online hypertext document: http://www.cudenver.edu/~mryder/coss.html

From Center to Periphery:
Shifting Agency in Complex Technical Learning Environments

Paper presented at the meeting of the
American Educational Research Association, March 27, 1997, Chicago.

Abstract

This paper describes the development of COSS, a collaborative learning support system which evolved, growing with increased sophistication under the care and feeding of its users. The system took shape through a process of participatory action research involving users in the central roles of design, development and evaluation. This process reflects a shifting locus of control and agency, empowering users with tasks traditionally ascribed to instructional designers and knowledge engineers.

Introduction

The technology treadmill is in high gear. Like Chaplin's Little Tramp of Modern Times, we are compelled to keep up, producing greater results, always faster, better and smarter. Skills, once mastered, quickly become obsolete. New tasks suddenly emerge with their own specialized skill requirements. Complexity increases with each iteration of the cycle, adding new technologies over the old like layers of an onion.

As systems become more complex, there is a natural tendency toward decentralization of information and control (Kelly, 1994). This pattern holds true for a swarm of bees (Hoffmeyer, 1995), a market economy, or a learning infrastructure. The aim of our research has been to discover tools and techniques which enable peripheral, distributed control of complex knowledge systems.

This paper reports on the development of an experimental support system for technical practitioners. This Collaborative Online Support System (COSS) has been implemented among a team of engineering and support personnel whose job function is to provide advanced helpdesk support for users of complex computer storage peripherals. The group is chartered to handle the problems that field engineers and helpdesk personnel aren't equipped to solve: design flaws, oversights in product code, errors in documentation, or unusual customer implementations.

The initial goals of COSS emerged in the context of corporate downsizing, a time when product development cycles had matured and when planning for new products was not on the corporate agenda. The reduction in force was pleasing to stock holders (see Smith and Kliner, 1995), but it created a knowledge crisis within the organization. Knowledgeable engineers began exiting the organization at an alarming rate. As skilled employees left the workplace, they took with them critical technical knowledge accrued in the process of day-to-day engineering problem-solving. Knowledge was suddenly disappearing about the organization's products, the tools used to design and test them, and special troubleshooting procedures used to maintain and support the products. Maintenance engineers were left in place to support newly released products without having access to the knowledge and experience of those who designed the products. While many skilled maintenance engineers left the organization to take new jobs, those who remained became acutely aware of the need to document their day-to-day experiences into a collectively accessible knowledge base.

Action Research: the Development Methodology of COSS

The COSS system was not the result of formal objective design, with a front-end analysis producing a set of design specifications from which the product would be developed, tested and evaluated. Instead, COSS emerged through the iterative processes of participatory action research, a process which merges designer, user and evaluator into a common functional role. Action research involves the pragmatic exercise of abductive reasoning (Peirce, 1902) in contrast to more positivistic models with a greater split between research and practice.

The term participatory represents a significant shift in social science practice (Lewin, 1946). It recognizes the value of involving practitioners as essential participants in research, both as objects of study and as fellow-researchers sharing in all aspects of the project from design to implementation and evaluation. The term action implies a focus beyond mere data collection. The project involves direct activities which lead to qualitative change. The term research denotes a systematic effort to reflect on each implementation, identifying opportunities for improvement, constructing knowledge about the enfolding solutions and constantly generating new ideas. Those involved with the problems contribute equally to the solution process. It is a collaborative model which places the researcher in position as a partner working alongside those affected by the problem, rather than as an objective observer who might impose changes from the outside. In contrast to "basic" research, the focus is on specific desired changes in a unique situation (Deshler & Ewert, 1995).

Action research submits each new innovation to the test of practical use. The reflective insights of the researcher/ practitioner form the basis of the developmental process (Schon, 1983). Participatory action research offers a way to redirect resources from those who wield power to those who bear its consequences (Thomas, 1995). By empowering participants, the production and use of knowledge is transformed into a collective enterprise with value to all participants. Participatory action research thus serves the agenda of bringing theory and practice together.

The development of COSS draws from multiple theoretical positions spanning multiple disciplines including cybernetics and systems theory, cognitive science, activity theory, critical theory, and performance technology. The aim here is not to rationalize the design of COSS from firmly-held theoretical positions, but rather to provide a conceptual framework for understanding the support system.

Information, Knowledge, and Knowing

The commonly accepted approach to information and the use of information systems is representationalism. Good old-fashioned artificial intelligence (GOFAI - see Haugeland, 1985) is predicated on the assumption that knowledge is a commodity that can be stored inside of a machine. Such representationalist views approach information systems as designed models of an external reality which exists independently of the observer's actions. This view assumes an objective reality "out there" and that the goal in information development is to model that reality as accurately as possible.

In contrast, situated approaches view knowledge as an attribute of an interaction (Winograd & Flores 1986). In contexts of human activity, information is something that is evoked (Rumelhart and Norman, 1981), not recalled. The human mind is an organism under constant change as it interacts with its environment. Human long-term memory is a rich repository of past experience, but human knowledge cannot be reduced to stored information. As the capacity to interact, knowledge has no existence apart from its realization in activity (Clancey, 1993).

An objective repository, whether an encyclopedia or a rule-based expert system, is a closed system whose contents are fixed in time and space. Closed systems lend themselves to quantification, stability, and control. In contrast, the human mind is an open system which resists quantification and adapts readily to its environment. Information is not stored anywhere in particular. Rather, it is stored everywhere. We subscribe to the view that knowledge is not a fixed commodity, but a function of our interactions with external resources including tools, media, other humans. (Gibson, 1977; McLuhan, 1964; Norman, 1993; Vygotsky, 1978.) The act of knowing is the the ongoing mediation between past experience and present reality.

COSS is an open system meant to facilitate that mediation. Like an encyclopedia, COSS contains a large body of fixed information. But COSS is not strictly an information repository. The system  is intended to support users who are engaged in the act of knowing. The aim of COSS is to evoke information exchanges, to place the user in touch with all of the available means of support - not just objective information - but tools, media, and the experience of others, all within the context of an immediate task. The act of knowing begets new knowledge. As new experiences can be represented, the entailments are fed back into the system for the benefit of others (Pask, 1975).

Complexity Theory

COSS is a complex cybernetic system whose component parts are themselves wholes, including users, functional groups, and information systems, all of which are in a constant state of adjustment, interdependency, and fructifying change. The knowledge base interacts with users in multiple ways. All of the informaton that is contained in COSS was input by its users. Users look to COSS for hints, tips and solutions to technical problems. Each time a user query is rewarded, the system wins user confidence for subsequent queries. A user's willingness to share new information with COSS is a function of her personal history with the support system. How tightly each user adopts COSS into his or her practice will impact the amount and quality of information in the overall system. Consequently, each user influences the patterns of use demonstrated by all other users..

One test of a complex cybernetic system is its ability to adjust to the loss of critical elements. Remove a cogwheel from a clock and the clock will cease to function. But remove a generator from a power grid, and currents instantly route to other generators in the grid with no loss of energy consumption anywhere in the entire system. Cybernetic systems are self-correcting. Complex autopoietic systems demonstrate robustness that can adjust to multiple imbalances. Such information systems offer redundancy, with multiple modes of inquiry, validation, and correction.

Quantitative differences beyond a certain point pass into qualitative changes. Hegel - 1830

COSS as a socio-technical phenomenon is the product of a reflective group of software engineers within a well-equipped high technology domain. High technology is not an essential ingredient for collaborative knowledge building, but the consciousness, the skill set, and the tools available to this team certainly influenced the possibilities that enfolded.

Conscious development on this project began in 1993, at the threshold of the World Wide Web. The http protocol was in development and the the first web browser had not been released. But an infrastructure was in place that enabled the beginning of the COSS environment. The work group had a network of interconnected computer workstations and software which enabled shared resources, including a common file system.

In that year, the team began placing useful technical information within a central repository. A top-level directory was created, providing read and write access to all team members. The group began placing technical tips, question-and-answer notes, and full-scale technical documents in this directory. Engineers enthusiastically moved privately-held information into this common repository.

The Problem of Structure

It was only a matter of time before the structure of the resource began to inhibit its use. Subdirectories were created to organize the information in some agreed-upon logical order:

While such structure served initially, the engineering group soon discovered the futility of collectively organizing large bodies of information in such a manner. One person on the team was primarily responsible for maintenance and administration of the knowledge archive. But the administrator became increasingly frustrated as the resource continued to grow. Negotiation and coordination between the contributors, the users, and the structured repository became more and more demanding.

In the real world, information objects resist pigeon-holing. Most day-to-day problems stretch across multiple subject contexts (Spiro, 1991). As the structure became more complex, users had great difficulty deciding where an information object should be placed. Even worse, the same user had trouble retrieving information which he or she had earlier placed within the structure. Other users spent too much time browsing from one section to another in search of specific information. In many cases, information becames inert because the author's logic in placing the object did not match the reader's logical search patterns. Standard UNIX search utilities such as grep offered some assistance, but as the file system continued to grow, these tools showed their limitations.

In 1993, several technical innovations emerged that would enable the group to overcome the constraints of structure. The first was the GLIMPSE search engine (Mamber and Wu, 1994), an indexing and query system developed at the University of Arizona. This utility enables quick ascii text searches through a large UNIX file system. It allows for approximate matching (e.g., finding misspelled words) and Boolean (AND, OR) queries which can filter the search field to extract precise matches. In late 1993 the University of Arizona placed this tool in the public domain.

The development of html (hypertext markup language) and the emergence of powerful browsing tools such as Mosaic and Lynx enabled the possibility of converting a structured file system into an unstructured hypertext web. With these tools, information objects within the file system were accessible with the click of a button. Hyperlinks could now connect a document with any related document within the entire body of archived information. Where deliberate links had not been created, the search engine could link to any object identified from the user's search criteria. Now files could be directly accessed with a simple keyword. Searching and browsing tools overcame the constraints of structure. Together these tools enabled users to traverse the labyrinth with a focused intent. For those seeking information, the problem of structure seemed to be solved.

But the group's own paradigm of structure continued to inhibit the process of contributing information. Each time a user had information to share, it was necessary to identify an appropriate directory where the object should be placed within the common file system.

Another problem had to do with file security and collective ownership. In order to distribute the process of information development, it was desirable to provide alterable privileges to all users. However, this rendered the archive vulnerable to mistakes. On occasion, a user might erroneously "correct" valid and accurate information. In one case a user accidentally deleted an entire branch of the archive. Though daily backups were maintained, the need to secure the resource from such mistakes became apparent.

A third problem had to do with the matter of ownership. Who should be accountable for the information that resides in this common archive? Who should be responsible for corrections or updates? Should one user be free to alter the information contributed by another? Who is ultimately responsible for the veracity of the information? While contributors were encouraged to place their name in each submitted document, this was not always done. In many cases, the objects were orphaned. In other cases, an object might have multiple owners.

All of these issues were resolved once the group reorganized the file system. The resource became manageable the moment that content was divorced from structure. The structured file system described above was completely eliminated and each team member assumed possession of the information he or she contributed. Each participant now has a personal directory for shared information. All team members can read the contents of each personal directory, but only the owner can alter its contents. Any information which an engineer intends to share with colleagues is placed in this personal directory. The individual directories reside within a common searchable file system. At the same time, each personal directory is linked symbolically to the home directory of the individual's own workstation. This gives each contributor immediate access and total control over the information which he or she submits to the networked resource. Contributers structure their own directory however they wish. But structure is no longer an important consideration since every file is immediately locatable with the help of the search engine.

The revised structure leads to a stronger linkage between people and information. By structuring the resource according to the information owner, the URL pathname of every object automatically reveals its source. Users can now consult the originator whenever a situation requires further clarification. When information becomes obsolete or inaccurate, a user can advise her colleague of the desired correction. Alternatively, a user can copy a co-worker's original object, edit it, then place the corrected version within her own directory. Subsequent searches will reveal the original object along with any and all subsequent revisions. This strategy extends the capability of updates to everyone, but only the originator can alter or remove the initial document.

In this manner, ownership can be extended to all users without destructive consequence. From the standpoint of content, actual structure gives way to a virtual structure. Subject content is dispersed among the various personal directories of the team contributors. The objective detail of the file system reveals itself only in the context of user queries. It doesn't matter where the information resides. The search engine makes any relevant document available within the structure of the problem that prompts the inquiry.

The Problem of Ownership: Further Evolution of the System

The shared personal knowledge just described is one very important aspect of a multi-dimensional knowledge base. This aspect is informal and dynamic (Wilson & Ryder, 1996). It complements released documentation by offering timely updates, special exceptions, notes, reflections and informal advice. This knowledge space mediates between the crystalized representations of released documents and how that particular subject content is realized in the actual world.

We use the term public knowledge to denote formal, released information that was actually intended for broad distribution. In the technical support context this includes installation manuals, user guides, source code, industry standards, published FAQ's, product specifications, and instructional media.

In the COSS experiment, public knowledge has been especially difficult to integrate into the knowledge base. On the one hand, resources which could be obtained from the Internet were easy to appropriate. These documents included industry standards for data communication including network communication, SCSI and serial communication standards. There was also an abundance of information in the form of frequently asked questions dealing with the UNIX operating system, hardware platforms, standard computer peripherals, utilities, and tools.

Ironically, though, internally produced, company-owned documentation was more difficult to obtain. The publications group and the corporate training organization had their own procedures, their own tools, and their own set of goals. Those goals focused strictly on customer documentation in hardcopy format or training materials including overhead foils and student handouts. While the writers and instructional designers often consulted with the support engineers for subject-matter expertise and formative evaluation, they did not immediately recognize the value of integrating their own work within a collaborative knowledge base. The primary tools for tech writers and training developers were proprietary products which did not lend themselves to open system standards. Customer documentation was not easily portable to simple text or html format where it could yield its contents to an ascii-based search engine. Furthermore, initial requests to access unreleased drafts of technical documents were denied on the basis of "quality control". At one point, the writing group's management wanted to approve the manner in which the support engineers made use of their documents even though such use did not apply beyond the support group's own domain.

Extraordinary measures were undertaken to include customer documentation within the COSS system. Once this information was in place, it inherited the pliable, dynamic characteristics of electronic media. It was easily changed and updated as new information unfolded. Hyperlinks were added to connect the customer document with engineering notes, source code and field incidents. Engineering notes were actually inserted within the body of the customer document so that support engineers could benefit from tips and technical procedures contributed by peers.

Integrating software source code into the knowledge base was also difficult.True trade secrets were not distinguished from common algorithms implemented from a base of public knowledge. In reality, the unique trade secrets within the body of source code represented only a fraction of one percent of the code. Rather than identifying these unique algorithms and securing them from general internal access, every single line of source code was considered "secret". Making this general information available on an intranet web was construed by some to compromise corporate security.

The group which implemented COSS is chartered to support a software product. Having access to the source code of that product was essential. Again, extraordinary measures were taken to include this information within the knowledge base so it could yield its contents to keyword searches. Password protection had to be implemented to satisfy security requirements within this portion of the knowledge base.

Another rich source of product knowledge is a database of daily field incidents involving the product. When customers have problems or technical issues with the product, they call the Customer Services support line for assistance. Each of these calls is logged in an incident database. A description of the problem, the various symptoms, and the problem resolution are all logged within this database for each and every customer call. Like source code, this information is also considered secret.

Further extraordinary measures were necessary to include the field incident database within COSS. Password protection was again implemented. In addition, it was necessary to export the information from the proprietary incident database into the open-system knowledge base of COSS, thereby yielding its contents to natural language queries.

Field information can also be filtered, adding hyperlinks from the incident text to technical notes, to product documentation, to source code, and to related incidents. Where the text in an incident log makes reference to a previous incident, the filter creates a hyperlink to the previous logged incident. Where the text refers to a specific module in the product code, the filter inserts a hyperlink directly connecting the reference to the actual text in the source code base. Where an error code is referenced in the incident log, the filter creates a hyperlink to the exact node in the product document where the error code is defined. Where an interface standard such as SCSI or RS-423 is referenced, the filter creates a link from that reference to the top-level index of the SCSI or RS-423 standard. There is nothing magic about the filter. The technique can be implimented in standard shell or perl scripting language. It becomes magic when linking capabilities are enabled because all relevant documents reside within a common hypertext domain.

COSS Architecture

The content in the emerging knowledge base can be divided into four catagories, described by Jacobson & Levin (1995) as knowledge spaces:

• Public knowledge
• Collaborative knowledge
• Shared personal knowledge
• Experiential knowledge

Public knowledge is comprised of information which has been "published" or made public. This is in contrast to personal information or information that is shared among a small group of people. Once "published", a document is assumed to have some degree of veracity. It is assumed to have undergone the test of peer review or some other means of verification and the information rests in a stable state. At the same time, because it is stable, it is not likely to change in the near term. In the COSS environment, public knowledge includes user documentation, engineering specifications, industry standards, source code, FAQ documents, and other stable sources of information. Because the information is stable, maintenance is fairly low. A search index is rebuilt each time a new document is added. The public knowledge space is maintained and updated by a team member who has administrative responsibility for the resource.

Collaborative knowledge in the COSS environment generally refers to the field incident database. This information is volatile and is always in a state of development until the incident is resolved. A customer problem is reported. Questions are asked pertaining to the specific configuration and the apparent symptoms. Troubleshooting procedures are suggested, system traces may be collected and analyzed. Finally, a possible solution is recommended. The solution is applied, and if successful, the incident is closed. In a support organization that serves a global customer base, this database is in a constant state of change. COSS takes a snapshot of the database many times each day. With each snapshot, incident logs are filtered to build the hypertext links and a search index is rebuilt.

The shared-personal knowledge space contains the notes, memos, job aids and miscellaneous items archived for one's self, but also made available to colleagues. This space is maintained by each individual contributor. A search index is rebuilt once a day for this resource, though contributors can update the index upon command. The group strategy is to encourage individual contributions to this shared resource. Consequently there are no requirements defining standards, format, or content of the information that is placed here. The only general guideline is the pragmatic use of the resource. Team contributors understand that colleagues would query this knowledge space in search of technical assistance. Consequently, colleagues would not include administrative memos, bulletin-board announcements, and other general items which would not relate in a technical context.

COSS contains a fourth, experiential knowledge space. This is a set of tools, test resources, and emulation systems which can simulate any of several customer operating environments. A support engineer can access these systems from the Netscape browser just as if she were accessing any other knowledge space within COSS. The user selects the appropriate hyperlink associated with the desired system, and Netscape opens a login session to the target test system. This allows the problem solver to experiment in a safe environment, looking up system variables, testing bug fixes, or altering system parameters without impacting the actual customer installation.

A distinguishing feature of COSS has been its open architecture, minimal structure, and totally searchable access. Such openness shifts the locus of control from center to periphery: from designer to user; from author to reader; from expert to learner. In effect, the learners themselves are the developers of this system, drawing on generic tools and non-proprietary data structures.

COSS vs EPSS

How does COSS differ from a conventional Electronic Performance Support System? If such a conventional EPSS exists, it probably resembles the model suggested by Gloria Gery (1991). Gery suggests that "the goal of an electronic performance support system is to provide whatever is necessary to generate performance and learning at the moment of need" (p34). This is precisely the goal of the COSS system. Where practitioners are continually called upon to solve unusual, novel and complex problems, the need for dynamic learning support is critical.

Gery suggests that an EPSS differs from a conventional help system in the degree to which it "integrates information, tools and methodology for the user" (p34). Tools and methodology are richly integrated along with information in COSS. Where a process is defined, if it is feasible to automate the process or to simplify the procedure, users exploit the opportunity to implement the idea themselves. This practice frees all users to perform increasingly complex tasks.

Here is an example to illustrate the point: A user contributed a technical note which explains to other users how to transfer customer log information from a corporate ftp (file transfer protocol) server to their own work station. Another user later discovered an easier method to accomplish the same task. The second engineer copied the original document and added a simple hyperlink. Now, embedded within the very text that explains the ftp process is a hyperlink which takes users to the exact remote directory where the log object resides. With click of a mouse button and their Netscape browser, users can select the object and download it instantly, without ever having issued a single ftp command. The same browser that displayed the technical note also circumvented the process which the note describes, achieving the desired end performance.

Gery explains that the content in an electronic domain can be layered, allowing detailed and seemingly insignificant information to co-reside with general and fundamental concepts (p35). Since hypertext is not linear like conventional text, the user is not encumbered with massive volumes of irrelevant information that may relate to a given topic. Yet the information is there, waiting to be recalled if necessary. COSS and EPSS are alike in this respect. With its top-level hypertext indexes integrated together with deeply rooted search indexes, COSS allows the user to traverse any subject to any depth within multiple domains: from customer documentation, to training materials, to design documentation, to source code, to field incident logs, to application notes, to documented industry standards, to archived peer exchanges. It is the user who must frame the inquiry on the basis of the problem at hand and on the basis of her own enfolding understanding of the problem.

So what are the differences between COSS and conventional EPSS? The primary difference has to do with design methodology. The difference is in the approach and in the various roles that are undertaken throughout the development process. The Gery model is an objective approach which separates the designer from the user. This model evolved out of a tradition of instructional systems design (ISD), a tradition which extracts information from a target population, processes the information, then presents it back to the target group in a packaged format. At the front end, the designer works with experts from the population. The designer's product is ultimately delivered to novices within the same target group.

EPSS designers determine the needs and requirements of the user population. The designer defines the objectives for the EPSS, evaluates the audience and establishes user-interface and media requirements. The designer consults with experts to develop the EPSS structure and logic. The designer establishes the evaluation for each phase of development (Gery, p. 218-19). While all of these tasks are undertaken in collaboration with a subject-matter expert, the resulting product is separate and distinct from the group which will be asked to integrate the tool within their work environment.

The COSS system is not the product of objective design. It evolved out of conscious development through the interaction of the users themselves: experts and novices working together toward a common goal; practitioners involved in problem resolution by means of collaboration, the sharing of knowledge and information. Each individual on the team is both expert and novice, depending on the task at hand. The assumption is that information flows in all directions, that people learn from each other. There are no passive participants in the process of collaborative inquiry (Scardamalia and Bereiter, 1991).

Each question that is framed reveals a path toward a solution. Where the path leads to basic information, it is here that the design criteria emerge for an information system. When a group embraces the goals that Gloria Gery defined for an EPSS, it is only a matter of time before individuals come up with their own ideas to shorten the path to needed information in the process of performing a task. The available technology for an EPSS infrastructure has mushroomed since Gery's ideas were published in 1991. What was then an exotic high-tech implementation requiring technical specialists (Gery, p. 226) is now rapidly becoming common practice in networked practitioner communities.

In function, then, COSS can be seen as an EPSS. Yet in its open structure and situated, distributed development, COSS represents a significant departure. In contrast to the rational-planning model traditionally employed by ISD and, indeed, by engineers themselves, COSS was neither specified nor designed. There was no deliverable in mind! Yet it exists today as a fully functional and fully integrated tool within a productive work group. The best way to describe the evolution of COSS is by using organic metaphors. The form and structure of the support system undergoes continual evolution as users discover new ways to share and extract information.

COSS is a self-organizing structure whose elements continually adjust and change into coherent new patterns. These patterns are not decreed, designed or imposed by any specific individual. They simply happen. They may subsequently dissolve and leave little trace or they may have a longer lasting effect and change the structure of the organization. Understanding organizational change as a gradual, incremental process helps to enable self-organization as a means of creating new innovative tools and new patterns of behavior.

Summary and Reflections

Over the space of three years, COSS has evolved from a rigidly-structured information repository to a powerful, flexible performance support system with redundant, self-correcting capabilities, distributed control, and multiple points of access. For as many features that have evolved into COSS, there are many more that have yet to be discovered. The most formidable challenges have involved:

• capturing expert/novice and peer-to-peer exchanges, making this information readily accessible to anyone who requires it at a particular moment and context.
• developing a strong sense of ownership among team members, leading to greater levels of collaboration, sharing, and knowledge documentation.
• finding processes and modes of access that best fit the working styles and preferences of team members.

Of special interest in this study has been the process of capturing and archiving representations of new knowledge: information derived out of the solution of problems, novel procedures constructed in response to field incidents; elaborations and corrections of released documentation in response to questions from day-to-day product support, and personally constructed notes and job aids. A key development objective has been to achieve a painless archival process and an easy strategy for retrieving new information.

An additional goal has been to enable support engineers who are new to the product to quickly function productively in a capacity as knowledge engineers. The role of new users is not a passive role in this environment. New team members are particularly helpful in identifing weaknesses in the knowledge base. The unanswered query of a novice reveals the tacit knowledge of an expert (or an expert system). In the solution of complex problems, the path from object to individual and from individual to object passes through another person using the technology of language (Vygotsky, 1978). As experts and novices engage in conversation, the challenge remains for COSS team designers to capture and archive this linguistic entailment.

Our research interest in peripheral-control learning systems has raised some fundamental questions pertaining to the collective development and use of information systems. Questions such as these serve as the springboard for future discussions and continuing inquiry.

Who owns the information? Organizations produce information as well as knowledge. Because digital information is so easily copied and distributed, organizations rightly seek to exercise some control over it, to prevent its misappropriation by careless employees or competitors. At the same time, this controlling mindset can hinder efforts to distribute and share information across a problem-solving space. The present challenge is to discover when it is appropriate to secure information and when it is appropriate to empower practitioners with that information to promote greater productivity.

Who maintains the information? We found that difficulties arise when control of information becomes disengaged from its author. In general, quality is best served by maintaining the author-ownership linkage. Normally, however, such a linkage results in territoriality and hoarding of resources. The COSS system directly addresses this tendency by keeping access open to large repositories of personally controlled information. When it is necessary to update an information object, there are two remedies: (1) the information owner can provide the update; or (2) a co-worker can copy the item, provide the update, then "own" the revised document. In either case the interests of the organization are adequately served.

Who assumes mediating roles, assessing value, and worthiness? Information resources are amenable to a number of quality indicators. Some, like institutional review, can be labor- and time-intensive. Quantitative indicators can help somewhat in assessing quality, such as number of links, number of times accessed, and user ratings. In general, performance-support systems such as COSS function best when they seek to (1) distribute responsibility for quality across the participants; and (2) integrate quality functions into the normal interactions with the system. But in the final analysis, it is the user of an information object who alone can assess its value. Like a piece in a jigsaw puzzle, an information object cannot be assessed apart from the context in which it is used.

What central tasks remain? Leadership--both administrative and community-based--is important for the success of COSS-like systems. At the same time, leadership is likely to be quite different across sites and systems. In distributed systems, users share essential tasks which keep the system operational. Where central tasks are assigned to individuals because of special expertise, time, or commitment, methods of backup, redundancy, and shared responsibility should be instituted as the system becomes more complex, and as users become more dependent on its continued operation.

What factors promote or inhibit the sharing of knowledge? Models of technology use and adoption offer a number of psychological, social, and cultural factors that can inhibit or foster knowledge sharing (Rogers, 1995; Farquhar & Surry, 1995). Is there a correlation between a user's demand for quality and completeness, and that user's willingness to contribute? What is the relationship between frequent contributors and frequent users of the system? What significant experiences have changed how a user interacts with the system? All of these are legitimate questions awaiting further study.

Who is accountable for the veracity of new information? In the past, a clear rift existed between the printed versus the spoken word. With the emergence of e-mail and the Internet, casual, information interactions can leave a documentable trail. In this new setting where nearly everything is retrievable, two questions emerge: (1) how do you determine which information is accurate and reliable; and (2) how do you develop a sense of trust, truthfulness, and comfort among participants? As information becomes more volitile, it becomes unreasonable to require the author or some mediating entity stand behind the veracity of every information object. Consequently, it is left to the user of the information to determine its veracity and appropriateness for the given context. Public knowledge can assume quality control mechanisms such as peer review. Shared personal knowledge does not. "If you hold me accountable for everything I share with you, I am apt to share very little."

Are there identifiable patterns of growth or learning, e.g., accretion, tuning and restructuring of information (Rumelhart & Norman, 1978), within a collaborative knowledge base? Developmental models of learning, such as those developed by Piaget and Belencky and colleagues (1986) have proven helpful to understanding individual learning trajectories. Do groups of interacting workers engage in similar stages or levels of growth and development? COSS has undergone several episodes of major restructuring since the early attempts. Each time, the resource has become more powerful and dramatically more useful. At the same time, each restructuring has resulted in a more abstract representation of the knowledge base as seen from the user interface. Future analysts may discern waves of expansion and consolidation, reflected by knowledge-base contributions that show growth, resolution of conflict, or new levels of understanding among participants.

What patterns of use can be seen between expert practitioners and new learners? Expertise in using knowledge resources is partly a function of time with the system, but also of effectiveness in using resources--That is, did the user find the information needed? Did the information help solve the problem at hand? Recognizing the limitations of expert-novice research, we need profiles of proficient and novice users. What differentiates new users from experienced users? And what differentiates effective from ineffective users of the system? One experienced troubleshooter, for example, took to the system immediately, submitting queries, getting answers, refining queries, and traversing multiple domains within the knowledge base. This person contributed multiple simple notes into his personal shared directory. Each note contained a discrete item of technical information. In contrast, a less experienced practitioner approached the resource in linear fashion, reading one section of a document, then the next like a book. The type of contributions found in his personal shared directory tended to be structured, lengthy, and somewhat pedantic. Since the team is a small group, it is difficult to make generalizations from such a small sample. In addition to problem-solving expertise, additional factors also come into play, such as a person's social inclusion into the local community, and a person's skill and confidence in written communication.

What self-organizing patterns emerge as users take control of the environment? In this paper, we have traced the evolution of the present COSS structure. Where will the system go from here? As job requirements continue to change, as the technology continues to advance, and as group members fall into regular patterns of interaction, how will the system evolve? We fully expect to see patterns of systemic change, with unpredictable but understandable changes taking place over time, and with major changes occurring at times when the system is in a state of disequilibrium. Major change will likely be prompted by external disruptions such as organizational changes or new technologies.

How has the environment changed the way that support practitioners approach their job? One result of the current system is higher levels of productivity. The typical team member now has access to more information, and that information has a good chance of being helpful and accurate. Greater levels of communication across departments are valuable side effects of the system, with improved understanding of overall processes and individual contributions to that process. The system can more effectively absorb the shock effects of personnel changes, and can more easily be updated and kept current. All of these effects are reflected in the way individual team members approach their work.

The performance-support movement has renewed appreciation for the learning that can legitimately take place in the work setting. Systems like COSS take the point one step further and demonstrate that people, given the right tools and work conditions, can develop effective systems to support their own learning and performance on the job. COSS in many ways parallels the innovations found in constructivist learning environments. That is, COSS is to a work-performance setting what a community of learners (Scardamalia & Bereiter, 1994; Campione & Brown, 1996) is to instruction. In some ways, however, COSS goes beyond the research in classroom-based learning communities. The development of COSS did not rely on any particular theory or model, nor on authoritative subject experts or designers. In this sense, participatory action research as a method of development is also appropriately de-centered, giving voice and control to the periphery.

Belencky, et al. (1986). Women's ways of knowing. New York: Basic Books.

Bilson, Andy (1994) Facts, figures and fantasy: A constructivist approach to professional training in the use of client information systems. Online document. http://enith.asfh-berlin.de/ bilson.htm#fn0ba

Brown, A.L., & Campione, J.C. Psychological theory and the design of innovative learning environments: On procedures, principles, and systems. In L. Schauble & R. Glaser (Eds.), Innovations in learning: New environments for education. Hillsdale NJ: Erlbaum.

Clancey, William (1993) Guidon-Menage revisited: A socio-technical systems approach. In the Journal of Artificial Intelligence in Education 4:1, 5-34.

Deshler , David and Merrill Ewert(1995) Participatory action research: Traditions and major assumptions. Online Article: http://tdg.uoguelph.ca /~pi/pdrc/articles/article.1

Farquhar, J. D., & Surry, D. W. (1993). Adoption analysis: An additional tool for instructional developers. Educational Technology and Training International, 31 (1), 19-25.

Gery, Gloria (1991) Electronic performance support systems: How and why to remake the workplace through the application of technology. Boston: Weingarten Publications, Inc..

Gibson, J.J. (1977). The theory of affordances. In R. Shaw & J. Bransford (eds.), Perceiving, Acting and Knowing. Hillsdale, NJ: Erlbaum.

Grundy, S. (1982). Three modes of action research. as cited in Kemmis,S. and McTaggert,R. (eds) (1988). The Action Research Reader (3ed). Geelong: Deakin University Press

Haugeland, J.,1985, Artificial intelligence: The very idea, MIT Press/A Bradford Book

Heylighen, Francis (1995) Epistemological constructivism. In Principia Cybernetica Web Online Document: http://pespmc1.vub.ac.be/construc. html

Hoffmeyer, Jesper (1995) Swarm-semiotics. Paper presented at the 5th IASS congress in Berkeley, June 1995. Proceedings are forthcoming at Mouton Gruyter, Berlin/New York (edited by Irmengard Rauch).

Jacobson, Michael J. and James A. Levin (1995) Conceptual Frameworks for Network Learning Environments: Constructing personal and shared knowledge spaces. International Journal of Educational Telecommunications 1:4

Jonassen, David, Terry Mayes and Ray McAleese (1993) A manifesto for a constructivist approach to technology in higher education. Online document: http://www.icbl.hw.ac.uk/ctl/msc/ceejw1/paper11.html

Kelly, Kevin (1994) Out of control: The rise of neo-biological civilization. Reading, MA, Addison-Wesley

Kemmis, S. (1993) Action research and social movement: A challenge for policy research. Educational Policy Analysis V1, N1. Online document: http://info.asu.edu/asu-cwis/epaa/v1n1.html

Lewin, K. (1946). "Action research and minority problems." Journal of Social Issues, 2, 34-46.

Manber, U. and S. Wu (1994), GLIMPSE: A tool to search through entire file systems Usenix Winter 1994 Technical Conference, San Francisco (January 1994), pp. 23-32. Also, Technical Report #TR 93-34, Dept. of Computer Science, University of Arizona, October 1993.

Masters, J. (1995) The history of action research. Online document: http://www.cchs.su.edu.au/Academic/CH/teaching/AROW/rmasters.htm

Maynard, M. R. (1995) Action research in information management: Decision-making and implementation on the run. Online document: http://online.anu.edu.au/CNASI/pubs/OnDisc95/docs/ONL45.html

McLuhan, Marshall (1964) Understanding media: The extensions of man. New York: McGraw Hill.

Minsky, Marvin, ed., Semantic information processing. Cambridge, Massachusetts: The MIT Press, 1968.

Norman, Donald (1993) Things that make us smart: Defending human attributes in the age of the machine. Reading MA: Addison Wesley.

Pask, Gordon (1975). Conversation, cognition and learning. Amsterdam and New York: Elsevier.

Panagro, Paul (1996) Cybernetics: a definition. Online document: http://www.pangaro.com/~pangaro/published/cyber-macmillan.html

Pea, Roy D., Daniel C. Edelson, and Louis M. Gomez (1994) Distributed collaborative science learning using scientific visualization and wideband telecommunications. Paper presented at the 160th Meeting of the American Association for the Advancement of Science, February 22, 1994.

Peirce, Charles Sanders (1902) Application to the carnegie institution for support of research in logic, Memoir 19, On Arguments. available online "gopher://gopher.ttu.edu/00/Pubs/ransdell/On-Line Library/Peirce's Writings/MS L75: Application for support for his logic (1902)/Part 6 (32k)"

Resnick, Mitchell (1994) Learning about life. Artificial Life 1:2

Rogers, E. M. (1995). Diffusion of innovations (4th ed.). New York: Free Press.

Roth, Bennett, Woods (1988) Human interaction with an 'intelligent' machine. Cognitive Engineering in Complex, Dynamic Worlds (Hollnagel, Mancini, Woods; eds.); Academic Press, 1988.

Rumelhart, D. E., & Norman, D. A. (1978). Accretion, tuning and restructuring: Three modes of learning. In Cotton, J. W. & Klatzky, R. L. (Eds.), Semantic factors in cognition (pp. 37-53). Hillsdale, NJ: Lawrence Erlbaum.

Rumelhart, David and Donald Norman, A Comparison of Models. in Hinton and Andersohn (eds), Parallel Models of Associative Memory. Hillsdale, N.J.: Lawrence Erlbaum.

Scardamalia, M. & Bereiter, C. (1991) Higher levels of agency for children in knowledge building: A challenge for the design of new knowledge media. The Journal of the Learning Sciences 1(1), 37-68.

Schon, D. (1983) The Reflective Practitioner: How Professionals Think in Action. New York: Basic Books

Smith, Bryan and Art Kleiner (1995) Is the purpose of a corporation to maximize profits? in Peter Senge, et. al. The Fifth Discipline Field Book. Online document: http://www.fieldbook.com/

Spiro, R. J., Feltovich, P. J., Jacobson, M. J., & Coulson, R. L. (1991). Cognitive flexibility, constructivism, and hypertext: Random access instruction for advanced knowledge acquisition in ill-structured domains. Educational Technology, 31(5), 24-34.

Wilson, Brent G. and Martin J. Ryder (1996) Dynamic learning communities: An alternative to designed instructional systems. Paper presented to the Association for Educational Communications Technology, Indianapolis February, 1996. Online document.http://www.cudenver.edu/~mryder/dlc.html

Vygotsky, Lev S. (1978) Mind in society: the development of higher psychological processes. Edited by Michael Cole, Vera John-Steiner, Sylvia Scribner, and Ellen Souberman. Cambridge, MA: Harvard University Press.