7. Meta-knowledge, mental workload

Topics in this section :

7.1. Choice of how to meet a given 'cognitive goal’ on the basis of meta-knowledge.

7.2. Goal-means independence.

7.3. Control of mental workload.

7.4. Multi-tasking.

Summary of main points in Section 7.

Section 7 : Meta-knowledge in choosing how to meet the cognitive needs, with implications for the control of mental workload.

Building up behavioural complexity from a cognitive processing element.

Lisanne Bainbridge

So far the discussion of behavioural flexibility has focused on the sequence of thinking about different aspects of a task (Sections 4 and 5). In the cognitive mechanism proposed in this paper, this choice is made by 'sequencers' including conditional processing elements. These 'sequencers' both build up an overview of what is happening in the task, and then use this overview as the context for choosing the optimum cognitive need to think about next.

Another important aspect of behavioural flexibility is the way in which different working methods can be used to meet a given goal or need, depending on the circumstances. For example, the steelworks operator sometimes obtained a numerical value by calculation, sometimes by judgement. So there must be some mechanism which can choose whatever working method is most appropriate at a particular moment.

So far, the stepped arrow part of the cognitive element has just been used as a symbol for : to meet this need use this method ('routine'). The stepped arrow is actually a much richer symbol : for a decision mechanism which provides the flexibility in how a given need is met, which has other interesting powers. The notion is that the cognitive need and the working method(s) for meeting it are independent, and that the link between them is the point of choice between working methods, by comparing meta-knowledge about the methods with the context at the particular moment. The place of this link in the control of mental workload is discussed in part 7.3. The way in which it might be expanded to account for multi-tasking is noted very briefly in part 7.4. Its possible role in learning is discussed in Section 8.

Repeat of Figure 3c.4 - a cognitive need may be met by linking to a 'routine' (or by referring elsewhere to where it was found previously) . The stepped arrow link to a routine symbolises that this link is also a choice point between alternative methods for getting the result needed.

For more discussion of meta-knowledge and mental workload see :

(1974) Problems in the assessment of mental load

(1978) Forgotten alternatives in skill and workload

(1975) Working memory in air-traffic control.

(1993a) Types of hierarchy imply types of model

(As usual in my papers, the word 'skill' means level of expertise from beginner to expert, not a particular type of cognitive processing.)

7.1. Choice of working method based on meta-knowledge

The method-choice mechanism suggested is that each working method has associated with it meta-knowledge about its properties, for example how long it takes, how accurate its result is, how much mental effort it takes, etc. This meta-knowledge about the method is then compared with the context (which is in working storage) : how much time and mental effort are available, what level of accuracy is required, etc., as a basis for choosing the method which is the best fit to the present situation. For a working method to be usable for many purposes, especially in contexts in which it has not been used before, its meta-knowledge would also need to include data about the general nature of the outcomes that can be obtained by using it. Which dimensions of meta-knowledge are needed will depend on what is important at this point, see briefly Section 8.2.1.

The meta-knowledge, and what it is compared with, may be information about :

- the state of the task.

- the 'state' of the person who is doing the task, how tired, interested etc. they are, i.e. what they are willing and able to 'invest' in doing it. For example, someone’s working behaviour may change if there’s someone nearby who they want to entertain or impress, or who will report them to the shop steward.

The stepped arrow in the processing element diagrams represents the point at which the choice between alternative working methods is made. Figure 7.1 below gives a simple example. In his verbal protocol the furnace operator, when predicting the power usage after a change in the steel-making stage a furnace was in, sometimes made this prediction by judgement and sometimes by calculation. Figure 7.1 suggests some of the dimensions the operator might consider in choosing between these two methods. Note that in the operator's verbal protocols there was explicit evidence that an operator used different methods for doing the same task, but there was no direct evidence about how the choice between them was made. As there is no explicit information, Figure 7.1 is based on analyst's inference. When there is no direct mention of this decision in the verbal protocols, this might suggest that the decision is made unconsciously.

Figure 7.1 : Obtaining a quantitative answer, the choice between judgement and calculation (Bainbridge, 1975, Figure 2).

Upper box - meta-knowledge about method.

lower box - working method.

Figure 7.2 gives another instance. Rasmussen and Jensen (1974) found that electronic maintenance technicians used three alternative strategies for doing their fault diagnosis, and Rasmussen and Jensen suggest reasons why the technicians might choose between these strategies. Figure 7.2 is a possible model of the choice between two of the strategies, using the mechanism proposed here. It shows two strategies, each with three properties which are compared with dimensions of the context at the time of choice. This mechanism was suggested in Bainbridge, 1975.

Figure 7.2 : The choice between topographic and functional strategies for fault diagnosis (Bainbridge, 1975 Figure 5, based on evidence in Rasmussen and Jensen, 1974).

Two important additions are needed to those Figures.

Firstly it is useful (see below, part 7.3) to distinguish between task related dimensions of the meta-knowledge about a working method (e.g. what accuracy of result it gives, how long it takes) and personal dimensions of the meta-knowledge (e.g. how difficult the person finds it to do, or their expectation of success).

Table 7.1 describes the same example as in Figure 7.1, but expanded to make these two aspects explicit.

Table 7.1 : A possible tabular representation of the meta-knowledge in Figure 7.1 (adapted from Bainbridge, 1978, Table 1).

Secondly, it is suggested that a working method is chosen by comparing this meta-knowledge about the alternatives with dimensions of the context in working storage : for example, how much mental effort the person is willing to make, how much time is available, what accuracy is required.

These points suggest several additional theoretical considerations. If the choice between methods is made by comparing meta-knowledge with the task context, this expands the notion of context used in previous Sections, to include not only aspects related to task needs, such as how urgent it is to make an action or what accuracy is needed, but also aspects describing personal needs, such as how much effort the person feels like making, or how much they want to succeed.

So the choice mechanism, with its parameters, could then be the locus of operation for several personal effects. For instance, the 'self' aspects considered by someone when choosing a way of doing things could include whether it is amusing or interesting, as well as aspects of self-image such as expectation of success, learned helplessness, fear of failure or success, and other positive or negative emotions.

The meta-knowledge about what contextual task criteria a particular method needs or meets can only be learned by using the method in doing the task.

A decision mechanism is needed, for comparing the meta-knowledge about methods with the context to make the choice between methods. The reference to context suggests that the two sorts of choice involved in adaptive behaviour which have been discussed (i.e. the choice of which part of the task to do next and the choice of how to do it) are made by interrelated mechanisms. Both choices involve comparison with the context, and similar dimensions (such as how urgent an action is) are involved in both choices. The contextual choice between cognitive needs, the choice about what part of the task to do next, has been described as being made by conditional elements in the 'sequencers' (Sections 4 and 5). The contextual choice between working methods for meeting a given cognitive need has been described by conditional elements in a 'routine' (e.g. Figure 5.4) or by multidimensional comparison of meta-knowledge (above). These are not necessarily two distinct types of decision mechanism. The conditional elements in a 'sequencer' actually carry out multidimensional comparisons, but in a way which has several advantages, see Figure 3c.4 and discussed a little more in Sections 5.4 and 5.6.1. It is possible that the meta-knowledge decisions could also use this double mechanism, of referring to the available data in parallel, or using a working method to find missing meta-knowledge if necessary.

If the factors determining what is the best way of meeting the current cognitive need (which may not be in conscious awareness) are already known and could be described by a multi-dimensional look-up table, then the suggested mechanism is as in this Section.

If choosing how best to do the required thinking itself needs some thinking, then that thinking might be described by a 'routine', example in Figure 5.4 (and is an important aspect of learning to do a task, see Section 8).

This can be illustrated by an alternative representation of some of the behaviour described in Figure 5.4. This 'routine' determines what is the best current method of assessing the control state, depending on various aspects of the current context.

The operator uses three strategies for assessing the control state, depending on whether the time is in roughly the first 5 minutes of the half hour, the next 10 minutes, or the second half of the half hour. In Figure 5.4 the choices between these strategies have been represented by conditional statements. But they might equally well be represented by a multi-dimensional table.

So the idea is that meta-knowledge about alternative behaviours is compared with current task requirements, which are in the overview.

What is the relevant information best used to decide between alternatives, and the relevant values of those dimensions, must be learned by experience of doing the task.

Whatever may be the mechanism for deciding how to meet a current task need, the personal aspects must be involved in a person’s control of their mental workload, as discussed in Section 7.3.

The mechanisms involved in the choice of both what to do and how to do it must be involved when a person is multi-tasking, as mentioned very briefly in in Section 7.4.

And also in problem solving, planning, and learning, as discussed briefly in Section 8.

7.2. Goal-means independence

If there can be several methods for meeting any one cognitive need, this means that the cognitive need and the means for meeting it must be independent (except as discussed in Section 8.3 on learning). The stepped arrow part of the cognitive element represents a point at which there is a choice about how to meet the cognitive need named in an element. The need and the means for meeting it (the 'routine' or, more generally, the working method) are independent. This independence leads to the mechanism having some more useful and important properties.

It allows the potential for adapting the working method used to the circumstances. Any one need can be met by several working methods, and any one method can be used to meet several needs or purposes. For example, if one wants to hammer a nail in, and a hammer is not available, then anything heavy and solid can be used instead - a brick, a tin of beans ? And a hammer is not only useful for hammering nails in, it can also be used for pulling nails out or as a paperweight.

If the working method and the need for which it is used are independent, then it is important to store the result of using a method with the need, rather than with the method. Then the method can be used for other purposes. This occurs in the cognitive element mechanism, as the results of the main 'routines' are stored in the overview structured by the 'sequencers'.

Hierarchy or heterarchy

Often a given task need is only expanded into a method for meeting it at the time it is considered, to allow for the specific circumstances at that time. This expansion into 'lower' levels of greater detail may not be permanently fixed in a hierarchy (though see Section 8.3 on learning). For maximum flexibility of behaviour, individual items may stored as a heterarchy or network of potential links. The hierarchical nature of behaviour organisation, into different levels of detail, is an emergent property of this network when something is done, one which appears when the behaviour is carried out rather than being fixed in advance, as suggested in Figure 7.3.

Figure 7.3 : Items stored independently appear as a hierarchy when used. If the situation is repeated frequently, then this hierarchical relationship may become more permanent, an aspect of learning, see Section 8.

7.3. Control of mental workload

It is possible to suggest that this choice between alternative working methods, using meta-knowledge about them, accounts for two aspects of mental workload :

- how mental workload is maintained at an acceptable level when task workload changes. This is discussed here.

- the way in which mental effort decreases with increased skill. Meta-knowledge could be involved in the control of learning and the choice of mode of processing ('type of skill'). This will be mentioned in Section 8.3. (see note on meaning of the word 'skill')

There is good evidence that people do choose how to do a task to control their mental workload. These data might be accounted for by a meta-knowledge mechanism.

Sperandio (1972) studied local air-traffic controllers. A local air-traffic controller’s main cognitive goal is to plan how to bring aircraft into an airport safely. Sperandio found that air-traffic controllers could use two strategies to bring aircraft into the airport. The strategies differed in the amount of mental effort required to handle one aircraft. When there were more aircraft to bring in, the controllers changed to the strategy which required less mental work per aircraft, thus maintaining their overall mental workload at a steady level while the task workload increased. Sperandio summarised this finding as in Figure 7.4.A.

Figure 7.4 : Effect of changing working method on relation between task demands and mental workload (Bainbridge, 1974b, part A adapted from Sperandio, 1972).

This is a simplified diagram to make the main point. Of course air-traffic controllers did not exclusively use either Method 1 or Method 2, making a discontinuity in the level of mental workload experienced. They actually mixed the strategies, using Method 2 more often as the task workload increased.

So there is not necessarily a monotonic relation, or direct mapping, between task demands, mental workload, and task performance, as there is in physical work. This means it is important to distinguish between these three when discussing mental workload.

The mechanism of using meta-knowledge in behaviour choice could produce this adaptive behaviour.

Suppose that the controller uses meta-knowledge about task aspects of working methods to choose a working method which meets the performance demands imposed by the task. This adjustment could be modelled as in Figure 7.5.

Figure 7.5 : The adaptation of performance to task demands, via choice of working method (Bainbridge, 1974b, Figure 2).

Suppose also that the controller uses meta-knowledge about self-related aspects to choose a strategy which keeps the amount of mental work done to within unstressful limits. This adjustment could be modelled as in Figure 7.6.

Figure 7.6 : The adaptation of mental work done to available mental capacity, via choice of working method (Bainbridge, 1974b, Figure 3).

The process of coming to a balance between the two demands, of meeting the task performance requirements while keeping mental workload within acceptable limits, could then be summarised as in Figure 7.7.

Figure 7.7 : The balancing of working method to meet both task demands and mental capacity (Bainbridge, 1974b, Figure 4).

The overall result of the operation of this mechanism can be described in several ways. The outcome which emerges from this mechanism, described as the interaction between the factors involved, is represented in Figure 7.7. The overall outcome which emerges from this mechanism, described as the relation between task demands and mental workload, is represented in Figure 7.4A. This way of controlling mental workload was suggested in, and these figures come from, Bainbridge (1974b).

The choice of how to do something could interact with the choice of what to do next. For example, someone might choose to do something only if a method for doing it was available which could be done in the time available. This interaction of what and how might account for some human errors in behaviour organisation, such as when people under stress do what they find easiest rather than what has the highest priority.

This sort of account could presumably be expanded to include multiple resources. There is also the possibility that this choice point could be the locus of operation for various aspects of behaviour which are studied in other areas of applied psychology, such as the effect of motivation and satisfaction on performance, studied in industrial psychology, or the effects of fear of failure and fear of success, studied in clinical psychology.

7.4. Multitasking

(Only a very brief comment !)

Multitasking involves interleaving several different tasks at the same time. Doing this requires a mechanism which can choose which part of which task to do next, on the basis of their urgency, importance, pertinence, etc.

Two areas of choice, underlying behaviour flexibility, have been described in this paper :

- the choice between different parts of the same task, between different main cognitive needs (Sections 4-5),

- the choice between different ways of doing the same task, of meeting the same need (this Section).

These are both decisions which involve comparing the values of several dimensions against the prevailing context in working storage.

I have not worked out an explicit example of how the mechanisms proposed here could be expanded to deal with multitasking. I have not had access to detailed data on someone doing multi-tasking, so anything I could do would just be imaginary speculation. However, I would take these choice mechanisms as the starting point. I assume that the choice of what to do next within one main task, and the choice of which part of which task to do next when multi-tasking, are done by similar mechanisms.

Amalberti (1992) developed a meta-knowledge based mechanism for doing multitasking, in relation to planning, which I hope is compatible with the ideas presented in the present paper.

Summary of main points in Section 7

* The cognitive need and the working method for meeting it are independent. This goal-means independence means that several working methods might be used to meet one cognitive goal, or one working method might be used to meet several goals.

* The result of using a working method is stored with the cognitive need which called on it, rather than with the working method itself.

* Working methods ('routines') have stored with them meta-knowledge about their properties. Some of these properties are task-related, such as how long they take and their outcome, and some are person-related, such as how difficult the person doing the task finds them to do.

* The best working method to use, for meeting a given cognitive need at a particular moment, is chosen by comparing the meta-knowledge about possible working methods with the context of the choice between them : how much time there is, how much effort the person is willing to put in, what they know how to do, etc.

* The choice between working methods allows people to choose their working method so that their mental workload is maintained within acceptable limits.

* Working methods are not stored in a fixed hierarchy, but a hierarchy emerges in practice, as choices about how to meet cognitive needs are made at the time of use.

Access to other papers on this site via Home page

In this review, there are 3 main groups of topics.

1. Introduction.

Basic element, sources of data which meet cognitive needs.

2. The cognitive processing element.

Meeting an information need :

3a. by finding it in the environment.

3b. from a stored knowledge base.

3c. by working through a 'routine',

or referring to the result of using a routine elsewhere.

3d. more on knowledge bases.

Choosing what to do.