Tony Lambie
t.lambie@ucl.ac.uk
Effective Work and the Autonomy of Design
The Ergonomics Unit was John Long’s creation, and the seminal ideas of the Unit are best summed up, I think, as ‘effective work and autonomy of design’. John was an outstanding teacher equipped with these burgeoning ideas. They were contentious ideas too – the best kind; at first blush, simple ideas but with hidden dangers. We knew they were burgeoning because everyone who came to the ‘Unit’ took them on. Not only were they challenging and fruitful, but they imposed discipline, with the added value of John’s supervision; excellent ideas upon, and around which, to build PhDs.
My thesis was on the status of these ideas as elements of a distinct engineering discipline, and how that discipline could be entirely autonomous of science and yet not come adrift from it epistemologically. But I was also interested in how John Dowell (another PhD student) along with John Long, had formalised the ‘Conception’ of those two ideas, among others: an engineering ontology for HCI. Projects I have been involved in have required effective collaboration and I have wondered how that ontology could be augmented to accommodate multiple agents at the interface.
The following is a brief exposition of an attempt to adapt those ideas to that end. It is a revised theoretical framework, part of a consultancy project on which I worked with John Long, but whose agreement with it should not be assumed, though I use the first person plural pronoun. However, I hope he might be sympathetic to the general direction of my views, and their good intentions.
1. Views of Coordination and Work
1.1 Introduction
Computer Supported Cooperative Work (CSCW) has been dominated by views which have an anthropological or sociological origin. These views are appealing because they address a less constrained, but arguably more realistic, set of problems than those traditionally tackled by human factors specialists. Socially-oriented practitioners choose to approach the complex process of group behaviour in this way because they do not consider the paradigm of a model or framework appropriate for such a protean phenomenon. It is this reluctance, we believe, which is largely responsible for the difficulty faced when attempting to integrate conventional systems development work with an understanding of the part played by the joint activity of those carrying out the work.
The challenge is to promote the idea of a framework or model to aid design reasoning, without imposing constraints: in short, to manage the design problem of cooperative work without practising reductionism; and at the same time, to maintain a consistent and systematic overall view so that a common basis might be established between the different stages from the understanding of the problem to the setting out of a solution. This common framework or model might support a language of representation for use by diverse designers to enhance mutual understanding throughout the different stages of the problem formulation and solution.
2.1 Some Frameworks or Models
Schmidt’s conceptual framework of cooperative work [1] is a comprehensive review of the aspects of collective behaviour at work, and covers its gross and obvious as well as its fine and subtle features, and Schmidt, indeed, has something interesting to say about the core notion of cooperation which we believe is associated with the idea of teamwork. However, just as we shall not address classes of coordination, such as ‘augmentative’, ‘integrative’ etc., nor the work organisation nor the modes of cooperation (all dealt with in Schmidt’s paper), except incidentally, so we are not concerned with whether the cooperation is between experts and experts, or experts and non-experts, as Falzon’s [2] paper is. Likewise, whether design is not contingently a cooperative activity but essentially and necessarily a collective one, as Bødker [3] asserts (“Design is a collective activity….”), need not detain us at this point.
All these writers recognise collectivity or potential collectivity of work, but none offers a framework which underpins the workers’ interaction, nor do they try to forge a connection between the systematic (and established) activity of the software engineer and this multiple and mutual interaction. Falzon, for example, deals with it as an instance of dialogue governed by the general principles of dialogue, and it is of course useful for that. The mechanics of the dialogue of any cooperative activity must be understood and their representation is going to be a key component in any specification or diagnosis of group work effectiveness. However, as Brehmer [4] points out, “cooperation can be achieved without communication”. That is to say, we can look below explicit dialogues for the principles of cooperative behaviour, and, arguably, we can go further as Schmidt [1] does when he writes, “conceptualisation is in principle a coordination of viewpoints”: i.e., that individual and collaborative cognitive behaviour is fundamentally inseparable in group work. It is in this region that we shall look for a ‘device’ which will help us to reason about cooperative work design problems.
Storrs [5] clearly deals with some sort of group work scenario but his paper constitutes, as he himself describes it, an ontology and, thus, is more descriptive and classificatory than ours, which will also attempt to account for, or provide a rationale for, the multiparty interaction – in the interests of the systematic specification of teamwork design, i.e., it comprises a prescriptive element.
Others, such as Rasmussen [6] and Vicente [7], who aspire to some form of engineering, deal with group interactions to carry out work, and provide frameworks of a kind. However, our conceptualisation (not to be confused with Schmidt’s term [1]) is different from theirs, and will extend it by re-interpreting the role of agents vis à vis one another and the complex of tasks which make up the work.
Our conceptualisation of cognitive coordination is an addition to, or an elaboration of, a conception of design peculiar to the Ergonomics & HCI Unit at University College London (see Dowell & Long [8] & [9]) – a version of cognitive engineering. So we start out with an account of the background against which the conceptualisation has taken place. We shall not provide a separate account of the Conception of a Cognitive Engineering discipline [8] & [9], but will introduce the features relevant to the interaction and the domain of work, and the way in which it might be re-conceived in part to permit multiple agents.
2. The Conceptualisation
2.1 Performance and Engineering
There are two main issues to be addressed when one considers the part the agents play in any work domain: firstly, how they relate to that work domain, i.e., how the interface should be represented, at least for the designer; and, secondly, how they relate to one another in the work environment. Perhaps the essential feature of cognitive engineering, as we understand it, is that it is a cognitive design activity which takes place against a background of scarce resources, as other kinds of engineering do. In order that a design problem can be defined and solved, a clear understanding is required of what constitutes performance. For this to be possible, agents doing work are, to one degree or another, conceived as set over and against the work domain, in order to facilitate reasoning about the design problem, e.g., to consider the merits of different specifications of agents’ behaviour to achieve the same work result, thereby answering the requirements, and enabling an assessment of the performance’s effectiveness. It is true that there is a dynamic and reflexive relationship which develops between the work being done and the work to be done, and that the agent contributes to this developing situation, i.e., the workers transform the object of their work, and consequently have to deal with a new object, sometimes with unanticipated properties. However, in order to get something done to advance the problem, i.e., to facilitate design reasoning, a decision has to be taken as to where the dividing line is between the worker, or workers, and the domain of work.
Thus a distinction is made between the work to be done and the means of getting that work done; the design problem being how to improve the means in order to get the work done more effectively. Enhancements to both must be relative to some criteria. These criteria, pertinent to each, are the resource costs of implementing the means, in the case of the agents’ behaviour, and the quality of what is done, in the case of the domain. That is to say, the doing of the work implies commitment of time and effort and, for the agents, the aim is to make these acceptable in the context of the requirements and the constraints of the setting. The work has to be performed as desired. The simplest way of putting this is to say that the costs of the work and its quality have to be weighed in the balance; and it is that balance which constrains the process of finding a design solution.
To put the point conversely, unless we have a clear idea of the agents’ behaviour and the domain of work, we cannot accurately and consistently measure the costs of their behaviour relative to the quality of work and, therefore, our design solution, or solutions, would be baseless, i.e., we would have no way of measuring performance.
Figure 1 Interactive Work System & Domain
As Dowell & Long (Dowell & Long, [8] & [9] have it, the conception of the design problem involves the fundamental duality of agents’ behaviour and the transformation of the domain contents, its objects and attributes. These are, respectively, the Interactive Work System (IWS) and the Domain of Work (the Domain) (Figure 1).
This duality is the basis for the measure of performance; and is why we can consider such a design problem specification to be, potentially, an engineering one. It is important, however, not to conclude that this separation, which is well-defined in order to permit such a calculation, leads to a radical detachment of the IWS from the Domain.
2.2 Autonomous Design, Human Factors, & the User’s View of the Work
As Dowell & Long [9] have also indicated, referring to Simon [10] and Neisser [11], it makes no sense to consider one in isolation of the other. They write, “If the worksystem is well adapted to its domain, it will reflect the goals, regularities and complexities in the domain”. Thus, what we are considering is a conception which not only permits a clear specification of a design solution determined by the calculation alluded to above but also, in spite of that rational basis, is arrived at through an understanding of the values of the work to be done. It is not, in any sense, a solution dictated by factors, or knowledge, external to the design problem, such as that, for example, drawn directly from the psychology of cognition or perception, though such knowledge might be the inspiration for a ‘conjectured’ (Popper [12]; see also below) design model to be defined as engineering knowledge through systematic design practices. The solution is at once domain-driven and IWS-constrained (in a way which awaits the particular knowledge of the domain and user behaviour).
In order that there is a means of connecting these limiting forces during the design problem-solving (i.e., the forces of motivation, or goal-fulfilment, on the one hand, and the constraints, representing the costs of so doing, on the other), there has to be a semantic connection between the IWS and the domain. This semantic connection is that of the intentions of the agents and the goals in the domain. These are linked, in the sense that the intentions result, when implemented meaningfully, in the transformation of states of the domain of work. In general terms, intentions in the work system are continuous with the goals in the domain, in the sense that objects in the domain accommodate those intentions (the affordances of the objects, as elements of ‘states of affairs’). An intention is not an intermediate goal to be fulfilled before the domain goal is specified or fulfilled.
However, though they are one in this sense, they are not one and the same, because the structures which support the behaviour of the worksystem and those which underlie the possible transformations in the domain are different depending on the work done. In one case, they may underlie the behaviour, and in other, support the work done. In any one interactive work system/domain of work they will relate the one to the other. The ‘reflections’ which Dowell & Long [10] (quoted above) allude to, in the well-designed system, amount to a congruence between these distinct structures. It is this double aspect of semantic continuity and syntactic disjunction which allows us to maintain the intimate connection between the users and the work, while rigorously separating them so as to define, and tackle, the design problem in a systematic manner.
The relationship, then, between the IWS and the Domain can be articulated
- so that performance may be measured, and
- so that a specification of the semantics of the design problem may be made and the solution may be addressed.
2.3 Human Factors and Joint Cognitive Behaviour
We have written throughout the last two subsections of the agents’ (plural) interaction with the objects in the domain, but in the Conception the agent is usually associated with the single user, as in a human-computer interaction. The relationship between several agents might produce odd results if we begin to consider each other as objects in the domain of work and not in the interactive worksystem.
However, the relationship between the agents in the IWS is not one of means-to-ends in a typical design problem, though people do use others in this way and one can envisage a design of a Machiavellian kind of problem. Indeed, this may be feature of the conception’s application to collaborative design problems which CSCW researchers think intractable. The agents must then be ends-in-themselves or they would be objects in the Domain, coming and going in and out of it, muddying the specification. When we refer to agents and the work they do, we mean that they are essentially all active and autonomous agents. So, in what might this relationship of cooperation consist?
It is the intention of this paper to introduce a further concept consistent with the framework, one which underpins this relationship. In other words, the paper will introduce a structure, or mechanism, which supports joint behaviour in the IWS, permitting the inclusion of social aspects in both the definition of the design problem and the specification of the design solution.
The mechanism is intended to fulfil a cognitive and a social function at one and the same time. However, our approach is still able to aspire to the title of engineering because of its emphasis on performance as the accomplishment of effective work of a desired quality at the expense of acceptable costs. It is often assumed that such an approach is reductive and mechanistic, and is unable to accommodate adequately the subtler concepts of cognition. It might be thought, for example, that such an approach would work if it is limited to the simplest case, but that ‘true’ CSCW would be beyond it, because in the class of socio-technical systems the focus is at the social end of the spectrum, which features strong social influences less constrained by technical knowledge, and composed of the co-workers’ diverse sets of values or perspectives (Vicente [7] points this out but fails, we believer, to deal with just those values and perspectives). The development of our conceptual framework for cooperative work should offer just this possibility.
3. Other Engineering Approaches
There are other workers in the field of human factors who also espouse an engineering approach, notably, when one considers the question of the behaviour of collective systems: researchers in cognitive systems engineering (CSE), and others comprising Rasmussen, Vicente, Woods, Hollnagel, Roth, etc. Their work is very rich and provides many means of representation for cognitive design purposes which assist design problem expression. They, for example, offer levels of description of agents’ behaviour in terms of skills, rules and knowledge, and as such their means of representation of this behaviour is consistent with another group of human factors workers, those espousing activity theory (AT) (see Draper [13]). However, as we have pointed out above
- these distinctions, in the first place – and with respect to, CSE, for example – concern individual agents, and do not relate those representations in social terms
- in the second place, and with respect to AT, although the group shares the cognition, the AT theorists can only advance design knowledge to a limited and uncertain extent without an explicit framework to support this collective cognition
- and in the third place, none of the above groups has an explicit and objective means of measuring performance, the pre-requisite of effective design products, in the sense pertaining to our cognitive engineering approach, outlined above.
Vicente and the CSE workers, in addition, have a concept of task which is left less well specified, apparently in order to leave room for evolutionary design development, i.e., not to limit the designer’s freedom nor that of the worker implicated in the designed system. However, we believe that it is possible to specify the task clearly as long as one is not confined to the conventional level of work (or technical) tasks; that is to say, if the notion of task is deepened to comprise the cognitive process, and is understood as shared in a more fundamental way (this should become clearer in our conceptualisation of cognitive coordination).
Vicente [7] argues for a review of the concept of task analysis, and we take account of this difficult concept. However, we feel that there is no need to make too much of Vicente’s distinction between “instruction-based” task analysis and the “constraint-based” form, but simply to recognise that task analysis must always be qualified by the limitations of domain knowledge. In other words, we hold to the position that a normative (i.e., a systematic and prescriptive) approach to design is required, but do not conclude, as Vicente does, that this implies dictated and rigid design solutions. The reason we are in a position to do this is that we take the view that such rigidity, which we agree is a danger, follows from applying, for example, psychological knowledge and failing to recognise the particularity of the design problem; as well as from assuming that the way things are done determines the way things should be done. We make no such assumptions and pursue engineering, not scientific, knowledge of workers’ behaviour,. Thus, though our output is prescriptive it is not a prescription determined by the way things are done in some domain of work nor the scientific findings of psychology which may or may not be appropriately scoped for the design problem under investigation. This tendency to lean heavily on psychology, we believe, stems from the perceived need for science to underwrite the guarantee for good design knowledge.
It is taken for granted that secure factual or descriptive knowledge is warranted by science, but as Popper has pointed out, even science needs to establish its knowledge through its practices: the hypotheses are bold speculations (Popper [12]). As we are trying to secure engineering knowledge we must, analogously, establish it through engineering practices, which are distinct from those of science and have different goals: diagnosis and prescription in the case of engineering, and explanation and prediction in that of science. To begin with, then, and analogous with Popper’s description of the acquisition of scientific knowledge, we start with a ‘conjecture’– the conceptualisation – but we do so with the intention of integrating it within a larger framework of cognitive engineering and, crucially, with the intention of operationalising, and ultimately testing and generalising some solution to the design problem.
After the conceptualisation, the exposition of the rationale for the conceptual structure or framework, there follows what counts as the next step in the acquisition of early design knowledge: seeing how it assists the analysis of a particular design problem. We might call this an intermediate stage between that of conceptualisation and that of operationalisation properly speaking: a stage where the concept benefits from feedback provided by its employment in the diagnosis of design issues – in this case, coordination issues.
The operationalisation has a converse and an obverse: the framework needs operationalising as part of the process of validation; and the properties, in this case of teamwork, need it too so that the design problem can be made more tractable, in such a way that reasoning about a design solution can be facilitated .
4. Conceptualisation of Cognitive and Social Coordination
Shaw & Fox [14] consider three types of system which might be thought of as sharing the task of problem-solving: collaborative reasoning systems; distributed problem-solving systems; and connectionist systems. Among the considerations which these types of group reasoning share (i.e., a component of a common framework for such systems) is what the authors call the ‘coordination mechanism’ (a term which we shall adopt, enhance, and integrate into the above conceptualisation of the design problem): “since each problem-solving agent only possesses a local view and incomplete information, it must coordinate with other agents to achieve globally coherent and efficient solutions”. They believe that “the design of coordination can be viewed from three different perspectives: the information content, the exercise of control, and the coordination mechanisms”. They go on to state that “coordination can be achieved through passing different types of information among the agents, such as data, new facts just generated, partial solutions/plans, preferences and constraints”. We may think, then, of these as the objects or contents of the coordination process. The authors suggest that “the initiative to coordinate may result from a variety of means of control: it may be self directed, externally directed, mutually directed, or a combination of them, e.g., coordination by synchronisation, coordination by negotiation etc. In short, therefore, the information etc. which is passed could be thought of as the content of the process, the “means of control” might be understood as the direction and force of the passing of that content.
If we look at Schmidt’s framework [1] we find him asserting, as we have seen, that “conceptualisation is, in principle, a coordination of viewpoints”, that “conceptual thinking is a cooperative effort”. We come then, in his view, enabled for coordination; and there is no inherent difficulty for us in coordinating: we start with a primitive basis for mutual knowledge, without which we could not communicate, i.e., it allows us to coordinate our attempts to communicate; unlike Shaw & Fox’s agents, which only possess “a local view”. It is part of what it means to have a conceptualisation of the world that it is shared. In contrast, the artificially intelligent agents under discussion by Shaw & Fox require an explicit input under something like the categories of ‘object’ and ‘means of control’ – what they call the “coordination mechanism” – in order for them to be coordinated (Fig.2). Like Schmidt, Hutchins [15] writes, “the distribution of labour can only be negotiated if the distribution of knowledge and ability is at least partially redundant”. That is to say, cooperative behaviour presupposes a common or shared view – a fundamental coordination, such as Schmidt writes about.
What we may now consider is how this natural, and fundamental, coordination which Schmidt writes might be characterised in a more formal and explicit manner by reference to the idea of a ‘coordination mechanism’. If this is possible we could be better equipped to articulate problems concerning the design of group behaviour.
What we are looking for then is a means of amalgamating the analysis of Shaw & Fox, which exposes the elements of a coordination mechanism but minus the assumption of mutual knowledge. There is a convenient concept associated with linguistic behaviour which has striking similarities in structure and in function, we shall argue, to the coordination mechanism as exposed by Shaw & Fox [14]: that of the Speech Act (Austin, [16]; Searle, [17]).
The Speech Act is a structure with two components, the ‘illocutionary’ and the ‘propositional’, where the illocutionary might be the act of commanding or asserting something etc., and the propositional content might be the representation of some state of affairs commanded or asserted. Together, such components might be exemplified by the imperative ‘put the cat on the mat’, which is composed of the mode of expression – a command – and the state of affairs envisaged as the object to the command, the cat’s being on the mat. Thus even an utterance that the cat is on the mat carries a mode of expression – the stating of a fact for a purpose plus the bare state of affairs concerning the cat; perhaps thereby informing someone of that fact to some end. Even though this assertion “The cat is on the mat.” seems indistinguishable from an exposition of its content (i.e., the propositional component) it is, nevertheless, not merely representational of a state of affairs. It is a speech act, and comes with the intention of having an effect. Speech Act theory claims, as the name suggests, that speech should be understood as another form of action.
Figure 2 CM & the Speech Act
These components then can be described in terms of the content or state of affairs – the cat being on the mat (unasserted), and the exercise of control – some communicative act over the content or state of affairs: in other words, an analogue of the coordination mechanism as described by Shaw & Fox [14] (see Fig.2). In the first case – the statement of a fact – the exercise of control is in the form of an assertion of the state of affairs expressed in the proposition, the cat’s being on the mat, and can be true or false; in the second, it might be in the form of a command that the state of affairs should exist and that someone should effect it, or, say, in the form of a warning, recommending some urgent response. Both may be tasks to be effected or requirements to be fulfilled.
We can recall the injunction by Dowell & Long [9]: “If the worksystem is well adapted to its domain, it will reflect the goals, regularities and complexities in the domain”. This expresses the rationale for uncovering links between the complex of tasks and the structures of the worksystem. If we take another example of an utterance in the form of a warning “The bull is charging”: one of the component of the CM (enhanced variety of coordination mechanism) is determined by perceived goals (Do we try to identify the breed of the bull, wonder at its strength, or run for our life?); the other refers to the state of affairs (The bull is at A and charging towards B, and you are at B)
Two things should be emphasised, and the one is linked to the other. The first is that these utterances are expressed for a purpose in the normal way of things. A statement of facts is made, for example, to satisfy someone’s curiosity, perhaps allowing him or her to make or modify some plan of action: to use the vacuum cleaner now or later. A command, or a warning are perhaps more patently also links in a chain of actions. They too are actions, and they are goal-oriented. The second thing which needs to be taken note of is that, as such, they have to be coordinated, like other group actions, for communication to take place at all; and that, as Clark [18] puts it, “language use and joint activity are inseparable”.
So, what we are suggesting is that it may make sense to suppose a more fundamental structure – the CM – which is goal-oriented, might underpin models such as distributed cognition, may support all of what Clark means by ‘joint activity’ (including the use of language) and comprise coordination in CSCW, i.e., working or behaving together effectively.
The difference alluded to above between the artificially intelligent agents and their human counterparts is that, in the case of the AI agents in [14] “each problem-solving agent only possesses local view”. By contrast, human agents often share information knowledge publicly and transparently, and necessarily share knowledge of conventions and values, as a basis for mutual understanding: jointly, these allow them to coordinate – including what is required for communication – on the basis of recognising objective states of affairs and of employing mutually understood means of controlling them.
Thus, to bring together the observations of Hutchins and Schmidt: cognition is already distributed, at least to some degree, for conceptualisation to take place, and the CM is the formal structure which can account for this distribution in the familiar terms of goal-oriented acts; at the same time, it brings together the terminology and conceptual apparatus of distributed cognition (DC) and distributed artificial intelligence (DAI). (Figure 3)
Figure 3 DC+DAI=CM for Language & other Joint Activity
To distinguish this CM from the speech act structure we shall refer to the element which exercises ‘control’ (in Shaw & Fox terminology [14]) as the ‘attitude’; and the object, data or representation of a state of affairs (designated ‘information’ by Shaw & Fox [14]) will be referred to as the ‘content’. In order to characterise the mutual knowledge underlying this fundamental communicative process, which supports both social behaviour and linguistic communication, we can adapt a famous account of non-natural meaning (in this case, the understanding of the intentions of others) by Grice [19] as follows: he writes, “‘The speaker meant something by x’ is (roughly) equivalent to the speaker intended the utterance of x to produce some effect in an audience by means of the recognition of this intention”. If we substitute ‘speaker’, ‘utterance’ and ‘audience’ for ‘actor/agent’, ‘execution’, and ‘spectator’, respectively, we get: The actor/agent meant something by x is (roughly) equivalent to the actor/agent intended the execution of x to produce some effect in a spectator by means of the recognition of this intention. We can see, therefore, how it might be possible to embrace indifferently both language and more general social activity with the more fundamental and general concept of the CM as an expression of a mutually understood state of affairs or ‘content’ governed by a meaningful ‘attitude’ in language, or in behaviour more generally.
5. Operationalisation of the CM
The foregoing is the case for employing a particular conceptual framework in certain situations of design involving group work.
- The employment itself of such a conceptual framework in these specific situations is what we call its operationalisation – with its feedback and feed-forward as suggested above.
- Series of such operationalisations to solve design problems amount to a validation of the concepts.
However, this is not a scientific endeavour and we are not trying to prove how things are but rather we are bringing frameworks and processes to bear on design problems, and if they have utility then to that extent they realise their meaning, and their value. If a conceptual framework, for example, assists the language of design allowing people in the field to communicate their ideas more easily, then this amounts to some success. The operationalisation, and a series of the same, amount to the engineering practices, which will constitute cognitive engineering knowledge.
We are arguing that the conceptual language currently at our disposal for cognitive engineering is inadequate for the purpose of tackling the design problems of group behaviour, and are attempting to correct this state of affairs, while maintaining this approach within a cogent discipline framework. The conceptual framework we offer should be considered as ‘early design knowledge’ which is in need of the operationalisation which results from addressing a particular design problem. At the current stage of the development of these ideas even operationalisation, strictly speaking, is problematic.
6. Conclusion
What we have tried to do is provide a way of understanding how agents might be conceived as working together in some domain or other, being equal members of an interactive worksystem, maintaining their independence but sharing their understanding of the task and working effectively to perform it. The CM is part of the structure which underlies collaborative activity, and allows the designer to specify the design problem of such collaboration for effective work by identifying the attitudes and contents of the CM. It permits a designer, for example to analyse dialogue and action into descriptive items reflecting states of affairs in the Domain and normative items reflecting goals of the work and intentions of agents in the IWS. It does this without resulting in a reductive account, and is consistent, we believe, with the ontology and related methodology of the original HCI Conception.
References
[1] K. Schmidt, Cooperative Work: a Conceptual Framework, J. Rasmussen, B. Brehmer & J. Leplat (Eds.), Distributed Decision Making. England: J Wiley & Sons 1991. [2] P. Falzon, Cooperative Dialogues, in: J. Rasmussen, B. Brehmer & J. Leplat, Distributed Decision Making J Wiley & Sons, 1991. [3] Bødker, S. (1998). Understanding Representation in Design. Human-Computer Interaction, 13, 107-205 [4] B. Brehmer, Some Notes on the Literature, in: J. Rasmussen, B. Brehmer, J. Leplat (Eds.), Distributed Decision Making, J. Wiley & Sons, England, 1991 pp. 3-14. [5] Storrs, G. A Conceptualisation of Multiparty Interaction., Interacting with Computers, 6, ii, (1994) 173-189. [6] J. Rasmussen, Modelling Distributed Decision Making, in: Distributed Decision Making J. Rasmussen, B. Brehmer J. Leplat, England: J. Wiley & Sons, 1991. [7] K. Vicente, Cognitive Work Analysis: Toward Safe, Productive, Healthy Computer-based Work, Lawrence Erlbaum Associates, 1999. [8] J. Dowell, and J. Long, Towards a Conception for an Engineering Discipline of Human Factors. Ergonomics, 32, 11(1989) 1513-1535. [9] J. Dowell, and J. Long, Conception of the Cognitive Engineering Design Problem (Peer Review Paper), Ergonomics, 41, 2 (1998) 126-139. [10] H. A. Simon, The Sciences of the Artificial MIT Press, 1969. [11] U. Neisser, From Direct Perception to Conceptual Structure, in: U. Neisser, Concepts and Conceptual Developments: Ecological and Intellectual Factors in Categorisatio, England: Cambridge University Press, 1987. [12] K. Popper, Conjectures and Refutations: the Growth of Scientific Knowledge, Routledge, London & New York, 1963. [13] S. W. Draper, Review of Designing Interaction: Psychology at the Human-Computer Interface, in: EACE News, Views and Reports, Int. Journal of Man-Machine Studies, 37 (1992) 812-821 [14] M. J. Shaw, M. S. Fox, Distributed artificial intelligence for group decision support: integration of problem solving, coordination & learning, Decision Support Systems, 9, , (1993) 349 – 367. [15] E. Hutchins, The technology of team navigation, in: Galegher, Kraut & Egido (Eds.), Intellectual Teamwork, Lawrence Erlbaum, Hillsdale, N.J., 1990, pp. 191-220. [16] J. L. Austin, How to Do Thing with Words. Oxford University Press, 1962. [17] J. Searle, Speech Acts, Cambridge University Press, 1969. [18] H. H. Clark, Using Language, England, Cambridge University Press, 1996. [19] Grice, H. P. (1959). Meaning. Philosophical Review, 66, 377-388