DISTRIBUTED COGNITION

The capture/codify/store approach suggests that knowledge relevant to work can be externalised, made explicit, and embedded in representations that are unproblematically storable by others. A good example of such knowledge is an encyclopaedia. Recent work in the field of Distributed Cognition (DC) has analyzed work practice from the perspective of the way information is represented and propagated around a network of people and technology and shows the important way in which physical artefacts come to embody knowledge relevant to the work practice.

Distributed Cognition is a theoretical framework which has been developed by Hutchins, (1990, 1995a, 1995b) and colleagues. Its aim is to explain cognitive activities as being embedded in the work settings where they occur. Rather than seeing the individual as the unit of analysis it takes "a culturally constituted functional group" (Hutchins & Klausen, 1991), because, generally, most cognitive work is not done in an individual's head but is spread out between people and across artefacts and time (Halverson, 1994). The importance placed on enculturation indicates that this may be a view that can help refine the notion of soft knowledge.

Hutchins (1995a) points out that DC is to be found in all collaborative work and therefore is of use to the field of Computer-Supported Co-operative Work (CSCW), which has paid much attention to shared artefacts. As a methodological framework with the intention of aiding design, DC has been found to be suitable for application in areas where work activities are complex and socially distributed and use a variety of artefacts. It has been applied to a range of areas including marine navigation (Hutchins, 1990, 1995a), airplane cockpits (Hutchins & Klausen, 1991; Hutchins, 1995b), air traffic control (Halverson, 1994; Fields, Wright, Marti, & Palmonari, 1998), and aircraft failure management systems (Hicks, Wright, & Pocock, 1998), end-user programming (Nardi, 1993), the flow of work in a German Ministry (Mambrey & Robinson, 1997), work in a hospital radiology department (Symon, Long, & Ellis, 1996), and ambulance control (Martin, Bowers, & Wastell, 1997).

The DC view of knowledge being embedded in artefacts and representations that are propagated across different media and states is interesting, but it needs further exploration to ascertain whether it is predominantly hard or soft.

What is the Distributed Cognition View of Knowledge?

With the DC view of cognition, representations are viewed differently from "traditional" views of cognition (Zhang & Norman, 1994). They are not regarded as schemas and knowledge structures—more emphasis is given to external representations of both information and knowledge, that is, how information is processed across both the internal mind and the external environment. This is perhaps best explained by going to a key point of DC: how a problem might be represented to make its solution easier. If a problem is presented in one way, it may well be difficult to solve, but if presented in another way, the solution becomes immediately apparent. It is clear that not every problem can be easily re-represented to a state where its solution is obvious, and different individuals will see the solution more or less quickly depending on their knowledge and experience.

Hutchins (1990, 1995a) applied this idea when he undertook an ethnographical study of a navigation team on a US warship. He observed traditional computational problems, such as plot fixing, and saw the artefacts, such as navigational tools, that were being used to assist with the navigation. The tools, for example, an alidade^[4], a nautical slide rule, logs (the depth recorder's book and the bearing logbook), and a chart were designed to make the task much easier—they had represented the problem in a different way. Also of interest is the way these representations are propagated across different media and states (Hutchins & Klausen, 1991; Rogers & Ellis, 1994).

The fix cycle provides a good example of this. It determines the current position of the ship and predicts its future position. According to Hutchins (1995a), "The fix cycle is accomplished by the propagation of representational state across a series of representational media" (p. 117).

Landmarks are selected from the chart by an officer. These landmarks are then assigned to the bearing takers (one starboard and one port) by the bearing recorder. The bearing takers sight the landmarks and measure their direction from the ship using an alidade. When given the instruction to mark the bearing, the time is recorded in the time column of the bearing log, a fathometer operator reads the depth of the water from the display, speaks the depth into the phone circuit, and records the time and the depth in his own book. After the depth is given, the command "mark" is given by the bearing recorder, and the bearing takers can speak their bearings. The Quartermaster Chief records the bearings in the correct column in the bearing record log. The plotter reads the bearing as it is being written in the log and using an one-armed protractor draws lines on the chart to match the lines of sight between the ship and the landmarks.

In his example, Hutchins (1995a) is not breaking down the individual task of each member of the team; rather he is focusing on the way knowledge is being transmitted between the team members and how information is being propagated across different artefacts. The representation of the ship's position moves from the alidade, to the spoken word, to the bearing book in the domain of written numbers, to the plotting tool, and finally to the chart that is the domain where the information is now more easily interpreted. Rogers and Ellis (1994) describe it as:

• Remembering the name and description of the landmark: mental coordination of representational states.

• Coordinating name and description of landmark with external sighting using alidade: technologically mediated coordination of representational states.

• Bearer reports readings over phone circuit to bearing recorder: socially distributed and technologically mediated coordination of representational states.

• Bearing recorder records bearings in logbook: technologically mediated coordination of representational states.

• Information passed to plotter either verbally: socially distributed coordination of representational states.

• Information passed to plotter by plotter reading from logbook: mental coordination of external and internal representation states.

• Fix plotted on chart using instruments: technologically mediated coordination of representational states.

The use of the artefacts is central to DC because of the representations that they carry and the idea that they can contain embedded knowledge. The tools mentioned above were not only re-representations of a problem—they also represented the knowledge of previous generations of mariners via whom the artefact had come into being. For example, the chart represents the collection of more observations than any individual could make and therefore embodies generations of observation, experience, and measurement. The knowledge is embedded in the artefact—the knowledge and practice are crystallised in physical structures:

Physical artefacts became repositories of knowledge, and they were constructed in durable media so that a single artefact might come to represent more than any individual could know (Hutchins, 1995a).

This shows how knowledge relevant to work practice can be captured, codified, and stored in artefacts in ways that are more subtle than the explicit propositional forms of encyclopaedias and expert systems.

Hutchins (1990, 1995a) also observed that artefacts do not have to be physical. In some cases, procedures act as artefacts. Standard Operating Procedures are standard practice and are commonplace in navigation. Hutchins (1995a) describes the procedure as a mediating artefact, that is, mediating the relationship between the performer and the task. He emphasises that it is not to be seen as something that stands between the two but as one of the structural elements that are being brought into coordination.

The emphasis on information artefacts and propagation of the representations suggests that as far as Hutchins goes in his books (1990,1995a), his model of DC is representational and not constructionist. He believes in and argues for the distribution of information but he does not emphasise any differences in the types of information or knowledge. In Hutchins' model of DC, there is no distinction made between physical representation and representation in the head. There is some consideration and acknowledgment of cultural issues, but the view appears to be that cultural knowledge can somehow be embedded in the material artefact, and again the emphasis is on the importance of representational artefacts.

Even with the emphasis on representational artefacts, Hutchins (1990,1995a) downplays the importance of other forms of knowledge that are required by the users of such artefacts. For example, Goguen (1997) argues that all representations have both wet and dry components; Robinson (1997) in his study of models and procedures argues that the knowledge to use such things is different; Berg (1997) has shown how expert systems have failed to capture the social, organisational knowledge required to make sense of the output of expert systems; and Orr (1997) points to the importance of the more subtle aspects of knowledge required to work with explicit representation. An example of this is the case of a street map and a telephone book—if positions are simply shown as addresses it may well be difficult to get an idea of distance between them. This shows that other knowledge is needed to use the artefact and is not in the artefact itself. This is supported by another aspect that is emphasised by Martin et al. (1997): many artefacts (as opposed to communication technologies) afford possibilities and occasions for social interaction in the workplace. They expand on this with different knowledge being found in the interaction between people and between the users and the artefact, not in the artefact itself.

The DC emphasis on representational artefacts and their propagation shows us that, although cultural aspects are acknowledged, the approach is primarily representational. It is clear that some other kind of knowledge is needed to use the artefacts, the representations. A key issue is also how soft knowledge can be managed in a distributed environment. The DC examples do not deal with virtual or distributed teams, therefore, further exploration is needed.

An area where soft knowledge may be more apparent, therefore, is where the artefacts (the representations) are used in different communities. We have already seen that different knowledge is needed to use the artefacts, and this will probably be exacerbated when the artefact is being used in different communities or social worlds as would be found in a distributed environment. As the artefacts move between communities, we might expect that difficulties would arise with the translation of the softer aspects of knowledge, that is, the artefact would have totally different meanings in the two communities.

Boundary Objects

Boundary objects address the problem of common representations in different social worlds that border one another, for example, across different disciplines. They also provide a further opportunity to consider the softer aspects of knowledge as they are highlighted by the interaction in different social worlds. Star and Griesemer (1989) use boundary objects as an analytical concept and describe them as objects that are found in different intersecting social worlds and that satisfy the informational requirements of each. They are strong enough to retain a common identity but are nevertheless flexible enough to adapt to local needs. Star (1989), and Star and Griesemer (1989) based their notion of boundary objects on artefacts that do not change but are able to convey information over a distance. The boundary objects are therefore robust enough to travel between communities and also have local interpretations. Although they have a structure that is recognisable in different worlds, they can have different meanings within those worlds.

Star (1989) and Star and Griesemer (1989) identified four types of boundary object:

Repositories

They describe repositories as being "ordered piles of objects that are indexed in a standardized fashion" (Star, 1989, p. 48). They give the examples of libraries or museums that can be accessed by the different social communities for their own purposes. Repositories show that, although based on immutable mobiles, a boundary object does not have to be mobile to be a boundary object.

Ideal Type or Platonic Object

Star (1989) gives the example of an atlas as being an ideal type. It does not accurately describe the details of any particular site or thing. It may be vague because it is abstracted from several domains. It is precisely for this reason that it is adaptable to a local site. It is a means of communicating symbolically with all parties.

Terrain with Coincident Boundaries

This is described as "common objects which have the same boundaries but different internal contents" (Star, 1989, p. 49). As an example they give maps for different groups of people working at a museum. Maps for collectors and conservationists emphasise features such as trails and campsites. The professional biologists, on the other hand, have maps (showing the same state boundaries) that emphasise shaded areas representing ecological concepts.

Forms and Labels

These are boundary objects deliberately designed for communicating across dispersed workgroups, for example, a standardised form. They are the boundary objects most closely based on the immutable mobile—they contain unchanging information that can be transported across long distance. They have the advantage that they delete local uncertainties. They may become part of a repository.

Boundary objects are part of the representationist approach and therefore are predominantly "hard." Seely Brown and Duguid (1998), however, observe the different local interpretations. They describe boundary objects as being objects that are of interest to different communities that are involved in some way. These objects are used differently, or at least, regarded differently by the different communities. Through the boundary object, a community can gain some understanding about what is common and what is different about another community.

The different local interpretations, or interpretive flexibility, of boundary objects throw a different light on the shared artefacts of Distributed Cognition. It shows that the knowledge embedded in an artefact is not simply re-extracted, but that a degree of knowledge is necessary to be able to use the artefact; that is, some knowledge is not embedded in the artefact, other knowledge is applied to make use of the artefact. This other knowledge is perhaps the "soft" knowledge that cannot be represented.

It would appear, however, that the DC and boundary object view is primarily representationist with the main emphasis on hard knowledge and that DC is falling into the same trap as KM. This means that we need to look elsewhere in order to further inform the conceptualisation of the softer aspects of knowledge.

Cole (1996) felt the interaction between an artefact and the different social worlds, an under-emphasised aspect of DC, to be of great importance in the way it helped shape the artefact. The interpretive flexibility of boundary objects also points to differences in the different communities touched by a boundary object, despite the DC emphasis on hard knowledge, that aim to let the artefact move smoothly between them. This suggests that the context of the different social worlds may be able to further inform the concept of soft knowledge. One of the main issues that defines a social world is the language used by the members. The following section will therefore move to explore the notion of Common Ground in language that Clark (1996) uses to demonstrate how interpretations are created in the context of a community.

^[4]alidade: special telescopic sighting device