlying mechanisms for knowledge representation have and are being suggested : propositions, production rules, frames, parallel-distributed processing, images, mental videos, etc. There is much debate in the psychological literature about whether different types of representational mechanism do exist in the brain, or whether they can all be reduced to one. For practical purposes this issue can usually be sidestepped. For example if some display format, perhaps a picture rather than a written description, can be used more easily by people doing a particular task, then this is more important than which of these representations might be prior in the brain.

2. The accuracy of the knowledge representation. A user's knowledge may be fuzzy, incomplete, or incorrect.

1.1.4. Factors affecting a specific mental model
The knowledge that a specific person has about their task depends not only on the task aims, the device and environment, their training, and the amount and type of experience, but also on the emotional, social and cultural context, and on the individual's cognitive style.

1.1.5. The investigator's model of the user
The model which someone who is describing a user/operator themselves finds most useful depends on their aim : whether this is to do elegant experiments, to simulate overt behaviour or underlying cognitive processes, or to make practical design recommendations, etc. In fact this is a recursive topic, as dimensions 1 to 4 also all apply to the investigator's model of the user's mental model!

It is important to consider dimensions 1.1.4 and 1.1.5, because any model must be a subset of what is potentially available to be represented. The most efficient subset to choose has a style and content of presentation which makes explicit what is most relevant, useful and usable in thinking about the current task, whether this is the task of the investigator or of the user. There is a lot of unnecessary debate about models, because investigators with one set of aims find models useless which were devised for other purposes (Bainbridge, 1986).

As there are many alternatives on each of these 5 dimensions, the 'problem space' of mental models is huge. This has at least 3 implications:

- it accounts for knowledge elicitation difficulties, as any one method can access only part of all this,

- it means that any one investigator is only concerned with a few cells of this space, and needs to be clear about which, to avoid unnecessary conflict,

- in my view it means that it is not particularly useful to use the term 'mental model', as this is not sufficiently specific. Instead it is important to be clear about which of all these possibilities one is referring to.


Of the above dimensions, this paper focusses on the uses of knowledge in cognitive processing. It will not debate the ways in which knowledge is represented in the brain, and it is not concerned with individual differences in mental models. It does present one investigator's model of the user operator. As for the aims of this paper, first, five general points:

- The data to be accounted for are verbal protocols and verbal reports collected from industrial process operators.

- The methodological approach has been to ask what cognitive model is needed to account for the cognitive behaviour expressed. The model concentrates on cognitive skill [doing known tasks] rather than on problem solving [coping with new situations].

- Both process operation and modelling process operation are huge tasks, with thousands of variables. As the size of the task to be accounted for goes up, so the elegance and strength of the proofs given, and claims made [about the task], must go down. This means that, in this sort of modelling enterprise, the ideas are justified when they are useful rather than because they are completely testable.

- This paper is concerned with 'meta-theory', with suggesting general mechanisms which underlay many applications, rather than with instantiating the details of any particular application.

- Although the concepts are presented more explicitly than they often are in papers on mental models, they have not been programmed, so people who like fully specified models will find this frustrating. The aim of this paper is not to consider what can be done in the way of cognitive modelling within the limits imposed by any available computer tool, but rather to start to identify what are the requirements for a model of cognitive processes in complex tasks, and to present a model which is in sufficient detail to have practical implications.

The main sections of this paper will be on:

- an example of process operation, to illustrate some of the issues involved,

- the cognitive processing modules which recur in process operation,

- the types of working storage, and knowledge bases, used by these modules, and the knowledge bases used in problem solving,

- the suggested cognitive mechanisms which underlay the modules, showing the place of knowledge bases, working storage, and meta-knowledge in cognitive processes. This will be in three sections, on :

  - the goals-means structure of cognitive processing,

- an example paper-and-pencil simulation,

- the nature of cognitive skill, and the integrated and redundant structure which is built up from knowledge bases, working storage and cognitive goals.

The final section of the paper is briefly on the implications for theory and for ergonomics practice.

More specifically, and in summary, the main features of cognitive processing in industrial process operation which can be inferred, which need to be modelled, and in which 'mental models' play some part, are that : 

- while operating complex industrial processes, the operators build up overviews in working storage of the present state of the process, what they expect it to do, and their own future plans of action. These overviews provide the context for their future decision making. Some of these cognitive processes can be modelled by concepts used in modelling language understanding and reading : 'frames' for structuring information and information search, structured 'blackboard' memories, and 'scenarios' for predicting events. But there are additional features in process operation :

    - the operators are interacting with a changing and independent external world. They know about process structure and function, and can mentally simulate its behaviour.

    - the operators' goals are not primarily cognitive goals of 'understanding', as in reading, but task goals of controlling the external world, and to meet these goals they have strategies and make plans. (The ways in which operators add goals, and develop strategies and plans, are not discussed in this paper.)

    - their behaviour is flexible, depending on the task context, the workload, and the operator's level of skill.  [A possible mechanism for this flexibility is suggested, see Figure 5.  But its relation to skill and workload are discussed elsewhere, e.g. Bainbridge, 1978, 1989b.]

    - much of the task thinking may be done non-verbally (this paper only mentions this in passing).


2. An example of industrial process operation

Tables 1 and 2 show a short sequence of operator activity while diagnosing and responding to a process fault. The tables are adapted from an Operator Decision Action Summary given by Pew et al (1981), who analysed what happened during the Prairie Island nuclear power station incident, after extensive discussions with the operators and other experts. [Table 3 gives an example of operator cognitive activity in another task.]

Table 1 

Prairie Island Nuclear Power Station Incident, time 1414

(adapted from Operator Decision/ Action Summary in Pew et al, p. B-12)

Table 2

also Adapted from Pew et al (1981), time 1421

Table 2 shows what happened 7 minutes later, after the alarm had gone off again, another possible interpretation had been checked, and several other pieces of information had appeared on the displays. Again, the operators inferred what was happening inside the process. Given the displays they had, there was no way in which they could get direct information about a leak (rupture). 

They then worked by anticipation. They predicted what further changes would happen in the process, and predicted what automatic safety actions would be triggered off by these future changes.  [Note the switch from bottom-up processing, reacting to an alarm, to top-down processing, actively looking for/ expecting information.]

They then predicted the effect of these automatic events, and chose and implemented actions to reduce the undesirable effects of the expected events. In this way they were using anticipatory control. They were not reacting by feedback to minimise an undesirable state which had occurred : they were anticipating, to try to prevent an undesirable future state from occurring.

Modelling industrial process operation involves modelling the cognitive processes of :

- interpreting incomplete data, 

- anticipating and planning actions with reference to expected events. 

In all these types of activity, having 'mental models' of the process, enabling operators to know the relation between evidence and inference and to predict process behaviour, are essential. The rest of this paper will discuss these points more fully.


3. The Cognitive Modules in Process Operation

This section gives more complete information about the types (modules) of cognitive processing involved in process operation. The full supporting evidence is not given in detail. [There is much more information in the Introduction and first half of Bainbridge (1975).] The activities which are proposed are sufficient to account for the data available on cognitive processes in four nuclear incidents (Pew et al, 1981; Woods, 1982), part of a steelworks (Bainbridge, 1974), gas distribution grid controlling (Umbers, 1979) and a commercial bakery (Beishon, l969).

To analyse/ understand complex cognitive activity, it is necessary to distinguish between cognitive goals and task goals. For example a task process control goal might be 'keep the temperature at 200º C'. Cognitive goals involved in meeting this task goal might be to find out: 'what is the temperature now ?' 'is the temperature acceptable ?', 'what is the best way of restoring the temperature ?' Data on operators' moment-to-moment thinking (as reflected in verbal protocols) show that their thinking can be divided into a sequence of short units, each of which has a different cognitive goal.

This Section aims to outline how these units or 'modules' of cognitive processing have been identified, to list the modules, and to suggest the main implications for mental models and cognitive processing in complex tasks such as process operation. Later sections will suggest more basic mechanisms which underlie, or implement, these cognitive processing modules.


In Tables 1 and 2 the operators' cognitive activity was divided into a sequence of sections, and the cognitive purpose of each section was suggested. Table 3 gives another example, part of a verbal protocol from an operator controlling the availability of gas supplies (Umbers, 1976).

Table 3

Adapted from Umbers (1976, p. 321) study of gas-grid control

Time 22.07

Main text - verbal protocol from gas-grid controller [cp. left of Tables 1 and 2]. 

Bold comments - inferred cognitive goal of each protocol section [cp. right of Tables 1 and 2].


I'll give you a quick resume of what the forecast was this morning, 

that is forecast estimate, 

and the estimate for this morning was 75 of MG [manufactured gas) and the NG [natural gas] send of 150 m.

(predict state)

the stock aim on both MG is 62.6 and NG 180.7

(review goals)

we have 3 plants at work, the 3b at C/hill, the 1c at Tipton, and No. 3 at W H, this gives us a total cum make per hour. of 4.2 approx without B/ air

(review present state)

B/ air will give us something in the region of -

sorry, a correction to the total available without B/ air would be approx 4.4 

and there's 0.8/hour. of B/ air, 

so that we can in fact make 5.2 with full NG, 

This is something in the region of 129-130 for the day.

(review action availability and effect)

we don't contemplate using this much


but there is the facility for taking NG off the reforming and putting Naphtha in, 

which will give us another increase of something in the region 0.30-0.35/hour.

(review action availability and effect)

so that we're not too badly off


if the weather forecast which they give for tomorrow has been mostly sunny and warm, 

with a max of 68 and a mean of 58 F.

(predict goals)

we shall cope with the send on MG with what we've got quite admirably. 


The sequence of analysis has been:

1.The cognitive goal underlying each piece of evidence about cognitive processing (phrase in the protocol) is inferred. (In many tasks, cognitive goals are rarely stated explicitly.)

2.The sequence of cognitive activity is then divided into subsections, by grouping together all the items concerned with the same cognitive goal, and defining a new section when the cognitive goal changes.

These subsections have been called 'modules' of processing, because they seem to be independent, in the sense that they can occur in any order, though they are not independent in the sense that they do not relate to each other (see Sections 4.1 and 6).


The evidence that cognitive activity can be divided into subsections each with a different cognitive goal comes from [analysis of] verbal reports made by operators while actually doing a task, as in Bainbridge (1972), Umbers (1976, see Table 3) and Reinartz (e.g. 1989). This notion has then been used in inferring the modules used in cognitive activities which have been analysed and described in less detail by Beishon (1969), Pew et al (1981), and Woods (1982). This widens the number of task types accounted for.

[In summary the cognitive analysts’ inferences about the cognitive tasks which operators do, in tables 1, 2, 3 are:

interpret what information implies about present state

review present goals/intentions

evaluate state

predict future state

predict future goals/intentions

evaluate future state

review actions available to meet goals

evaluate possible action effects

choose actions to meet goals.

Five of these modules are analysed further in Table 4: the possible knowledge bases used, and the working storage items used and added, by each module have been inferred by an analyst.]

Table 4 : Working Storage (WS) and Knowledge Bases (KB) referred to during five proposed cognitive processing modules.

Most Modules also refer to the environment.

[There are 4 levels of indent in this representation :

WS : items available in Working Storage and used in the processing (some WS items have been indented too far)

KB : items in Knowledge Base referred to by this processing

MODULE (cognitive processing task)

WS : items added to Working Storage by this module.]

In Table 4, the basic modules proposed are listed in bold type. This is a preliminary list for discussion and to illustrate the arguments, as only a small range of tasks has been surveyed. In particular, the modules used in problem solving are not included. And these modules are not all needed in all tasks.

The key concepts are:

- cognitive processing is done by modules each of which has a particular cognitive goal, which is to build up a temporary structure of information about the current state of the task.

- the primary concerns of the cognitive processing, before choosing and making actions to operate the process, are to:

- infer and review the state of the process,

- anticipate and plan future events and activities.

- the modules refer to information which is available in knowledge bases, and already in working storage (i.e. to 'mental models'), as well as to the environment.

The justification for this sort of notion of the basic cognitive processing module, and additional points about the modules, are given in Sections 5 and 6.


4. Summary of Knowledge Bases and Working Storage

Table 4 lists the sources used by, and outputs of, the basic modules. It shows the working storage and knowledge bases referred to during cognitive processing, on the left. All the modules refer to the environment and to the previous output of the same module, so these have not been mentioned explicitly. The working storage contributed from the processing modules is shown on the right.

These types of working storage and knowledge base are not based on direct evidence. They are inferences, on the basis of observed cognitive activity, about what would have to be the working storage and knowledge bases needed to simulate this module of activity, in the way outlined in Section 6.1.


Figure 1 : The main interdependencies between aspects of working storage.  

[These are all items which the verbal reports show the operators think about.

The arrows indicate that one item is used in finding another.]

Figure 1 lists the main contents of working storage, and illustrates how one part of working storage depends on another. There are times when this whole structure has to be built up from scratch and in sequence, for example when the operators come on shift, or when there is an unexpected event such as a major fault. Otherwise, the existing working storage [describing the current and future process state] is referred to and revised as appropriate (this depends on working storage capacity, but there is not space to go into this here, see e.g. Bainbridge, 1975).


The knowledge bases fall into two groups, those referred to by several process modules, and those which are referred to by only one module. (There is not space in this paper to discuss the underlying mechanisms by which the different types of knowledge are represented, see e.g. Bainbridge, 1988.)

4.2.1. Knowledge Bases used by several aspects of processing

There are three groups of knowledge which are referred to for several purposes. All of them are externally defined information about the task, and tend to be taught in the more formal off-the-job training of operators. The groups are:

- product targets and plant constraints, i.e. the criteria to which the operator is working,

- operating procedures, the formally defined working methods,

- dynamic models of the plant, and scenarios of events.

Within this last area of knowledge, the central notion is that operators use some sort of dynamic model of the process for predicting its behaviour over time. Indeed some authors restrict the term 'mental model' to this.

Figure 2 : Some of the necessary components in an operator's dynamic 'mental model' of a phase of a process

[circle : variable/ state

box : processing

circle with cross : comparison.]

An operator also needs to know that some process states initiate a major change in the process, for example:

- when temperature greater than 100º C, water turns to steam,

- when pH greater than y, product is precipitated,

- if pressure less than x, automatic shutdown is initiated,


These are called a change in 'phase' of the process, because different dynamic models are needed to predict how the process will behave, on each side of the discontinuity.

So it is possible to suggest that the operator's knowledge of the process has two levels, both of which are used in anticipation.

- a selection of dynamic models of process behaviour in various phases, for which the necessary components are illustrated in Figure 2,

- knowledge of scenarios (see Figure 3) describing the sequence of events (phases) :

- during start-up, shutdown, or batch processing,

- after a given fault has occurred.

Figure 3 : Some components of an operator's knowledge of the sequence of phases in a process

4.2.2. Knowledge Bases used by only one aspect of processing

The second group of knowledge bases are each used only by one processing module such as :

stimulus : identity, 

symptom : underlying cause, 

state : response required, 

effect required : action with this effect, 

action : effect, 

action : required preconditions.

In each case, in well-developed cognitive skill these items have specific links to identity, action, result, etc., not the use of a general principle [or 'mental model'] to generate the answer. 

Interestingly, at least for this inferred evidence, these knowledge bases used by only one module are of the 'association' type, i.e. 'if x then y', and need to be learned from repeated experience.


The above knowledge bases are used in cognitive skill, that is, when working methods and reference knowledge are well-established and readily available (see Section 6.2). When an appropriate working method and background knowledge are not known then problem solving is needed (Bainbridge, 1989b). It is generally assumed that operators then use information about the causal, functional and physical structures of the plant, and the goal structure of the task, to work out from first principles what is wrong, or what to do. Most reviews of operator knowledge focus on these types of knowledge, e.g. Rasmussen (1979), Bainbridge (1988), but there is not space to go into this here.


5. Goals-Means Structure of the Task

Some aspects of the basic cognitive processing module have not yet been discussed. These include the need for the third type of 'mental model' suggested: a skilled person's model of the properties of their own behaviour. Two main points will be covered in this section:

- the main features of a cognitive processing module,

- the goal-means structure of the task, the need for choice between alternative methods of working, and the need for the third type of 'mental model'.


Figure 4 : Components of a basic processing 'module'.

[transfer of control arrows on left]
The main aspect not shown here is knowledge about the properties of alternative methods for finding the answer needed to meet the cognitive goal, see next Section and Figure 5.

Figure 4 gives more detail about the suggested mechanisms in the basic cognitive processing module. (It may be easier to understand the reasons for points made in this section after reading Section 6, which describes a specific example.) There are six important features [evidence for these features is in e.g. Bainbridge, 1975] :

1. The aim of each processing module is to meet a particular cognitive goal, such as to find what is the present temperature, or to choose an action.

2. The 'answer' is found (stepped arrow) by referring to the environment or to knowledge, or by further cognitive processing. This processing is carried out by modules at a lower level (see 5.2 and 6.1). Each module might be likened to a complex slot in a 'frame'.

3. The modules actively search for the information they need, rather than passively responding to information as it arrives.

4. Finding the relevant answer is done within the context provided by existing working storage, represented by the left hand of the two boxes.

5. The answer, represented by the right hand box, itself becomes part of working storage. It provides the answer for a superior goal, and/ or it becomes part of the data used in later cognitive processing (see Figure 8). Working storage is therefore structured by its place in the goal/module structure of the task. 

6. The modules communicate with each other via working storage, and are sequenced via working storage.


The stepped arrow represents the link between the goal and the method for meeting it. The special arrow acts as a reminder that this is not a simple link. It is not a simple link because behaviour is flexible, and it is easiest to represent this flexibility by goal-means independence (Bainbridge 1975, 1978). A goal may be met by several methods. Which of these it is best to use will depend on details of the context. And a given method may be used in meeting several goals. So the goal-method link needs to be flexible, and to include the choice of method.

The suggested mechanism for this is that each method has stored with it data about its general properties : how long it takes, how accurate the result is, how much effort it needs, etc. On the basis of this information, the method is chosen which best fits the context. Figure 5 suggests, as an example, some dimensions of choice between topographic and functional strategies (Rasmussen and Jensen, 1974) in fault diagnosis.

Figure 5 : The operator's knowledge about the properties of alternative working methods, used in choosing between them.

(The example strategies come from Rasmussen and Jensen, 1974, the Figure does not.)

[Left column : names of properties.

Right column : comparison context information.

Inner columns : properties needed/ supplied by working methods.]

It is possible to suggest that one aspect of cognitive skill is that these data about working methods are ready available, so choosing appropriate behaviour is automatic/ unconscious (Bainbridge, 1978). The data, and the relevant dimensions of behaviour to consider, would have been learned from past experience. This information could be considered as a third type of 'mental model', of knowledge about the properties of one's own behaviour, an aspect of meta-knowledge.


6. Processing and Skill

The basic theme of this paper is that the development of process operators' cognitive skill has built up an integrated cross-referring structure, in which the working methods (structured relative to the cognitive goals) refer to knowledge about the process (a first type of mental model) and knowledge about one's own behaviour (the third type of mental model above) to build up in working storage an overview of the current state of the task (the second type of mental model above). This section gives an explicit example, and lists the implications for the nature of cognitive skill.


The example is a paper-and-pencil simulation of the activity in Table 1. Figure 6 shows a goal-means hierarchy on the left (hierarchy shown by level of indent). (This has been inferred for this example. In other data there is more direct evidence on the processing used, e.g. Bainbridge, 1972).

Figure 6 : An example of a working method described using the proposed processing 'module' [assumed to underlie the evidence in Table 1]. This Figure shows the links in transfer of control. For other aspects see Figures 7 and 8.

[Knowledge base in Figure (9.)7 :

possible causes - (a) in Figure 7

probability - italics in Figure 7

evidence - (b) in Figure 7]

References to the knowledge base and to the environment are shown on the right. The relevant knowledge base is shown as a network in Figure 7. It has been assumed that the working method is more general, while the knowledge base is specific to this particular alarm.

Figure 7 : The knowledge base referred to by the working method in Figure 6 [and representing 3 possible reasons for the alarm observed, as in Table 1]. It is suggested that the working method and knowledge base are structured relative to each other, as a result of practice.

[The arrows just indicate items are related, not intended to imply any particular storage mechanism.]

Working storage is also being built up, though it would be too confusing to show this in Figure 6 as well. Two types of cross-reference in working storage are involved, and are shown in Figure 8 :

- the working storage used as data, or context, for later processing [left box of pair],

- the answers to superior cognitive goals, which are transferred up the hierarchy[right box].

Figure 8 : The working method showing the 'transfer' of working storage items between different processing elements, and how this helps to bind the elements together.

[This diagram uses the two boxes representation, as in Figure 4.

The left box is input information, supplied by the output of previous processing.

The right box is the output of the processing which, if not filled by processing represented here, is otherwise as in Figure 6.  

The arrows in this diagram indicate cross references, not sequences of processing as in some earlier figures.]

The choice between alternative working methods, which is also happening, is not shown explicitly in these diagrams. (For an example see Figure 5.) It is possible to suggest that working storage provides the context information, i.e. time available, accuracy required, etc. against which the characteristics of the available methods are compared [right column of Figure 5.]. An example where this choice would be needed in Figure 6 is at the point where the operators choose a method for selecting between hypotheses about the possible causes of the displayed information. In this example the operators used Reason's (1982) 'frequency matching' strategy.


In summary, this model suggests that the main features of cognitive skill are well-developed working methods and mental models :

[knowledge bases]

- sufficient knowledge bases from which the operator can infer states and anticipate events,

- for 1:1 information, well practised specific knowledge bases can be accessed automatically,

[working methods]

- fully developed working methods built up as a structure of cognitive goal modules; this structure :

- knows how to access relevant information in the knowledge bases and environment,

- produces and structures working storage,

- is activated by, and works within the context of, working storage,

[meta-knowledge about methods]

- each working method has stored with it information about the quality and quantity of its needs and outputs; these are used in choosing the best working method as a function of the context [which is in working storage],

[working storage]

- working storage maintains overviews of the state of the task.


7. Implications for Theory and Practice, Some brief Notes

From the theoretical point of view, this whole structure provides a model for (some of) the mechanisms involved in complex cognitive skill. Section 6 illustrated what is meant by cognitive skill as integrated redundant processing. It is integrated and redundant because the goal-means structure, the working storage and the knowledge bases all cross refer to each other and reinforce each other. This illustrates the unified structure, but a great deal of work would be needed on specific instantiation before it could actually be used for cognitive modelling of users/ operators.

Although mental models are part of user modelling they are not the whole. To predict all aspects of user behaviour, user models may also need to include other aspects of cognitive processing, such as pattern handling, cued recall, the time to translate from one representation to another, the non-mathematically optimum use of evidence, and physical activity. In many practical situations, considering these aspects will be more important in improving users' performance than any consideration of mental models (Bainbridge, 1989a).

However, in general, the types of cognitive processing module and knowledge base, and the types of working storage built up, do have practical implications. The main basis to these recommendations is that no part of a complex task is done in isolation, as is illustrated by the model.

Training, task analysis, and job aids such as displays, need to be oriented to helping the operators to build up the contextual overviews, and to work oriented to the future (Bainbridge, 1991a).

The types of knowledge base used are also relevant in training and in display format design. The content of the different knowledge bases, and their type and relevance to different aspects of processing, need to be considered in the design of training schemes (Bainbridge, 1990a). Display formats should be compatible with the type of knowledge represented (Bainbridge, 1990b). However, many recommendations about details of interface design for process operators do not need to be based on an explicit model of cognitive processes at the level of detail presented in Section 6.1 (Bainbridge, 1989a, 1990b).

The need for the overviews, and the time taken to build them up from scratch, also have implications for job design. This is particularly true for the design of automated systems so that operators can either maintain an overview, or have time to build one up, and can maintain the cognitive skills needed, when expected [perhaps without warning] to take over manual operation from an automated system (Bainbridge, 1983, 1991b).

Much of the development of this paper was done while I was Visiting Research Fellow in the Ergonomics Workgroup, University of Twente, The Netherlands. I would like to thank Dr. Ted White and his colleagues for an excellent and happy research environment.


BAINBRIDGE, L. (1972). An Analysis of a Verbal Protocol from a Process Control Task. Unpublished PhD. Thesis, University of Bristol.

BAINBRIDGE, L. (1974). Analysis of Verbal Protocols from a Process Control Task. In : E. EDWARDS and F.P. LEES, Eds, The Human Operator in Process Control, pp. 146‹158. London : Taylor and Francis.

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, pp. 165-183. London : Taylor and Francis.

BAINBRIDGE, L. (1978). Forgotten Alternatives in Skill and Workload. Ergonomics 21, 169-185.

BAINBRIDGE, L. (1983). Ironies of Automation. Automatica 19,775-779.

BAINBRIDGE, L. (1986). What Should a 'Good' Model of the NPP Operator Contain ? In : Proceedings of the International Topical Meeting on Advances in Human Factors in Nuclear Power Systems, American Nuclear Society, pp. 3-11. Knoxville, Tenessee, April 21-24.

BAINBRIDGE, L. (1988). Types of Representation. In : L.P. GOODSTEIN, H.B. ANDERSON and S.E. OLSEN, Eds, Tasks, Errors, and Mental Models, pp. 70-91. London : Taylor and Francis.

BAINBRIDGE, L. (1989a). Cognitive Science Approaches to Process Operation : Present Gaps and Future Requirements. In : Proceedings of the Second European Meeting on Cognitive Science Approaches to Process Control, pp. 1-9. October 24-27, Siena, Italy.

BAINBRIDGE, L. (1989b). Development of Skill, Reduction of Workload. In : L. BAINBRIDGE and S.A. RUIZ QUINTANILLA, Eds, Developing Skills with Information Technology, pp. 87-116. Chichester: Wiley.

BAINBRIDGE, L. (1990a). A Note on Training Sequences and Simulator Facilities for Industrial Process Operation. In : Proceedings of the Meeting on Operator Training and Acquisition of Cognitive Skills. School of Psychology, University of Wales College of Cardiff, September 27-29.

BAINBRIDGE, L (1990b). Multiplexed VDT Display Systems. In: G.R.S.WEIR and J.L. ALTY, Eds, Human-Computer Interaction and Complex Systems, pp. 187‹208. London : Academic Press.

BAINBRIDGE, L. (1991a). Cognitive Context Analysis. International Journal of Human Factors in Manufacturing.

BAINBRIDGE, L. (1991b). Will Expert Systems Solve the Operators' Problems? In:

R. ROE and M.ANTALOVITS, Eds, Technological Change Process and its Impact on Work.

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, pp. 79‹90. London : Taylor and Francis.

EDWARDS, E. and LEES, F.P., Eds (1974). The Human Operator in Process Control. London : Taylor and Francis.

GENTNER, D. and STEVENS, A.L., Eds (1983). Mental Models. Hillsdale, NJ: Erlbaum.

JOHNSON-LAIRD, P.N. (1983). Mental Models. Cambridge, England : Cambridge University Press.

KRAGT, H. and LANDEWEERD, J.A. (1974). Mental Skills in Process Control. In: E. EDWARDS and F.LEES, Eds, The Human Operator in Process Control, pp. 135-145. London : Taylor and Francis.

PEW, R.W., MILLER, D.C. and FEEHER, C.E. (1981). Evaluation of Proposed Control Room Improvements Through Analysis of Critical Operator Decisions. Palo Alto, Calif: Electric Power Research Institute, NP- 1982, Research Project 891.

RASMUSSEN, J. (1979). On the Structure of Knowledge - a Morphology of Mental Models in a Man-Machine System Context. Roskilde, Denmark: Riso National Laboratory, Report M-2192.

RASMUSSEN, J. and JENSEN, Aa. (1974). Mental Procedures in Real-Life Tasks: A Case-Study of Electronic Trouble Shooting. Ergonomics 17,293-307.

REASON, J. and MYCIELSKA, K. (1982). Absentminded? The Psychology of Mental Lapses and Everyday Errors. Englewood Cliffs : Prentice-Hall

REINARTZ, S.J. and REINARTZ, G. (1989). Verbal communication in collective control of simulated nuclear power plant incidents. In Proceedings of the Second European Meeting on Cognitive Science Approaches to Process Control, pp. 195-203. October 24-27, Siena, Italy.

UMBERS, I.G. (1976). A Study of Cognitive Skills in Complex Systems. Unpublished PhD Thesis, University of Aston in Birmingham.

UMBERS, I.G. (1979). A study of the Control Skills of Gas Grid Control Engineers. Ergonomics 22, 557-571.

WILSON, J.R. and RUTHERFORD, A. (1989). Mental Models : Theory and Application in Human Factors. Human Factors 31,617-634.

WOODS, D.D. (1982). Operator Decision Behaviour During the Steam Generator Tube Rupture at the Ginna Nuclear Power Station. Westinghouse Research Report 82-1C57-CONRM-82.

Access to other papers via Home page

©1997, 2022 Lisanne Bainbridge


Popular posts from this blog

Ironies of Automation

Types of skill, and Rasmussen's SRK schema

Complex Processes Review : References