Position Paper for the PDC/CSCW-Workshop
"Designing cross Borders: The Community Design of Community Networks"

One of the most revolutionary aspects of the internet is that everyone can be a sender and air her views world-wide. The opportunity to communicate asynchronously with others independent of time and place increases the usability of the medium. Our vision is to use that option to enable each and every interested person to participate in design processes of his (geographic) community.

Providing the grassroots option to Participatory Design for networked (geographic) communities brings possibly thousands of people together discussing with each other on some issue. Obviously, for the reader/participant of such a discussion it can be quite hard to keep track of the current state of the discussion, especially if she follows the discussion just from time to time. On the other hand it is likely to occur that a participant engaged in the discussion of one design aspects might miss to air views on important issues in other parts of the discussion. While in informal discussions or playground environments like Multi-User-Dungeons these might be aspects less important, the issue becomes crucial in discussions with design decisions which would have an impact on the participants real environment (e.g. discussions on design decisions in urban land-use planning).

Trying to introduce some structures and models seems to be a very natural way of addressing the problem. But similar approaches (Speech Act Theory resp. the Language/Action Perspective and its use in CSCW, i.e. the Winograd/Flores contributions [Flores et al., 1988 ; Winograd, 1988] and the Suchman reply ) caused vivid discussions on whether and how models ("categories") influence the communication itself and its real world context (i.e. workers and organizations). We consider Wanda J. Orlikowskis contribution to that discussion as the most relevant (). She pointed out (referring to Giddens structuration theory) that any category system ("model") enables and constrains communication since categories tend to stress some aspects of the categorized items, and thus dominating other aspects. What aspects are stressed and whether that should be considered as good or bad she states as dependent on the context and the content of the categories (being categorized by skin color makes a signif icant difference at the dermatologist resp. under an apartheid regime).

So we feel it is necessary to describe the context we assume for the following discussion. We consider a networked geographical community as a group of people who are having equal rights in the discussion, who are living within a certain geographical area and who are willing and able to use a computer-based medium for discussions. Further we assume that they have (in general) a positive attitude towards the community and that they are willing to solve the posed design problem.

Because of those assumptions we don't need to deal with the criticism on Speech Act Theory that modeling might amplify hierarchical relations and enable control. We claim that dependent on further context specifications structuring models can help in improving the aspects discussed.

We want to reflect upon existing models and ideas according to the costs and benefits for discussion participants the models can offer. We try to evaluate them by a qualitative analysis of these aspects.

Cost is, in the first place, the work the users have to invest to keep the model in a consistent state. The extent to which the model restricts the communication also counts as cost.

The benefits of a model are the services it can offer to users regarding the problems stated above (e.g. easy finding of argumentation lines, filtering, etc.). In certain situations (e.g. external time restrictions) even the restriction of communication bandwidth can help for a more focused and concentrated discussion style.

There are two ways of providing the participants with an orientation infrastructure. We can distinguish between approaches to model the discussion structure itself and approaches to model the state of the design issue currently discussed about.

The advantages of discussion-structuring approaches are that modeling is easier for the user and that he is provided with the complete context of the discussions. The disadvantages are that, if we have a design process which lasts very long, it can be quite hard for participants to read all the previous discussions just to know what is going on.

Issue-oriented modeling can provide users with an easily accessible, complete model of the actual state of the design discussion. But it is difficult to relate social information which occurred in the original discussion, to the issue model, since the information loses the context of the argumentation sequence. We will discuss further (dis-)advantages below along with the description of concrete approaches and ideas.

Apart from structuring, the discussion or the issue models can also provide automatisms (Query, ...) to support participants. We will deepen that below, too.

In this section as in the next one we distinguish the approaches according to their origin: Technical vs. Social/Psychological.

In our terminology full-text search engines can be seen as an automatism which does not require any modeling since it works on the original arguments of the discussion. We don't believe that search engines can provide appropriate solutions to our problem, since they can't reflect the argumentation structure of the discussion.

The IBIS method can be seen as the father of the two approaches described here.

• In 1988 gIBIS, a graphic-enhanced tool on the basis of the IBIS method has been presented (). The method covers three conversation acts (Issue, Position, Argument) and 8 relationships between them. Users have to categorize their texts on submission. The tool provides a graphical representation of the argumentation structure and the option to pose simple queries (ANDing of predicates). The authors mention the option of having a more elaborated query mechanism, which they did not implement, but "more sophisticated queries may be required, when the networks become very large". Which is certainly the case in our problem field.
• Another IBIS-based approach with extensions pointing to the direction of decision support systems has been published in 1997 (The ZENO Argumentation Framework, ,). In addition to primitives similar to those of gIBIS ZENO offers the opportunity to state preference expressions over previous discussion contributions as a discussion contribution and an evaluation function for discussion contributions and a computational framework to determine, what issue branches of the discussion have been agreed upon and what have been rejected. In the GeoMed project (at the GMD, St. Augustin/Germany) ZENO has been applied to discussion processes in urban land-use planning and is currently evaluated by users.

We note that in both Systems the option to lead meta-discussions on the discussion or the tools itself is mentioned, e.g. for building coalitions. The option for decision support has still to prove its effectiveness

In real world communication environments there have been developing routines to handle complexity based on trust. These routines could be (and - as far as we know - aren't yet) provided in a discussion forum:

• Recommendations: Participants are enabled to mark certain contributions or persons as "recommended" according to some criterion. These recommendations could result in a different presentation of the contributions/persons marked. Criterions could be content-related (e.g. a competence assessment like "engine specialist" in car design) or communication-related (e.g. "humor"), and thus can help showing not only the argumentation structure but also social aspects of the discussions. Recommendations can be a good starting point for exploring a discussion.
• Mandates: Mandates given from participants to other participants can to represent them in a branch of the discussion. Mandates could also be restricted in duration. The amount of mandates a person has can influence his or her representation.
• Delegation: Delegation establishes a relationship between two participants where one delegates his representation in an informally described part of the discussion to the other. What rights and obligations result from such a relationship (reports, obligations to "alert", etc) should be up to negotiations.

In general, trust abuse could also have an impact on the representation of the abuser, too, and can be a successful regulative (e.g. on ICS systems, which are MUDs for chess playing, an abuser is simply marked as an abuser instead of denying him access. Other players may choose not to play him by telling their "match assistant" to deny all requests from abusers.).

We feel insecure judging the effectiveness of those mechanisms in our context. Empirical investigations should be conducted on when a significant amount of participants feel that their point of view is not represented accordingly anymore.

This structuring approach aims at showing the current state of a discussion in showing the current state of what is discussed about. Since a design process usually can't be fully anticipated, discussions on the design process and the design issue itself belong to the domain.

Structuring domain knowledge is one of the main points in computer science, especially in Artificial Intelligence, Software Engineering and Databases. But of course most models have been build to enable an automatism to work on the information stored in a model and providing a result to the user. Our focus is to present a user an understandable representation.

Although some comments have been made on making formal models better understandable (e.g. for Object-Oriented Modeling), we believe that in our context the cost of building and maintaining a model which is up to date and detailed enough to be suitable for algorithms outweigh the benefit of the possible quality of representation. Those approaches only pay if there is a useful automatism providing additional service. Possible candidates include:

• Simulation: Simulation can be used for evaluating alternatives in a design process or for gaining a deeper understanding of the design issue. While with mathematical (numerical) simulation the relationship between cause and result is hard to comprehend, with qualitative simulations based on a logical model and an inference mechanism the way to a simulation result can be traced.
• Learning algorithms: Learning algorithms from symbolic learning in AI can be used to build a model from examples. Algorithms like Inductive Logic Programming () with the feature of "inventing" new predicates when necessary can collaborate with users in building a (first order) logical model of a domain. Algorithms like Explanation Based Learning can be used to reformulate an aspect of the collected domain knowledge by expressing the knowledge using a different, user-chosen set of expressions.

Looking at models in one of the most investigated design processes - software engineering (e.g. OMT [Rumbaugh 1991] for object modeling, Projects like EPOS [Conradi 1995], GOODSTEP [GOODSTEP 1995] and CoMo-Kit [Dellen & Maurer 1996] for process modeling) - where there is also a need to provide an accurate model for both, users (software engineers) and automatisms (compiler, profiler, etc.), we see that currently the modeling languages are dominated by automatisms and that users are specially trained to understand those models. Since these models are not intuitively understandable, we believe that formal models should only be used if there is a very strong need for the service of an automatism.

Another direction can be the presentation of "models", which are already existing and typical to the design domain, to participants for commenting. In the GeoMed project the users have been offered to make graphical annotations on digitalized land-use maps generated with the help of an Geographical Information System (GIS).

Several approaches to support the process of understanding complex domains by structuring have been developed. The Mind Map method guides the user to produce a graphical image which structures the domain of her ideas. Basically labeled nodes are connected with labeled edges, similar to semantic networks known from AI. The labeling is supported by the use of metaphors, pictograms and colors. There have been developed computer tools to build large concept maps where parts can be masked out if desired. Collective mind maps might be a way to structure a domain. An important issue here is whether users share a common set of metaphors and labeling expressions.

Concept Maps are similar. They also include methods for group decision support (statement clustering and rating). The design process shifts from a brainstorm focus to a rating focus. Trochim described the application of that method to planning and evaluation processes.

Though there have been developed tools to support both approaches it is not clear whether these approaches can be used for collaborative design. Especially the large number (and potentially large variety) of users might hinder a successful application.

We believe that up to now the judgement of the role of modeling in conversations has been merely subject to political discussions than to serious research. We plead for using discussion models where appropriate according to an accurate examination of the cost and the benefit of model usage. We believe that, in general, the more advanced a design discussion is and the clearer the design options become, the more formal/technical models can be used. The model choice also depends on the need for the service of algorithms like simulation. And, after all, the choice of a modeling option to use can be subject to a discussion, too.

We believe that those options should be subject to research to find out, to what extent the grassroots option can be used in Participatory Design. The research trajectory should begin with the informal structuring methods and introducing formal models and automatisms where there's a need for them.

Cited Literature: