9. Final comments

Section 9. Final comments

Building up behavioural complexity from a cognitive processing element

Lisanne Bainbridge

Justifying the claim that complex behaviour can be accounted for by a simple processing element does seem to require a lot of explaining ! In fact, most of the length of this review has been due to the need to explain the evidence which justifies the use and effectiveness of the box processing element account. The element itself is quite simple, it is the behaviour it has the potential for generating which is complex.

This review is an extended discussion of the box element mechanism, and what it might be extended to account for.

There are briefer introductions to this mechanism in :

(1975) The representation of working storage and its use in the organisation of behaviour.

(1992) Mental models in cognitive skill : the case of industrial process operation.

and more on the types of cognitive processing to be accounted for, and how processing changes with learning in :

(1989) Development of skill, reduction of workload (the 'types of skill' paper).

The focus in this review has been on accounting for the organisation of behaviour. The claim is that complex behaviour emerges from the use of structures ('routines' and 'sequencers' built up from the cognitive processing element), as a function of the external and internal environment (context) at a particular moment. Knowledge of the context is built up by previous cognitive behaviour, and then influences the choices about what cognitive task to do next, and how to do it. So flexible sequences of behaviour which are adapted to details of the circumstances are generated.

The key cognitive aspects of this account are:

* behaviour is organised in terms of cognitive needs and associated working storage, which are parts of cognitive processing elements,

* behaviour is active and need oriented ('top down'), rather than occurring only in reaction to environmental events ('bottom up'),

* the cognitive processing elements are built up into structures which carry out processing, and decide what best to do next and how to do it,

* cognitive processing builds up an overview of the situation, which is structured by the cognitive needs, and acts as the context for the choice of later behaviour,

* there may be two types of working storage:

i. the overview, which is built up by processing, and then is available as a reference context for later processing,

ii. the temporary storage of items needed during a particular part of processing, which are not retained for later reference.

* in situations in which there is no (not yet a) context, behaviour must follow a sequence of stages. This serial behaviour emerges from the contextual mechanism in particular situations, rather than being formally defined in advance. This is represented in the box elements by the operation of two types of two-arrow box, see Section 3c.3.4.

* the choice of how best to meet a cognitive need involves meta-knowledge about the possible behaviours.

* this meta-knowledge choice is also involved in the control of mental workload and learning.

This review describes how the proposed cognitive mechanism arose originally from an analysis of actual behaviour in an industrial process operation task. The data on real performance in a wider range of tasks suggest that several additional mechanisms are needed to account for complex human behaviour in many cognitive tasks. These extra mechanisms are needed to cover such points as :

* Working storage needs to be able to represent not only single independent variables, but also :

i. relations between variables,

ii. multidimensional constructs. The current representation includes only discrete variables. It only includes very simple ways of describing relations between them, or perceptual constructs, not Gestalt perceptions in which the sum is greater than its parts. So it omits a major way in which the brain works. We are very short in general of ways of understanding how this aspect of the brain’s functions can be described, so it tends to be ignored. Which can be a big failing.

* Active search for the information needed, which is organised by the 'routines' and 'sequencers', may be overridden by:

i. a highly salient event in the environment,

ii. some aspect of the environment which is relevant to a cognitive need other than the one currently being thought about.

But processing can return to a previous point in thinking, after an interrupt.

* Representation of knowledge.

Sections 3b and 3d, and 7 and 8 of this review explore extensions to this processing element, in particular two types of knowledge :

i. meta-knowledge about proposed working methods, which is used in the choice of how to do a task, and is therefore a focal mechanism in the control of mental workload and of learning,

ii. the knowledge base which is referred to by processing. In dynamic tasks this can involve a very complex network of types, and levels of detail or immediateness of relevance, of knowledge and processing.

* Behaviour may appear as serial but not be stored in a serial form :

i. In tasks in which it is not possible to build up a contextual overview, or an overview has not yet been built up, behaviour may follow a standard sequence. This standardised behaviour also emerges from the operation of the contextual mechanism in the 'sequencers' in oft-repeated circumstances.

ii. 'Routines' may be carried out as serial behaviour (because it often isn’t possible to do two things at once), but have an underlying parallel/ simultaneous representation (see Section 3c.3).

iii. A cognitive need is met by meeting other cognitive needs, i.e. overt behaviour has a hierarchical nature. However, the different cognitive needs are not necessarily stored in a fixed hierarchy of levels (Section 7.2).

See also list below of other aspects of behaviour which have not been explained by the box element and the constructs that can be made from it.

This review has suggested that this approach to behaviour modelling is an example of a complexity theory, as several features of complex behaviour are emergent properties of the activity of simple elements, rather than being explicit in the underlying representation from which the behaviour is generated. This complexity appears in the 'sequencers', in the 'routines', and between the levels of behaviour organisation.

Traditionally, in simulation of human cognitive processes, it has been considered a weakness if a representation has a large number of degrees of freedom, as they make it difficult to test whether a simulation does match some particular behavioural data. However, in a complexity theory, degrees of freedom are seen as a strength, as they mean that a large number of different behaviours can emerge from a simple substrate. This could also have an advantage in accounting for why the nature and probability of human behaviour are so difficult to predict. Attempting to predict human behaviour is like weather forecasting (though very much more complex) : it is not possible to be right, but it is possible to be useful.

It may not be possible to test or prove the validity of this 'cognitive element' approach by the conventional method of generating testable hypotheses. However, this approach works as an account of complex behaviour, as far as can be tested using pencil-and-paper simulation. And it has the strengths of parsimony, internal consistency, and the ability to provide an account of aspects of behaviour which it was not originally devised for. The approach might be tested indirectly, as it could be the basis for making testable suggestions about what would be effective designs for supporting human behaviour in complex dynamic tasks (see e.g. Bainbridge, 1983, 1991, 1993b, 1993c).

The mechanism described here is mainly oriented to accounting for 'familiar cognitive skills', as the verbal report analysed and represented in the most detail came from an experienced melting shop manager.

The mechanism proposed in this paper has powerful potential for understanding several crucial aspects of complex behaviour, but not all of them ! It is not at all possible to claim that this is complete as an account of cognitive processes. It does not, for example, explain :

* pattern handling,

* perceptual and motor skills such as recognition-primed decisions,

* some inference processes,

* multi-tasking and planning,

* problem solving,

* how learning occurs,

* the elements which would be needed as the basis for building up knowledge structures - it doesn’t for example say much about the nature/structure of knowledge about the external world which these processes access in thinking about the behaviour of externals.

It will certainly be interesting to see whether and how much these and other cognitive processes could be accounted for by extensions to the mechanisms proposed in this paper. Though no doubt, as the history of science shows about many theories in many fields, something completely different will be needed if we are going to get a more valid and complete account of our own behaviour.

Good Luck to those who attempt to develop it 👍

Access to other papers on this site via the Home page.

These are the papers on the main site which are referred to in this review as containing more discussion of these issues.

These are the main theoretical papers on the main site, the other papers are mostly on the implications for practical applications.

On techniques for analysing verbal protocols :

(1985) Inferring from verbal reports to cognitive processes.

The key results from that analysis :

(1974) Analysis of verbal protocols from a process control task.

'Types of skill' in cognitive processes :

(1989) Development of skill, reduction of workload.

Basic concepts used in modelling the observed cognitive behaviour :

(1975) The representation of working storage and its use in the organisation of behaviour.