8. Learning, modes of processing

Topics in this section :

8.1. What needs to be learned.

8.2. Developing new working methods.

8.2.1. Adapting methods used in similar situations.

8.2.2. Building up a working method from components.

8.3. Modes of processing ('types of skill').

Section 8 : Learning and modes of processing ('types of skill')

Building up behavioural complexity from a cognitive processing element

Lisanne Bainbridge

The mechanisms proposed in this section are by no means adequate to account for all the processes and changes involved in learning, but this section does indicate some of the mechanisms needed, and some aspects of how learning affects the cognitive difficulty of doing a task. Some of the points made in this section are speculative, they have not all been tried out in full, but they indicate how the proposed cognitive processing element might be expanded to have some interesting additional powers.

The points made here are very brief. I have written many comments on learning which are elsewhere on this site, so I don’t think they need to be repeated here.

So for a fuller discussion of some of these topics, with examples, see :

(1978) Forgotten alternatives in skill and workload.

Changes in cognitive processes with the development of skill, and the implications for mental workload.

(1989) Development of skill, reduction of workload.

This may be referred to as the 'Types of Skill' paper.

See also Section B on learning in the final section of the big review of processes underlying human performance.

(1995) Mental workload, learning, errors.

Note that, as usual in my papers, I use the word 'skill' in its general sense as used in psychology and English language, to indicate the progression from beginner to expert, not to label a particular type of task processing. 'Skill' is one of those words which causes a great deal of confusion because it does not translate easily from one language to another, see Bainbridge (1989).

This section is necessarily brief and speculative, as the expert furnace controller whose thinking was analysed in detail was doing a familiar task, so did not need to use problem solving. So what little evidence I have about problem solving comes mainly from observation, and a little knowledge of the literature. Verbal protocols were also collected from a group of inexperienced people learning the furnace power control task (university students), but these were not analysed in much detail. It took me long enough to find a methodology which worked for analysing the data about an experienced operator's thinking. No doubt, making sense of evidence about learning (if possible. . .) would have needed many more technique developments !

This section will discuss three main themes, starting with what needs to be learned for the box element mechanisms described so far in this review to be able to operate. The key part of the element, for this discussion, is the point of choice between alternative working methods on the basis of meta-knowledge, see Section 7. This section also suggests that this choice point plays a major role in the development of new working methods - see Sections 8.1 and 8.2, and in the flexible use of different modes of processing (one aspect of 'how to do it') - see Section 8.3.

8.1. What needs to be learned

There are many types of process involved in learning, and types of process which need to be learned. This paper is specifically concerned with cognitive processes, so will not discuss the mechanisms by which extended practice leads to the learning of associations or perceptual-motor discriminations and integrations.

There is not only the working method(s) and associated knowledge bases to be learned.

Any working method also needs to have stored with it (after learning) all the meta-knowledge involved in deciding whether to use this method :

- what it can deliver - speed, accuracy/ quality, etc,

- what is required to be able to do it - from environment, self,

which are compared with the task to be achieved, to choose the method for doing it.

More working methods may called on for finding these relevant properties of the environment and task if needed.

The cognitive element proposed in this paper has been developed for simulating the behaviour of a highly experienced industrial process operator.

Someone doing a familiar task is using 'familiar cognitive skills’, in which most of the thinking processes and knowledge bases needed for doing the task are already well established. However the nature of the elements does give clues to the cognitive processes which must be developed by a learning process. These learning processes must at least be able to :

* develop the cognitive working methods or 'routines' (see 8.2).

* build up the cross references in working storage.

* organise a knowledge base referred to by a 'routine' so the KB is compatible with the way it is referred to by the related cognitive processing (Sections 3b and 3d).

* learn what main cognitive needs must be met, and how to choose which of these cognitive needs is most appropriate in a given context, as represented in the 'sequencers'.

* learn the meta-knowledge related to each working method, and identify what are the relevant parameters of the meta-knowledge of a working method which need to be considered at any particular point of choosing how to do a task.

8.2. Developing new working methods

A focal issue in cognitive learning is how someone might develop a new cognitive working method, for a situation which they have not got an established method of dealing with. It is possible to suggest that the link between need and method is an important location in learning, as if there is not at least one method for meeting the need then it is necessary for the person doing the task to develop one.

This is a type of problem solving.

A mechanism for developing new working methods needs to be able to do several things. A new working method presumably has to be built up from one or all of :

- components of existing working methods,

- working methods used in analogous circumstances,

- other knowledge about the situation.

Two approaches to developing new working methods will be mentioned, by :

- adapting similar existing working methods, or

- putting together parts of existing knowledge.

In either case, the knowledge referred to may be more or less specific to the particular situation (e.g. Bainbridge, 1988, Figure 5).

8.2.1. Adapting methods used in similar situations

Perhaps the easiest way of dealing with a new situation is to adapt a working method used in a similar situation, that is, to work by analogy or by case-based reasoning. At least two mechanisms would be needed : for recognising similarity, and for adapting the existing method to the new situation. If I was trying to develop a process to account for this learning, I would use the choice-and-meta-knowledge mechanism as the starting point, particularly using meta-knowledge about the outcome of an activity (which was mentioned in Section 7.1 but not discussed).

Recognising similarities

Section 7.1 discussed the choice between alternative working methods for meeting a cognitive need, but did not discuss how the working methods might be grouped together to choose from in the first place. Several mechanisms might be needed in grouping items on the basis of similarity.

One mechanism would be for the meta-knowledge about an activity to include data about the general nature of the outcome which can be achieved by doing it.

A second mechanism would be needed if no meta-knowledge about the outcome is available, so an activity would have to be simulated mentally to find what the outcome would be. This mental simulation would presumably be done in working storage, which extends the notion of what working storage might be able to contain.

The third mechanism would be that working methods could be grouped into classes or categories of working methods which meet the same need, on the basis of them having similar values on several dimensions of meta-knowledge about a method’s outcomes. It is possible to suggest that a related mechanism would be used for choosing analogies or cases which are similar to the situation for which a working method is needed.

Adapting the suggested method

This review does not suggest an actual mechanism for adapting a working method to new circumstances, but again suggests that meta-knowledge about the outcome of behaviour could play a key part.

It could be suggested that a learner who is trying to achieve some outcome chooses a working method which they expect (on the basis of meta-knowledge or mental simulation) will have the desired result. Even people who just press all the buttons on a piece of equipment at random in the hope of getting it to work do not usually start by pressing buttons on equipment A in the hope of getting B to work.

If the learner finds from experience that the behaviour they have chosen does not work, then they adapt the working method and meta-knowledge, using the difference between the actual and expected outcomes as the basis for making these revisions. They then try the proposed behaviour, adapt it if necessary, and so on until they are successful. This process is summarised in Figure 7.2.1. (There is a cognitive version of this figure, and suggested examples of the states of knowledge before and after this learning process, in Bainbridge, 1989, Figures 1, 2 and 3.)

Figure 8.1 : Double use of feedback in improving the present action, and improving the next action of the same type (Bainbridge, 1978, Figure 1).

8.2.2. Building up a working method from components

The second way of developing a new working method is to build it up from components. These components might be parts of other working methods, or principles in the knowledge base. This Section does not suggest a mechanism by which this construction might be done, but does point out some features of the way in which this relates to and extends the element which has been proposed, in this case the cognitive need and working storage parts of the box element.

Note that 'find a working method' is itself a cognitive need. It is possible to suggest two aspects of meeting this need :

* that it is met by a general strategy for building up a working method, which is itself expressed in terms of cognitive needs, i.e. elements.

* that an element can be recursive, in that its working storage can contain elements, so that a working method consisting of elements can be built up in working storage. From observation, this recursion appears to pose heavy loading on cognitive processing capacity.

As an indication of the nature of a general strategy, it is interesting to look at evidence on behaviour in a complex dynamic task in which people do use a general strategy. In the furnace power control task, an experienced operator working with a familiar and relatively simple process had specific working methods for doing his task, described as 'routines'. But some tasks have inherent variability. For example, in the emergency services management task studied by Samurçay and Rogalski (e.g. 1988), managers need to be able to deal with a wide variety of fires, chemical spillages, road accidents, and so on. So they could not develop a standard method of working, or assume that the environment and resources they were working with would always be the same. Samurçay and Rogalski found, from analysing the managers' behaviour, that the managers did have a general method which varied in detail according to the unique conditions at the time it was used. They call this general method the 'method for tactical reasoning' (MTR).

Figure 8.2 : The Method for Tactical Reasoning described in terms of box elements (adapted from Samurçay and Rogalski, 1988, Figure 1).

Figure 8.2 shows the MTR re-described in terms of cognitive processing elements. These main cognitive needs are similar to the main cognitive needs in process operation, as listed in Table 3c.2, and Section 3c.4. However, the main cognitive needs in the MTR are more general, because no two situations in which it is used are the same, so there can be no standard 'routines' and no predefined states or resources.

Although the MTR is described as a sequence of activity, the managers do not follow a set sequence of thinking. The order in which the managers think about the cognitive needs is very variable (as in process operation). Although it has not been studied, I would expect that the managers build up an overview of the state of the situation, and that many cross-references between the different cognitive needs are involved, as the needs are highly interdependent. The outcome of meeting any one cognitive need will affect what is feasible in meeting others, and vice versa (also as in process operation).

However, while this is interesting and suggestive as a general strategy, the mechanisms described in this paper do not appear to account for how this strategy would build up a usable working method in practice. This probably involves several mechanisms. One would be for finding appropriate components in the knowledge base. Finding the items of knowledge to use when developing a working method involves a complex interplay of types of knowledge and degrees of generality or closeness to what is needed in specific circumstances. These are mentioned briefly in Sections 3b and 3d on knowledge bases.

8.3. Modes of processing

This section so far has mainly mentioned two general modes of cognitive processing, either devising a new working method (problem solving) or using an established method (familiar cognitive skill). Learning from experience leads to the development of established methods.

There are also a 3rd sub-type or processing mode, not so much involved in complex tasks except in interacting with the interface : perceptual-motor skills.

These modes of processing could all be called 'modes of expertise' or 'types of skill', because a considerable amount of previous experience and knowledge would be needed for any of them to be done efficiently, and this efficiency is a defining characteristic of 'skill'.

Much behaviour in complex tasks has to start with problem solving to work out what to do. As the situation becomes more familiar, so working methods are developed and parts of the problem solving behaviour drop out, they are no longer needed to do the task.

That leads to Figure 8.3, a summary of the way in which some aspects of cognitive processing can drop out as a task becomes better known. As tasks become better known, some types of processing are no longer needed, so a familiar task poses less mental workload.

This Figure comes from Bainbridge 1989, in which there is extensive discussion of these issues, with examples.

Figure 8.3 : A schema for the main mechanisms underlying complex skill.
Responses : aR = attention, cR = cognition, pR = physical)

Notes :

- perceptual-motor skills ('automated tasks') don’t need any conscious attention, i.e. make no use of working memory. At first, tasks done this way may need problem solving, to work out what might be a good thing to try. But after practice these tasks don’t need problem solving or working memory.

- similarly, familiar cognitive skills don’t need any problem solving about what best to do, though the point at which a task is possible this way may only be reached after much experience with this or similar tasks(s).

Fuller discussion of what underlies this Figure is in Bainbridge 1989, from which this is Figure 6.

This diagram summarises that the different 'modes of processing' or 'types of skill' are not distinct but closely interrelated, and may be combined in a flexible way according to details of the circumstances.

Using an established method has been described as involving a choice point, at which the best method for dealing with the present circumstances is chosen. Learning could affect this choice. If the circumstances in which a cognitive need is met are always the same, then the same method will always be used. So it would become unnecessary to make a choice, the choice and the associated meta-knowledge would become redundant (Bainbridge, 1978). If the circumstances are so regular that the processing is the same every time, eventually the behaviour could become 'automatic', that is, it could be done without conscious awareness, or need for working storage capacity to retain intermediate results. This means that the processing is done with minimum cognitive effort, it is highly efficient. But the cost of this efficiency is that this mode of processing is maladaptive if the environment does change, as then the person will continue to react automatically, which will lead to errors. Leplat (1989) discusses the ways in which practice leads to changes in how a task is done, including this process of 'overlearning'.

It is likely that the point at which a working method is chosen is also the point at which the mode of processing is chosen. If a cognitive need has no associated working methods, then it is necessary to develop a new one (problem solving skills). When there are established working methods, if the situations are so invariant that no choice is needed and the processing is always the same, then the need for choice and working storage disappears (automated skills). So the amount of mental work involved in doing a task can depend on the invariance in the task and the experience of the person. Which of these modes of processing is used depends on the circumstances of the moment. It is not necessarily the case that someone stays consistently within any one mode of processing for any length of time.

Here are some more brief examples (see more in Bainbridge 1989).

The mode of processing does not necessarily depend on the generality or specificity of the behaviour concerned. It is sometimes claimed that 'higher' levels of behaviour organisation involve 'higher' or more difficult modes of processing. For example, problem solving skills might be used at the strategic level, and automated skills at the perceptual-motor level. But this is not always the case. As a practical example, someone's choice of their route to work ('higher' level) may become automatic, but how they choose in more detail to travel along that route - by car or bicycle - ('lower' level) may depend on the weather. Or a problem at a lower level of detail may mean that the way a task is done at a higher level may need to be revised. For example if, during an emergency, process operators move to a special emergency control room which has a different format of displays and controls than in the everyday control room, then they may not be able to use their automated perceptual-motor skills for operating the interface, and may have consciously to solve problems in finding and interpreting the information they need or making the action they want. This problem solving could interfere with their thinking about the emergency. In general, as suggested in Figure 8.3, any of the choice points at any level of the task could be met by any of the three modes of processing, depending on how stable the task situation is and how much prior experience the person has.

It is not necessarily the case that skills always develop in the same way, either that learning always starts with perceptual-motor or automated skills, and leads up to a problem solving mode of processing, or that learning always starts with problem solving and develops until behaviour is automated. Most learning experiences will involve a mixture of processing modes. Which may be influenced by a person’s cognitive talents and life experience.

It is also not necessarily the case that the mode of processing is always related to the sophistication of the processing. For example, recognition-primed decisions (Klein, 1989) are an automated form of processing (a perceptual-motor skill) which is highly sophisticated, it depends on a great deal of knowledge and expertise. Nor is it necessarily obvious what is the mode of processing underlying a particular piece of observed behaviour. Kicking a piece of equipment which is not working properly could be an automatic (emotional) response, or it could be an unsophisticated example of case-based reasoning ("my friend’s television set stabilises when they hit it, so perhaps mine will too"), which might put it in the problem solving category - or a perceptual-motor skill ? (This was written in the long ago days when a TV included many hand-made solder points !)

Which mode of processing is used at a particular moment is not permanently specified in advance, but is something which emerges in the context of a particular moment, depending on details of the current state of the environment and the relevant expertise and spare mental capacity of the person. So this is another reason why any account of complex behaviour needs to include the context, and to have the characteristics of a complexity theory (see Section 1).

This also suggests that it might be unwise to design displays/ interfaces which support only a specific type of processing, if the same task may be done by different types of processing depending on the circumstances.

To greatly simplify, these modes of processing can be grouped within the five general varieties of learning described by Gagné (1977).

1. Gagné’s 'motor skills’ are what I have called perceptual-motor skills.

2. Gagné's 'intellectual skills', which involve using symbols, and range from associations through discriminations and concepts to rules, would include :

a. 'Automatic' processing, which requires no conscious attention or working storage, and is based on (unconscious) knowledge of associations.
b. Using established cognitive processing 'routines'. This type of cognitive skill is the behaviour accounted for in previous sections of this review. This requires both knowledge of working methods and of the knowledge used for reference.

3. Gagné's 'cognitive strategies' are skills for learning, remembering and problem solving, i.e. skills for controlling one's own cognitive processes. This would include developing new working methods. This sort of craft or professional skill depends on knowledge of categories, cases and general principles in the environment, the devices used, and the person's working methods.

4. Gagné’s 'attitudes' are included here as part of meta-knowledge about the personal aspects used in deciding how to do a task.

5. Gagné’s final variety of learning is learning 'verbal information'. I consider understanding language is a good example of a complex task which raises many of the same problems discussed here (see Winograd, 1972).

Gagné suggests that these different aspects of behaviour are best taught by different methods of training. While I agree with this as a general point, they are all so closely integrated in a complex task, I think that (like displays) they should not be taught only as completely distinct - a person also needs practice in using them together and flexibly.

Summary of main points in Section 8

* If there is not a working method available at the need-method choice point, then one has to be developed.

* Working methods might be built up by using a general strategy, itself expressed in terms of cognitive needs, which finds components of working methods and other knowledge which meet these needs, i.e. carry out the required processing.

* Working storage requires to have a form such that working methods can be simulated mentally in it, to find out their relevant properties if these have not been previously experienced.

* If it is possible to build up a working method, or try it out, in working storage, this suggests that the elements are recursive. An element needs to be able to represent another element in its working storage.

* Meta-knowledge about the outcomes of working methods could be the basis for categorising them as meeting similar needs.

* Learners choose a working method which they expect will have the required result, on the basis of meta-knowledge about the outcome. If it does not work, they revise the working method, on the basis of the difference between what they expected and what actually happened.

* If the circumstances/ context of choosing a working method are always the same, then the choice between working methods could disappear.

* If a working method is always carried out in the same way, then it could become automatic, i.e. using no conscious awareness or working storage capacity.

* There may be a minimum of three modes of cognitive processing : automated activity, using an established working method, or devising a new working method.

* Which mode of processing is used would depend on what is available and appropriate at the point, and particular moment in time, of choosing a method to meet a need. These would depend on the variability in the task, and the experience and current spare mental capacity of the person doing it. So which mode of processing is used emerges in the context of a particular moment, rather than being fixed in advance.

Access to other papers on this site via the 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.