3b. Knowledge Bases 1

Topics in this section :

3b.1. Knowledge used in the nuclear incident example.

3b.2. More on knowledge used in process operation.

3b.2.1. Types of knowledge.

3b.2.2. Levels of generality.

Section 3b : Ways of meeting the cognitive needs 2 : Knowledge base I

Building up behavioural complexity from a processing element

Lisanne Bainbridge

The data needed by cognitive processing may be obtained from the person's long-term memory, that is, by transferring control to search in the person's knowledge bases.

What can we say about how someone doing a complex task uses their knowledge of the task, and what does this knowledge need to contain ?

There are two sections on knowledge bases in this review, this one on some key aspects. Then another Section, 3d, after some more features of 'routines' have been mentioned.

This section makes no claim to account for all knowledge base issues, it just illustrates the types of issue which arise in considering them, in 2 main sections :

3b.1 - a specific example of interrelating a processing 'routine' with the knowledge base which it accesses, using the nuclear incident in Tables 2.2 and 2.3 as the example.

3b.2 - more general points about the types of knowledge an operator may have, and how they may be interrelated.

This Section is mainly about the knowledge base as a source of information about the process.

A person’s memory/ stored knowledge also includes information about their own potential behaviour, what actions they can do, what those actions achieve, and what is needed to carry them out. Those types of knowledge and their uses are discussed in later Sections, especially 7 and 8.

Several of the issues mentioned in this Section have been discussed in previous papers.

These are papers which are repeatedly referred to in this section, and expand on the issues mentioned here :

Bainbridge(1988) Types of Representation.

Bainbridge (1989) Development of skill, reduction of workload. 'Types of skill'.

Bainbridge (1992) Mental models in cognitive skill.

Bainbridge (1993b) Types of hierarchy imply types of model.

As much extended discussion of these topics is already available on this site, I didn’t think there was much point in repeating it here.

3b.1. Knowledge used in the nuclear incident example

Figure 3b.1, a fuller version of Figure 2.6 based on Figure 2.5, with the stepped implementation arrows completed, illustrates two of the basic types of direct data acquisition : from the environment (Section 3a), and from a knowledge base.

Figure 3b.1 : Working method for identifying the reason for an alarm signal (see Figure 2.5), with references to knowledge base (from Bainbridge, 1992, Figure 6). For knowledge base used, see Figure 3b.2.

Even in a modest example of operator behaviour, to account for the fault diagnosis behaviour reported in Tables 2.2 & 2.3, and underlying the processes summarised in Figure 2.5, it is too complicated to put everything which might be involved in the behaviour into one diagram. In clarification :

Figure 2.6 describes the behaviour in terms of cognitive needs, and meeting cognitive needs by subsidiary 'routines'.

Figure 2.7 shows the cross referencing links in working storage ('short term memory') in this routine.

Figure 3b.1 shows that cognitive needs are met from the environment or from the knowledge base.

Figure 3b.2 below illustrates what has to be in this knowledge base ('long term memory').

In the nuclear incident, the operators’ knowledge supplied the information that there were 3 possible reasons why a specific alarm had gone off.

In Tables 2.2 and 2.3, after thinking of possible causes, the operators awaited confirming evidence. What to look for, and how, are again suggested by their knowledge of the process. Given the nature of nuclear power stations, there is hopefully no way in which they could get direct information about a leak (rupture) !

In checking which of the three possible explanations of the event was actually the case, the operators used knowledge of two attributes of the possible explanations :

- the likelihood of an explanation, i.e. how frequently it has been the reason,

- other evidence which would confirm or disprove a given explanation.

That the operators have this knowledge is shown by their reported behaviour, and from what they mention in interviews about what might be relevant in this incident.

In Figure 3b.2, this knowledge is represented by a network with different types of nodes : different attributes are represented by different typefaces. The knowledge could equally well be represented by a 'coloured' graph with different types of links, by a table, or other multi-dimensional means. There is no necessary commitment to a particular form of representation here.

Figure 3b.2 : Knowledge base referred to by the working method in Figure 3b.1. This knowledge links underlying causes to evidence for them (from Bainbridge, 1992, Figure 7).

bold underlying cause

italic frequency

underlined confirming evidence

An important general observation is that, when people are doing a cognitive task with which they are experienced, the appropriate knowledge appears to be accessed automatically, without using cognitive processing capacity or conscious awareness of doing cognitive work to search for the data. This suggests that access to the knowledge base, and the organisation of the knowledge in that base, have become mutually compatible through use. This is represented here by having the originating cognitive need (in Figure 3b.1) and the relevant attribute needed (in Figure 3b.2) in the same typeface.

This notion of compatibility between access and storage is implicit in Bainbridge (1989), and explicit in Bainbridge (1992, 1993b).

In Section 3d there is some discussion that some knowledge bases may be specific to particular modules of processing.

This small example from the nuclear incident data only raises simple issues about the representation of knowledge. Some more aspects which need to be considered when representing the knowledge used in complex dynamic tasks are outlined in the next Section.

3b.2. More on the knowledge used in industrial process operation

Knowledge is referred to during cognitive processing, in meeting the cognitive needs by adding to the information available from the environment. So what forms might this knowledge take, and how might this knowledge be represented, organised and processed ?

This section will outline some of the issues to be accounted for, and raise some possibilities, rather than giving a complete or detailed account of how knowledge is represented (as the problems of describing knowledge are many and vast). In particular there are the interesting questions (which will not be answered !) : if the cognitive processing element proposed in my papers can account for (aspects of) complex task processing, could other types of element account for knowledge representation ? and, if so, how many different types of element are there and what might they need to be like - what mechanisms would they need to include/ account for ? (more on this in 3d.)

So the examples in this section are intended as indicators of how complex the issues are, not as serious suggestions of an answer.

I have found it useful (Bainbridge, 1988) to distinguish between the operators' knowledge of their own activities, in particular their working methods as described in this review (see Section 7-8), and their knowledge of the environment and devices they are working with (this Section). However, it is possible to question this distinction between knowledge of physical entities and knowledge of working methods. Evidently, in the brain something about cognitive processing having been done is conserved in some way, so that later processing on the same topic can be more effective. But it is not clear whether what is conserved is in the form of processing instructions or is a passive store of the results of processing. Bartlett (1954) and Neisser (1967) argue that remembering (accessing information experienced in the past) is a re-constructive process, rather than simply a matter of accessing stored data.

This Section 3b will focus on the types of knowledge which a process operator might have, and which therefore need to be represented.

In Section 3d there are some (incomplete) suggestions for the nature of the underlying mechanisms which might be required.

The knowledge, about the environment and devices they are working with, which is used by process operators is a complex interplay of kinds of knowledge and levels of detail.

3b.2.1. Types of Knowledge about a process

A process operator does need to know at least about :

- the physical structure of the environment and devices, their positions, functions, and causal relations,

- the dynamic behaviour of the plant/ device over time, and how to influence this behaviour.

(Bainbridge, 1988, 1993b).

It may be easier in the abstract, when discussing these types of knowledge and representing them visually, to use a different type of diagram for each type of knowledge (examples in Bainbridge, 1988).

But actually the process operators' mental representation must include the links between types of knowledge, e.g. how physical positions affect causal relations which affect behaviour (given of course that the operator actually knows those links).

Figure 3b.3 : Some of the links in the knowledge structure about a small part of a complex process (thick line = cause-effect relation). (This is a version of knowledge about flying, adapted from a process control example in Bainbridge, 1991, Figure 3.)

Figure 3b.3 gives a small example of possible links between knowledge about :

- physical structure (left column),

- cause-effect relations (mid column),

- the person's working method (right column).

Note it has only been possible to simplify this diagram into representing the 3 types of knowledge as a linear hierarchy of part-of relations, because each column is a tiny extract from a huge network of related information of the same type.

And this is a speculative example, rather than based in some way on data about an operator’s knowledge.

This diagram only represents simple issues. In complex dynamic tasks, operators may have 'mental models' of how the process behaviour changes over time in response to various inputs, and of multi-dimensional states. The furnace power control task was dynamically very simple, and I did not obtain any evidence about these types of knowledge.

In the area of complex tasks, 'mental model' is the label used for an operator’s knowledge of the plant dynamics, which they can use in understanding what the process is doing and predicting how it will behave in the future.

Their 'mental picture' is equivalent to the overview discussed in this review, the operator’s construct of information about the current process state. See more on this in Section 6a.

These figures are examples of types of dynamic knowledge a process operator might have (there are also others) :

Figure 3b.4 : possible components of a mental model of dynamic changes in over time in a multidimensional process, Bainbridge, 1992 , Figure 2

Figure 3b.4 is a possible attempt to describe the causal processes underlying changes in the state of a process.

While Figure 3b.5 describes the sequence in time of phases in the process, and what specifically causes the transition from one state to another.

Figure 3b.5 : possible knowledge about what affects a sequence of phases in a process, Bainbridge, 1992, Figure 3.

Again these are possible diagrams of what needs to be available in a dynamic knowledge base, not based on data from a specific process. See more on 'mental models' in Bainbridge (1992).

3b.2.2. Degrees of generality.

The previous comments, see Figures 3b.1 and 3b.2, suggested that the reference knowledge used by an experienced person may be specifically organised to be compatible with the processing which accesses it.

On the other hand, Section 8 on learning below (and Bainbridge 1989) suggests that when an inexperienced person is developing a new working method, as they seek relevant knowledge to use in their task, they may search through knowledge which is further and further removed from closeness to the particular problem.

For an industrial process operator it is possible to suggest several degrees of distance of knowledge from their specific process (Bainbridge, 1993b) :

1. Knowledge specifically referred to by an established working method (example in Figure 3b.2). More discussion of this in Section 3d.

2. This direct reference knowledge may be underpinned by plant specific explanatory knowledge, such as the physical, causal and functional structure of the specific plant, mental models of plant behaviour in particular states, general sequences of events, or specific case stories.

3. This plant specific explanatory knowledge could be supported by industry specific knowledge. This could be engineering knowledge, for example what type of plant design is needed to make a particular chemical. Or comparisons between the properties of the same mechanism made by different manufacturers. Or it could be case stories. For example, operators in many chemical and nuclear plants exchange stories about unusual events in other plant (and these stories may include words, or static or moving mages, or emotions). These become part of their knowledge about how to deal with situations they have not experienced themselves.

4. This industry specific knowledge could be supported by general engineering explanatory knowledge, such as mental models of various categories of physical, functional and causal structures (e.g. general types of pumps, heat exchangers, cooling systems), and their behaviour or general scenarios.

5. This general engineering knowledge could be supported by general physical and chemical explanatory knowledge, such as about nuclear reactions, physical laws or chemical reactions.

Items grouped together in this list are not necessarily all at the same distance from relevance to operating a plant. For example, some aspects of theoretical physics or chemistry may explain how the plant fundamentally works but may not be much use in operating it. On the other hand, knowing the law relating pressure and volume might be directly relevant when thinking about an operating problem.

More on issues about flexible/non-constant 'levels' of knowledge in Bainbridge, 1993b.

Evidently knowledge bases can be huge, and contain many different types of information, at many levels of detail, and with many different types of link between items in them. Whatever an individual’s knowledge base contains, depending on their nature and experience, it must be complicated.

- there are many different aspects of knowledge, such as physical form, function, cause-effect relations, etc.

- and there may be many different levels of generality, from specific instances to general principles.

- knowledge may be in different sensory forms : sights, sounds, smells, or 'imageless'.

- at many levels of detail.

- and there can be different types of links between them (see Bainbridge 1993b).

3b.2.3. Wider types of knowledge

One person may also have many different knowledge bases, which may have or may not have links between them (e.g. I have large knowledge bases about both sewing and cognitive processes in complex tasks. Isn’t the human brain amazing - I still seem to be able think about that knowledge while re-reading these papers, despite 30 years of complete neglect. And we can do all these tasks which we have little success yet in describing the mechanisms or results of.)

- there are many different formats of sensory knowledge : visual images, words, sounds, tastes, etc. (I’m good at recognising voices, but am amazed by people who can recognise what a specific piece of music is from hearing a few bars. As for the skills and knowledge involved in playing a musical instrument, composing music, or cooking at Michelin star level. . . When one gets into trying to represent these types of knowledge, the account gets even more complex and distant from the heart of what is being represented than it does for industrial processes.)

- within each of the senses there may be many different sub-types : different types of visual diagram (see Bainbridge 1988), different words with related meaning within one language and between languages, different genres of music, different types and cultures of food, etc.

- there is also 'imageless thought', not based on records of sensory images. When I was a psychology student in the late 50s, there was much discussion about whether 'imageless thought' was superior to image-based thought, and it was claimed that men do imageless-thought more than women !

So what simplifications might there be in all this ?

One might be that there are similar types of links within all types of knowledge : is-a, part-of, cause effect, occur together, etc.

Another is that some knowledge is specific to particular tasks.

So what evidence might there be on all this in a complex task such as process operation ?

Are there any simplifications of the knowledge bases used in complex tasks ? see more in Section 3d.

Summary of main points in Section 3.b

* Process operators need to know about many different aspects of the environment, for example physical, causal, dynamic, and the links between these different aspects.

* For an experienced person, after learning, the way of accessing reference knowledge, and the structure of that knowledge are mutually compatible.

* The knowledge used in devising new working methods, or for explanation of events, may be at several degrees of distance from direct relevance in operating the plant.

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.