Models for the Analysis of Work Competence: A Critical Review
School of Education
University of Manchester, UK
Cognition et Activitees Finalisees,
CNRS Universite de Paris 8, France
Paper Presented at the European Conference on Educational Research, Lahti, Finland 22 - 25 September 1999
In many workplaces, technological and organisational change are creating needs for new kinds of knowledge, skill and competence, and this is turn is creating new pressures on VET. This paper focuses on the assumptions of existing research methods for conceptualising and analysing knowledge and skill, especially in complex, dynamic work environments.
Research into the introduction of automation has drawn attention to the broadening of work roles. This development arises from the capacity of machines to take over the application of tools to materials, leaving the operators with the broader role of planning the production run, monitoring the machines, intervening where necessary, and assessing the quality of the output. The amenability of computer controlled production processes to integration into large-scale systems also broadens the operator's role, in that he or she now has to liaise with co-workers in other parts of the organisation.
Moreover, increased competition has created the recognition that to succeed, organisations must become "learning organisations". In a global market, they need to renew their product ranges more quickly, customise more and reduce the time between the design of a product and its delivery to the market. Thus the organisation needs to learn collectively at all levels. Instead of the earlier situation, when a worker could learn a task and perform it unchanged for a long time, there is now an ongoing process of setting up new production teams to learn how to make new products. The competitive pressures make obsolete a traditional work organisation based on a division between "the office" (where the planning is done) and "the works" (where direct work activity is carried out), mediated by "foremen" who translate office-made plans into procedures for operatives, and then check that they are performing them properly. Now, many enterprises have integrated planning with production in partially-self-managing teams, and have developed a flatter structure in which managers are supporters of the teams rather than controllers. There is more dialogue than before between the staff traditionally responsible for planning and decision making and the operatives responsible for direct work activities.
Hirschhorn and Mokray (1992) carried out in-depth interviews and used grounded theory to identify the nature of competence in a company making microprocessors. They identified two facets of competence: skills (specific pieces of work-related information and know-how) and roles (the work ethic, empowerment, responsibilities.) Changes in skill requirements were due to changes in tools and problems. For example, the introduction of an automatic sheet metal press involved a transition from a single tool to a system of tools, from sensory feedback when the tool hits the material to cognitive feedback derived from inspecting the print-out showing the results of the machining, from a narrow focus on a manual task to a wider focus on a broad range of activities. Problems became more abstract, or the level of abstraction changed. Even though there was more information, many of the employees emphasised that they needed to rely more on memory. To plan ahead, to monitor and inspect, they had to consciously picture the process in the abstract and remember the whole nature of the run.
The significant finding reported by Hirschhorn and Mokray (1992) was that competence could not be identified with the new skills that were required. Rather, competence was an interaction between these skills and the workers' roles within the organisation. Roles were shaped not only by the plant's social system, but also by individual factors such as the choices the individuals made about how to direct their own work (e.g. to take initiatives or to work by the book.) The essence of this interdependence can be grasped by considering Hirschhorn and Mokray's (1992) comment that, if someone is highly skilled, but their role gives no authority to exercise the skill, they will not be able to act in a competent manner. Hirschhorn and Mokray (1992) asked employees when they felt competent. Employees responsible for direct work felt competent when the wider system of work and organisation supported rather than disrupted their work activities. Employees responsible for indirect work felt competent when they could act on this structure, helping to shape it so that it facilitated the autonomous work of the former.
Many workplaces are undergoing changes similar to those described by Hirschhorn and Mokray (1992). The introduction of new technology both informates work and instruments cognition, and we need to find ways of capturing the complexity of this relationship. A clue is provided by Hirschhorn and Mokray's (1992) observation that competence is the result of an interaction between the new skills required by the technological development and the workers' role in the workplace. This is significant because it proposes that both individual and social factors should be included in the definition of competence. We now turn to consider how adequately existing theories make this definition.
Information-Processing Approaches to Competence
There now exists an extensive literature describing competence in knowledge-based work. For several decades, the dominant paradigm (in cognitive approaches in this field) has been the information-processing model. This is based on an analogy between the mind and the computer. Thus the way the supervisor of an automatic production process operates is explained in terms of a mind or cognitive system "between the ears" which serves as a general purpose symbol-processor. The information-processing model assumes a representational theory of mind, in which the contents of the mind mirror or reflect objects and events in the outside world. An example is Orchanine's theory of operative images: supervisors of automatic processes develop internal models of the plant which are not photographic likenesses so much as representations of the operator's own function, reproducing aspects of reality important for that function but omitting or distorting aspects that are irrelevant to it. Intellectual work is essentially an internal process, operating on the representations to transform them into new representations which would (for example) represent new states of affairs in the outside world. This is a dualist model: the representations drive action, just as a computer programme in a robot drives the actions of the robot arm. Most research within this paradigm has concentrated on the problem of defining the internal structure of the knowledge which was supposed to drive the action; knowledge-elicitation techniques were developed for this. One hypothesis was that knowledge was structured internally as semantic networks, whose nodes were linked by different relationships. Studies of knowledge in work situations (as opposed to knowledge of artificial laboratory tasks) stressed the complexity of these networks.
Consider, for example, the work of Samurçay and Hoc (1992; 1996) who investigated the competency of blast furnace (BF) operators. They show that, in this kind of supervision where the process has a very long latency, the control activity does not consist of adjusting inputs to outputs by information feedback and product quality is not a suitable criterion to guide supervision. From the operator's viewpoint, the task consists above all of meeting conditions for a "good functioning" of the tool (the process). This "good functioning" is interpreted in terms of the equilibrium of the system, decomposable into subsystems corresponding to the seven "descriptors" or pragmatic concepts (built by operators in action and for action). BF operators have been shown to use these concepts to understand the current operation and make intervention decisions. It has been shown that these schematic representations, which correspond to non-measurable variables, act as intermediary entities in expressing causal relationships between observable parameters in the plant. These variables, which were named "descriptors", refer to phenomena of various kinds inside the BF which are not directly observable. Their values are inferred from observable parameters. These representations took the form of descriptors of seven ongoing (mechanical, thermodynamical, chemical, etc.) phenomena inside this opaque plant, only accessible by the means of peripheral parameters (indicators). The plant is "opaque" in that the transformations are not directly observable, but must be inferred from instruments, which display parameter values at sites remote from where they are sensed. Comparisons between experts and novices would demonstrate that the former had more complex networks than the latter - this was assumed to explain why experts could deal with situations that novices could not. Moreover, the networks contained representations built out of experience of real systems - learning in the workplace. Thus the knowledge was adapted to the actual demands of work.
Production systems provided an alternative hypothetical representation of the mind (Newell and Simon, 1972). The mind was supposed to store a system made up of a complex set of "IF (condition) THEN (action)" rules called productions. Information was taken in from the outside world and stored in memory. It would trigger the IF part of a rule. The action specified by the rule would follow, usually triggering the condition for another production. The line of dominoes would continue falling until eventually one of the actions would begin an effector process, which would result in speech or physical action. Boreham (1989) used a production system to model the decision making process of doctors prescribing drugs. The validity of the model was tested by the technique of "cognitive simulation." The advantage of production systems was supposed to be that they were computable models, i.e. could run as a program. Thus the production system could be run, and compared with the performance of human subjects operating on the same problem. If the computer programme produced the same outputs from the same inputs as a human subject, it was taken as a good theory of how the latter was performing his or her intellectual work. Productions are attractive to researchers because they are supposed to make implicit processes explicit, and are teachable. However, many attempts to model work such as medical decision making in terms of productions failed to provide adequate explanations of cognitive skill. For example, in the Boreham (1989) study, certain aspects of the judgement process were not captured by the production system. Whether this was due to weakness in the theoretical model or failure to do the modelling adequately, remains a point for debate.
Connectionist models - otherwise known as neural nets, or parallel-distributed processing - are a third type of model within information processing theory. They attempt to model the physiological structure of the central nervous system directly. (It still counts as a computational model of cognition, as it followed the introduction of computers built on connectionist principles by Siemens, and computer programmes written on connectionist principles such as data-mining programmes.) Connectionist nets consist of neurone-like nodes which are connected together in such a way that a single unit has many links to other units. Units affect other units by exciting or inhibiting them, and a concept representing some outside event or object can be stored in a pattern of activation distributed throughout the whole network. Thus in place of the monolithic, unbending rules represented by productions, cognitive processes were modelled by the more fuzzy concept of patterns of activation that link certain ways of stimulating the network with certain types of output from it. A rudimentary kind of learning has been demonstrated by "training" neural nets to produce outputs from inputs - for example, to produce speech from text (Sejnowsky and Rosenburg, 1987). In this example, it may be said that the network has learned the rules of pronunciation. This model has two main contributions to cognitive ergonomics. It depicts the implicit nature of knowledge, i.e. representation without symbolisation. Second, the complexity of reasoning is based not on the complexity of the nodes (unitary concepts), but on the complexity of the relationships between them - patterns of activation and inhibition of simple units. However, connectionist models fail to take into account the context and meaning of the situation for the subject.
Despite capturing some aspects of intellectual work, many researchers working in the field find information processing models of cognition disappointing. While the models might explain performance in narrow and static tasks, today the challenge is to understand the competencies required in work situations characterised by flexible teams, organisational structures in which once-clear boundaries have been blurred, ever-changing roles and responsibilities, new technology and the constant need to learn new skills. The main weaknesses of the information processing paradigm are its inability to capture these dynamics of transformed work situations.
Towards a Contextualised Model of Competence
Recent years have witnessed the growth of alternative theoretical models which place these aspects of the modern work situation closer to the centre of the frame. The situated cognition movement emerged in the early 1990s, with the publication of a series of books expounding the anthropology of work and everyday life. Suchman (1987) addressed concerns in the design of office machinery, such as how to give directions for the use of photocopiers, and presented a theory of human action in which the cognitive activity of planning is not a distinct internal act that drives an external instrument, but is embedded in the action (and the instrument) itself. Developing the argument that cognition is embedded in social activity, Lave and Wenger (1991) defined learning as increasing levels of participation in a community of practice. The fundamental assumptions was that knowledge was to be found in the customary activities of that community. In general, situated cognition:
rejects the fundamental mind-body dualism of the information processing model, i.e. that knowledge and skill are plans stored in memory which drive action in the physical world;
as just indicated, regards knowledge and skill as distributed through the social relations which compose the culture to which individuals belong;
regards apprenticeship as the central paradigm for learning - not necessarily any particular form of apprenticeship, but the gradual process by which a neophyte is initiated into a community of practice;
explains the transmission of knowledge and skill as the progressive modification of a social organisation by new uses of tools and technology (including cognitive tools such as symbol systems.)
Many of the fundamental ideas underpinning this model have been derived from the Russian socio-cultural theorists Vygotsky and Leontiev. However, as it stands the situated cognition paradigm is a blunt tool for analysing competencies in particular workplaces. Its particular weakness is its claim that the major part of competencies are constructed by people spontaneously in action situations. But each encountered situation does not produce spontaneously structured knowledge; conditions such as work organisation or specific work conditions are necessary for this structuring. These are relevant indicators and cues that people retain from the situation in relation to future efficient action. Increasing experience with some categories of situations creates both routines and automatised procedures, and schematic representation. Conversely, enrichment of the conceptual knowledge and problem solving competences increases the possibility of benefiting from the experiences. The resources of episodic memory can play a crucial role in the development of individual and collective competences. Baerentsen (in press) has shown that, in the context of system control, episodic memories can function as a means for the retrieval of relevant information. He stresses also that they may also serve as a "virtual world" for joint experimental problem solving, and as a general means for the collective development and maintenance of the professional qualification of the operators.
Experience provides the circumstantial and episodic representations people acquire by encountering problem situations. It should not be confused with seniority in the profession or with the time spent in a given work position. Two dimensions are important: the density and diversity of the problems encountered; and the individuals ability to retain something significant from this experience. As postulated in Piagetian theory, to act and to understand are two different processes, so we propose that very often, the construction of experience requires that the individual engage in a specific activity oriented toward the analysis and the understanding of a situation. The conditions for engaging subjects in this constructive activity still remain an open question, although studies are beginning to appear on this issue (Hukki & Norros, in press; Pastré & Samurçay, 1998).
Boreham, N.C. (1989) "Modelling medical decision making under uncertainty", British Journal of Educational Psychology, 59, pp. 187-199.
Edwards, D. (1991) Categories are for talking. Theory & Psychology, 1, 515-542.
Heritage, J. (1984) Garfinkel and Ethnomethodology. Cambridge, UK: Polity Press.
Hirschhorn, L. and Mokray, J. Automation and Competency Requirements in Manufacturing. In Paul Adler (ed) Technology and the Future of Work. New York OUP 1992
Hoc, J-M., Samurçay, R. (1992) An ergonomic approach to knowledge representation. Reliability engineering and System Safety, 36, pp. 217-230.
Hollnagel, E., Hoc, J.M. and Cacciabue, C.P. (1995) Expertise and technology: "I have a feeling we are not in Kansas anymore", In Expertise and Technology : Cognition and Human-Computer Cooperation, Hillsdale: LEA.
Lave, J. and Wenger, E. (1991) Situated Learning: Legitimate Peripheral Participation. Cambridge, UK: Cambridge University Press.
Nardi, B.A.(Ed.) (1996) Context and Consciousness : Activity theory and Human Computer Interaction, Cambridge, Massachusetts: MIT Press.
Newell, A. and Simon, H.A. (1972) Human Problem Solving. Englewood Cliffs, NJ: Prentice Hall.
Peirce, C.S. (1931) The Collected Papers of C.S. Peirce. Cambridge, MA: The Cambridge Press.
Rosch, E. (1978) Principles of categorisation. In E. Rosch, B. Lloyd (Eds.) Cognition and Categorization. Hillsdale, NJ: Erlbaum.
Sandberg, J (1994) Human competence at work: an interpretative approach, Göteborg University Press, Sweden.
Samurçay, R. and Hoc, J.M. (1996) Causal versus topographical support for diagnosis in a dynamic situation, Le Travail Humain, 59, 45-68.
Sejnowsky, T.J. and Rosenburg, C.R. (1987) Parallel networks that learn to pronounce English texts. Complex Systems, 1, 145-168.
Suchman, L.A. (1987) Plans and Situated Actions. The problem of human-machine communication. Cambridge: CUP.
Taylor, (1981) A categorization approach to stereotyping. In DL Hamilton (Ed.) Cognitive Processes in Stereotyping and Intergroup Behaviour. Hillsdale, NJ: Erlbaum.
This document was added to the Education-line database 21 September 1999