3d. Knowledge Bases 2

Topics in this section :

3d.1 . Some implications for mechanisms : knowledge bases used in only one or in several aspects of processing.

3d.2 . More fundamental constituents of a knowledge base.

Sections 3d : Ways of meeting the cognitive needs : Knowledge Bases 2

Building up behavioural complexity from a cognitive processing element.

Lisanne Bainbridge

Two of the many possible more speculative issues about the nature of knowledge bases will be mentioned here :

- the implications for the possible modularity of working methods ('routines'),

- the more fundamental constituents from which a cognitive processing element or a knowledge base element might be constructed.

3d.1. General and specific knowledge, implications for the modularity of 'routines'

For each of the main cognitive needs, and the 'routines' for meeting them, which a process operator uses (see Section 3c above), it is possible to infer what knowledge is required in order to meet this need.

Table 3c.2 lists some of these possible inferences about the knowledge referred to.

The Table below shows the knowledge bases listed in Table 3c.2 as being used by each of the 'routines' identified from the furnace power supply operator’s protocol.

Cognitive need	Knowledge referred to
identify	stimulus : identity. stimulus : hypotheses about identity, with test information. state : response required. operating procedures.
infer/ review present state	symptom : hypotheses about cause, with test information. dynamic models. scenarios.
review/ predict events	dynamic models. scenarios.
predict state	dynamic models.
review/ predict task goals	dynamic models, product targets. plant constraints

review action availability and effects	action : effect. dynamic models.
choose best action	effect required : actions with this effect. criteria for choice. operating procedures.
identify need for enabling action	action : required preconditions.
plan future activities	criteria for optimise/ compromise.

Table 3d.1 : The knowledge referred to in meeting each of the main cognitive needs (see Table 3c.2). Some of them also continue to use knowledge referred to in previous stages of thinking.
(in several sections, colon indicates direct 1:1 link, i.e. a : b = a associated with b)

It could also be possible to make additions to this Table.

Although this table shows inferences based on analyses, rather than being based directly on data, some interesting points emerge. In particular, the knowledge bases appear to fall into two groups, those referred to by several of the main cognitive needs, and those which are referred by the processing ('routine') for only one of the main cognitive needs (Bainbridge, 1992).

Some 'routines' do not refer to a knowledge base, according to Table 3c.2.

The following 'routines’ make use of information now in working storage, obtained from a knowledge base by a previous ’routine’ :

- evaluate state,

- review compensatory events.

The following 'routines’ focus on direct interaction with the environment - again these processes may be directed by information found from a knowledge base in previous thinking:

- make action,

- monitor action effects,

- monitor changes.

Knowledge bases accessed by several of the cognitive needs.

There are several groups of knowledge which are referred to for several purposes, and so need to have a general form rather than being linked directly to the processing which uses them (see Figures 3b.1 and 2), such as :

* dynamic models of the plant,

* scenarios of events,

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

* operating procedures, the formally defined working methods.

These types of knowledge tend to be taught in the more formal and verbal off-the-job training of operators. So the operators may not have any practise in using this knowledge while doing the task.

Knowledge bases used during only one aspect of processing.

The following types of knowledge pairings are each used by only one main cognitive need, i.e. main 'routine'.

need : linked knowledge :

stimulus : identify identity, or hypothesis about identity

symptom meaning underlying cause, or hypothesis

state : response needed response required

effect required actions with this effect

action : result effect

action : pre-conditions required preconditions

In each of these cases, the knowledge is of specific associations (e.g. this particular pattern on the interface means that is happening on the plant), not general principles which could generate this answer.

In many cases an operator can only acquire this type of of associative knowledge by direct experience of operating the process, and using its specific interface.

If this can only be acquired by experience, then this training needs to be done on a full-scope simulator, so the eye and muscle movements used can become automatic/ ballistic/ not needing feedback adjustment (which takes attention, and time).

It is interesting that each of the associative types of knowledge is used by only one of the main cognitive needs. If a particular group of if-then associations are specific to a particular main 'routine', this could support the notion that the main 'routines' are independent modules of processing.

They appear to be independent, in the sense that processing within them is self-sufficient, in at least two ways :

* the working storage they contain is only available to processing within this 'routine' (Section 6).

* they may refer to a knowledge base which is used only by this 'routine'.

The three interdependent networks in a 'routine' (discussed in Section 2 and above),

i.e. the linked elements, the cross references in working storage, and the references to compatible knowledge,

once learned, could give a strong and stable organisation which is difficult to disrupt. This might also provide an explanation for perceptual set.

3d.2. More fundamental constituents

The above points may need some thought in relation to training. This section may have some relevance to theoretical cognitive models, but is rather remote from any practical use of cognitive processing models ! And it's only possible to mention these questions here, as verbal protocol data doesn't provide any evidence which could suggest answers.

Several types of knowledge constituent appear to recur, to be used as basic building blocks in both the processing element and in knowledge representations, e.g. Figure 3b.3. These are constituents such as :

- associations,

- attribute-value descriptors (e.g. the meta-knowledge for the 'routines', see Section 7),

- conditionals,

- part-whole groupings of items (e.g. processing elements linked together to make a 'routine'),

- is-a category groupings of items (e.g. grouping together working methods with similar outcomes).

Both part-whole and is-a category forms of item grouping can occur in several versions, see Bainbridge (1993a).

This could suggest that, however many types of cognitive element there might be, for processing and for knowledge representation, they are all different configurations built up from the same more fundamental constituents. To sort this out properly, it would be necessary to identify all the possible types of part-whole and is-a category relation, and to check whether they could all occur both in processing and in knowledge representation.

If there are these more fundamental constituents, then cognitive processing might be organised in at least four levels, each with different properties :

* the basic constituents of any type of cognitive element,

* the processing elements,

* 'routines' consisting of elements. These 'routines' meet the cognitive needs, and have associated meta-knowledge about their properties, as well as local working storage and reference knowledge (see Section 4).

* 'sequencers' consisting mainly of conditional elements. These sequence the main cognitive needs, and maintain the overview which provides the context for decisions about what to do and how to do it (see Section 5).

While Sections 4-5 suggest that both 'routines' and 'sequencers' might be emergent properties of the cognitive processing element.

There is no direct evidence about these sorts of issues, either from the verbal reports, or from the 'box element' diagrams developed to simulate them.

Summary of main points in Section 3.d

* Some types of associative knowledge may be specific to a particular cognitive need.

* 'Routines' might be independent processing modules, as they contain local working storage, and refer to unique knowledge.

* Knowledge stores might be constructed from a few basic types of element.

* Different types of cognitive element might be different configurations of the same more fundamental constituents, such as attribute-value descriptors, conditionals, part-whole groupings, and is-a category groupings.

* Complex behaviour may be organised in at least four levels, each with different properties : constituents of the elements, elements, 'routines' and 'sequencers'.

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.