Organizational Memory Systems - Semantic Scholar

Chapter in the Encyclopeadia of Computer Science and Technology Marcel Dekker, Inc., New York, Vol. 40, 1999

Organizational Memory Systems Guy A. Boy European Institute of Cognitive Sciences and Engineering (EURISCO) 4, Avenue Edouard Belin, 31400 Toulouse, France +33.5.62.17.38.38; [email protected]

Keywords Active documents, cognitive function analysis, computer supported cooperative work, design rationale, field studies, hypertext, intranets, socio-technical systems, software agents, traceability of information.

Aim and content of the chapter This chapter introduces an agent-based approach to organizational memory systems (OMSs). It is based on OMS work developed at EURISCO that is multidisciplinary and multidomain, focused on the construction of OMS concepts for the aeronautical industry (Attipoe & Boy, 1995; Boy, 1995, 1997, 1998; Durstewitz, 1994; Israel, 1996). In many ways, OMS problems encountered in the industry domain are very similar to those encountered in the education domain, even if the productivity issues are not quite the same. OMSs are also related to the development of Intranets, that will enable massive information transfer within an organization. But they do not solve the major problem of existence or availability of the right information at the right time in the right place, and in the right understandable format. In addition, organizations evolve towards more autonomy, cooperation and coordination between agents. Machine agents, taking the form of software assistants, tend to replace human agents. The emergence of these new agents is creating new jobs. This chapter presents and discusses a theoretical framework that enables the analysis, design and evaluation of organizational memory systems. This chapter will emphasize the multi-agent perspective of cognitive modeling of sociotechnical systems. It results from this analysis that the more a software agent provides the right representation at the right time, the more it affords to retrieve the right information. To this end, representing context is a crucial issue that facilitates appropriate information retrieval and understanding. A context representation will be proposed. Not everything can be remembered, and not everything needs to be remembered. The way things are remembered is essential to insure fast recall and understanding. People are unique at remembering things and in explaining why and when they were stored. In this chapter, I will introduce a new kind of memory support using current information technology: active documents. They are mediating tools that enable people to store and use both content and context. Examples will be taken from the domains of aeronautics and education. This concept will be used to develop emerging issues of cooperation and coordination. In the balance of the chapter, a discussion will be started. Knowledge management and traceability will be specifically analyzed, and OMS research perspectives will be proposed.

Page 1


1. Introductory analysis Background and motivation Human beings have always tried to preserve information and knowledge one way or another. Sumerian literature is the oldest literature in history (6000 years old). Sumerian inscriptions found on clay were administrative, economic, and legal documents, including inventories, receipts, marriage contracts, wills, and court decisions. Egyptians recorded information on papyrus (after the 5th century BC). Papyrus was superseded by parchment in the 4th century AD. When Gutenberg developed printing, the number of documents started to increase, so that they were not only targeted towards a few initiated people. The intellectual world started to open. Changes started to appear incrementally with the development of books and journals in the nineteenth century. Today, we are living a new evolution phase, we are moving from paper-based information support to computer-based information support. This does not mean that paper documents will disappear. However, electronic documents offer new interesting possibilities. These possibilities include the integration of documents into complex software systems. For example, these systems become self-documented and interactive. Ed Barrett introduced the notion of Sociomedia that incorporates text and its human or artificial authors/readers (Barrett, 1992). In this chapter, I take the dual viewpoint that information technology (IT) influences directly the way people interact and work together, and that the evolution of our society influences the use (and thus acceptation, adaptation and development) of this information technology. It is important to notice that legal regulations follow this fast evolution of technology, often not without difficulties. When a safety-critical system is being certified, for instance, it is mandatory to « certify » new certification rules also. The current major issue is that technology is developing much faster than human kind can handle. If this human-centered phenomenon often slows down technology, it is at the same time very useful for controlling its evolution. Human-centered design is at the same time participatory work, and a ratification mechanism in the legal sense. Libraries, books and records of all kinds have existed for a number of centuries. People have learned to write, read and understand their content. Today, people want to be able to use information technology without having to read user guides. Writing and reading paper documents have progressively become design and interaction with information technology. Today, most people expect to watch television, to use computer software or to type documents without any extra burden. This evolution has induced the current development of computer-supported communication via electronic documents. In addition, commercial corporations, social associations and public organizations are living entities that develop their own knowledge and expertise. They acquire knowledge from interactions among their own agents in a similar way as an associative memory, as well as from interactions with the external world. An organization forgets because its agents forget things that they were doing everyday because they change jobs, retire from or leave their organization. Today, technology is changing much faster than before. People move much more than before. This permanent and fast evolution tends to increase organizational forgetting and the need to improve the structure and the processes of organizational memory systems (OMSs) which were naturally handled by humans in more stable environments. Today, rapid technology evolution and human Page 2


adaptivity show some incompatibility that we would like to further investigate. We would like to better understand and master the emergence of current human practice that is acceptable according to appropriate cultural and philosophical criteria. This chapter provides a socio-technical framework for further investigations.

Rapid societal evolution and memory needs emergence Computers are the latest tools that have emerged from a mathematized world. Today, computers are almost everywhere in our occidental society: at work, in administration, in amusement places, at home, etc. On the negative side, computers define an artificial world where reality is made of simulations. People may accommodate to the simulated world, loosing the sense of reality. On the positive side, computers enable people to produce, store and use various kinds of information. They were originally designed to process calculations, they have become mediating tools that can improve cooperation in the same way telephone does. In addition, they support information storage and access. An organizational memory system is an environment that mediates information exchanges among human and software agents at the same time or at different times, at the same location or at different locations. The World Wide Web is the largest environment of this type. Using the same technology, Intranets are transforming the office by enabling employees to cooperate on tasks, to retrieve information previously stored, and reconstruct the context of the production of this information in order to access the sense of it. The integration of new information technology in an organization is not only a technological issue. It poses the question of role (re)distribution, cultural heritage preservation and knowledge transferal improvement. Three main issues motivate this chapter: • Will information technology give birth to new OMSs? • What would the role of human beings be in such OMSs? • What will be the repercussions of this evolution on the way humans work? Humans are often the victims of new information technology because they do not assimilate or integrate it in an appropriate way, and/or at an appropriate time. The use of new information technology leads to the creation of new cognitive functions enabling the management of knowledge and action. Rapid societal evolution has become a rule to the point that tradition and local cultures are no longer crucial issues in our globalizing world. Everything needs to be new: new design, new shape, new function, new everything... In addition, people are moving from one job to another more than before. Organizations need to cope with this evolution also. Finally, information technology has induced new legal issues that were not anticipated before. Designers need to provide explicit design rationale when it is necessary, for example during certification or accident investigations. Organizational memory has two major orientations (Israel, 1998): a legal-based orientation and an asset-based orientation. The former emphasizes what should be explained to appropriate authorities to justify organizational decisions. Corresponding organizational memory systems should provide traceability mechanisms that enable certification teams to link evaluation criteria to organizational decisions. The latter emphasizes the organizational ability. In order for this ability to be preserved, knowledge and skills developed by an organization must be easily and readily available Page 3


to its employees. Training and documentation are usual means of transfering assets within an organization. Corresponding organizational memory systems should be as rich as possible.

Organizational memory metaphors Memorization of information is a usual cognitive process that we perform everyday. There are various ways to model this process. The following models can be used as metaphors for the analysis and implementation of tools useful in the development of an organizational memory system. Explicit versus implicit memory In cognitive psychology, a major distinction is made between explicit and implicit memory. Explicit memory is consciously stored in the hippocampus, and includes declarative knowledge composed of facts related to people, locations, objects, and numbers for instance. Implicit memory is unconsciously stored in the cerebellum, and includes procedural knowledge composed of perceptual and motor skills such as ability to play piano, write, or ride a bicycle. Access to information stored in the memory is a function of frequency and recency of access. The more you use implicit information the less you forget it. Information vanishes with time, i.e., when it is not used. When a piece of information is similar to a number of others, an interference process takes place and people do not remember. Every morning, I cannot remember my parking place since I am confused by previous places I used on previous days. Context of storage is often very different from the context of recall, and causes memory troubles. The way the information is stored and retrieved influences very much memorization (see the Art of Memory presented later in this chapter). This distinction between explicit and implicit memory can be extended to organizational memory systems. Since organizations incrementally build their own rules, they document these rules after facts that they try to regulate. These rules are basic declarative knowledge characterizing the organization, i.e., they explicitly say what should be done. In addition, people within the organization have their own ways of doing things. Implicit practice takes place in everyday activities, and persists while people remain within the organization. Human resources personnel try to make some of this practice explicit, and transform it into explicit rules by adapting it to the requirements of the overall organization. It is often the case that organizational rules are developed within a context that is likely to have changed when the rules are being used. Thus, organizational rules should be designed with their attached context of use. Short-term versus long-term memory In classical models of human information processing, the short-term memory (STM), that is seen as a set of buffers of a cognitive processor, is distinguished from the long-term memory (LTM) that constitutes (as a computer analog) the mass memory (hard disk metaphor) of humans (Card et al., 1983). The short-term/long-term memory distinction ads several characteristics to the implicit/explicit memory distinction: • The LTM is constituted of a network of chunks accessible through the short term memory by indices. The probability of retrieving a chunk of knowledge in the LTM is strongly related to the number of associations to this chunk.

Page 4


• The STM can be considered as a subset of the LTM. This subset is activated by the cognitive processor. Several chunks can be organized into a larger-sized chunk. For instance, the letters N, L, A, I, P, A, E, R constitute 8 chunks. These same letters, in a different order A, I, R, P, L, A, N, E constitute in general, for someone speaking English, a single chunk. Chunks can be related to other chunks. Thus, a chunk activated in the LTM may itself activate other chunks and so on (notion of associative memory). Activated chunks are added in the STM. • As new chunks are incrementally activated, old chunks disappear from the STM (limited capacity). The capacity of the STM is about 3 (2.5~4) chunks. It can reach 7±2 chunks thanks to the combined use of the LTM (Miller, 1956). • Chunks are not directly added to the LTM. The insertion of a new chunk from the STM into the LTM essentially depends on the possibility to construct associations between the new chunk and existing chunks. This activity necessitates the activation of chunks in the LTM that use a part of the STM capacity. Thus, storing new chunks takes longer than a regular access time. • The LTM contains both facts (declarative knowledge) and formal procedures (procedural knowledge and know-how). This distinction between short-term and long-term memory can be extended to organizational memory systems. The organizational short-term memory takes place in the current practice of the organization. It is usually constituted of active relations that evolve according to current events happening in and requirements decided by the organization. The organizational STM is sensitive to both external and internal events, as well as corporate knowledge stored in the organizational LTM. The organizational LTM includes all facts and rules that were previously designed and refined by the organization. Distributed memory systems Previous memory models describe mono-agent processes and functions. The evolution of information technology and the influence of anthropology and ethonomethodology (Hutchins, 1995) induced new developments of socio-technical models describing distributed cognition among agents. Agents are taken in the sense of Minsky's terminology (Minsky, 1985). In this perspective, an organizational memory system is a cooperating « society » of artificial and human agents. With the evolution of information technology artificial agents have become software agents (Bradshaw, 1997), also called intelligent assistant systems (Boy, 1991a), that are computer programs facilitating humanmachine interaction, as well as human-human communication. Software agents and metaphors can be seen as remembering facilitators in the same way mnemotechnic methods are. There is a motto that I always remember when I am trying to solve a problem: « it is much better to ask someone who knows that hiring hundreds who search. » It is often the case that someone in your environment knows what you are looking for. So why don’t we ask this person to help. This is where cooperation takes place. Obviously, you need to know first that the person knows and can help you. How do you know this? Either you know the person very well, or you know that he or she belongs to a wellknown class of experts. For instance, I have told many times to my students this episode that I experienced when I was starting to use Microsoft Word. My goal was to display the current footnote. At that time, Word did not have all the current useful functions. I Page 5


started to use the documentation provided by Microsoft. After a quick search in this documentation, I did not find any relevant information enabling me to display the footnote. During this information retrieval, a colleague of mine walked through the door of my office, and I suddenly understood that my problem would be solved in a minute. I knew that my colleague was mastering Word (this is an example of identification of an expert). I asked him: « How do you display the footnote, please? » He immediately replied: « Press simultaneously ---, and the same to hide it! » It goes without question that finding this complex command was quite impossible without expert help. I recalled this episode many times and finally tried to analyze it. This episode influenced the design and development of the CID system (Boy, 1991) that will be briefly described later in the chapter. In this episode, it is obvious that the kind of information I was looking for was not in my head and I could not easily find it in the appropriate manual. It was known by someone else. That is to say another agent provided me with this information. This example deals with two agents. By increasing the number of links among (expert) agents, we tend to increase the potential of the overall resulting memory. This is an example of distributed memory that is commonly used by people everyday. Space-time windows memory systems Historical facts and events are usually remembered using space-time windows. For example, we remember the period of the Second World War within several space-time windows according to the focus of interest. If we are interested in French issues, a possible space-time window could be the period of the resistance on the French territory. More specifically, we could be interested in the 1942-1945 period in the Vercors maquis. Within this space-time window, a network of agents enter into play with their actions and fact productions. Space-time windows specify contexts of validity of the described facts and events. The main difficulty here is to describe this highly dynamic context. It helps to figure out what the invariants of the context are. The notions of persistence and obsolescence are crucial to categorize and describe contextual conditions. If we consider a space window, the town is more persistent than the building, the building is more persistent than the room. If we consider a time window in aviation, a flight phase is more persistent than a flight sub-phase. Context can be organized into context islands with respect to its attributes and the persistence of these attributes. Context islands may be (Boy, 1998): • hierarchically dependent, they are then defined as mutually inclusive context patterns with respect to their relevant attributes and the degree of persistence of these attributes; • independent, they are then defined as mutually exclusive context patterns with respect to their relevant attributes and the degree of persistence of these attributes; • interdependent, they then share some context patterns. Any artifact, whether it is a physical tool, a concept or a piece of information, may become obsolete because the current context is totally different from the initial context in which they were generated. In addition, its life cycle may be very short, and it may become obsolete very quickly. Even if they are « formalizable », if they are not used for a long period of time, they are progressively forgotten, and eventually replaced by other artifacts. Note that there are some universal artifacts that remains valid across spacetime windows.

Page 6


An organizational memory system should be seen as a complex dynamic process where context plays a crucial role, instead of a context-free database management system. It is composed of human and software agents exchanging information that can be concretely stored either in people’s minds, on paper or electronically. What remains in people’s mind and is not stored in concrete media is forgotten in the long term. The major issue is to better understand space-time windows that characterize vivid remembering. When this vivid remembering pertains to people, usually time windows are around forty to fifty years if there is no information transfer to other people. What will be the time window for the information stored on CD-ROM’s? One reason why I introduced this concept of space-time window is because I believe that more research efforts should be started on the topic. I will not further formally develop the concept within this chapter, but I will take a few illustrative examples and concentrate on the associated notion of context instead.

External memory of cognitive functions Cognitive function definition In highly dynamic complex organizations, employees develop cognitive skilled processes that are very context-sensitive. These numerous skills can be approximated by cognitive functions. By definition, a cognitive function enables its user to transform a (prescribed) task into an activity (effective task). It represents a human cognitive process that has a role in a limited context using a set of resources. The role of a cognitive function covers the concept of responsibility (who is in charge?) Eliciting a cognitive function requires one to specify its context of use (where and when this function is relevant and usable?) Unlike goal-driven models, such as GOMS (Card et al., 1983), that tend to valorize smaller numbers of methods, context-driven models such as cognitive functions try to elicit organizational context patterns (Boy, 1998) that facilitate the access to an appropriate information at an appropriate time. A cognitive function is implementable when it is linked to the right resources that are cognitive functions themselves. Cognitive functions constitute the asset of an organization, and define an organizational memory system explicitly or explicitly. They are incrementally categorized according to context. The Cognitive Function Analysis (CFA) methodology that was designed and used for the human-centered automation of safety-critical systems (Boy, 1998), can be extended to the human-centered design of organizational memory systems. It is based on a sociocognitive model linking the artifact being designed, the user’s activity, the task to be performed, and the organizational environment. Cognitive functions can be allocated to humans or machines. They are characterized by their role, context definition and associated resources. The methodology is supported by active design documents as mediating representations of the artifact, the interaction description and cognitive function descriptors being designed, redesigned and used as usability criteria to evaluate the distribution of cognitive functions among humans and machines (see section 3 of this chapter). This methodology enhances user-centered and participatory design, and traceability of design decisions. It was successfully tested on three main applications in the aeronautics domain.

Page 7


The issue of forgetting in terms of cognitive functions People need to forget. A café waiter, for example, usually develops a tremendous capability of memorizing customer orders. He has developed cognitive functions facilitating information retention and recall. These cognitive functions are explicitly or implicitly based on a memorization method (see the description of the Art of Memory later in this chapter). The waiter uses his visual memory associating clothing details, age and location to the orders. He then organizes the drinks on his tray according to the associative visual cues. He rarely makes mistakes. Fortunately, he forgets these configurations once he has used them, otherwise he would be confused by overloading his memory. An electronic memory never forgets, except if it is programmed to do so. For this reason, appropriate access mechanisms should be developed to manage large amounts of information. Does an electronic memory need to forget? Let us take into account our electronic mail (email) system, for instance. We receive several messages everyday. Which ones should we keep? Which ones should we delete? Among the ones that we keep, how should we organize them? Should we revisit the messages that were stored and decide to delete some of them? What are the criteria for deleting email messages? Are there software cognitive functions that would assist the user in managing email messages? The interest of one message may be more persistent than another, and will be kept. We tend to categorize email messages to facilitate retrieval. We also tend to revisit old messages (categorized or not), and test the persistence of their interest. Messages may be deleted anytime. There are several levels of forgetting: individual, group, enterprise and community. The first three levels are usually organized hierarchically. The fourth level is transversal. People tend to recognize themselves belonging to a community of thought, a professional community or a religious community for example. A community is usually active across several space-time windows. At the individual level, we usually want to forget all information that is not in our current focus of attention. The focus of attention is defined here by intentions, past activities and any context of potential interest. Even if we delete an information in an electronic memory, we might never forget it. This is not true at the group level; a group seldom knows what an individual has deleted except if all members of the group are aware of what was deleted. At the enterprise level, forgetting is managed according to top-level policies and social practices. At all levels, whenever memorization and forgetting take place, a modeling process is implemented. Individuals do not store the same information, nor do they organize information the same way as others. All individuals who store and organize information in similar ways tend to implicitly or explicitly model a community. Groups and enterprises provide guidelines for memorization and forgetting by proposing meeting report templates, technical report formats and a culturally-oriented content production.

Rehabilitating the Art of Memory « Celestial navigation capitalized on the European virtues of mathematical theory and on instruments of high technological sophistication. In contrast, navigation in Oceania emphasized the deliberate refinement of people’s intuitive sense of direction and the learning of direct perceptual cues from the natural environment. From a seaman of

Page 8


Oceania, making a voyage is conceptualized as being within a pattern of islands, the positions of which are represented in his cognitive map » (Oatley, 1977). Like the Polynesians who used dynamic cognitive maps to navigate across the Pacific more than 1,000 years ago, could we use software agents to extend our short-term and long-term memories to handle our « navigation » in our modern world? The art of memory may take a larger place within this framework in the future. Computers should not be only used as calculation means, but as memory extensions facilitating remembering. The art of memory, invented by the Greeks, is seldom used today. People have almost forgotten it after the invention and practice of printing. This art enables someone to memorize loci (locations) and images imprinted on his/her memory. It is usually considered as a mnemotechnique. A locus is easily remembered, e.g., a house, a balcony, an angle, etc. Images are forms, distinctive signs or symbols of things that we need to remember. The art of memory is like internal writing. Even if it is not necessary, people who know the letters of the alphabet are able to write and read. Similarly, people who know the mnemotechnique are able to put what they have heard into specific loci and repeat it by heart (Yates, 1966). If we want to remember many things, we need to have a number of loci. A major condition is that the loci must be organized into a series that needs to be remembered in correct order. Training is a crucial issue. This way, one can go forwards or backwards from any locus. Yates considers that the loci have attributes such as: put distinctive signs every five loci; create these loci in isolated places; create memory loci that are different from each other. The art of memory is a particular indexing mechanism that enables people to invent loci and images (indices) that help remember things. Emotional events tend to facilitate the formation of such loci and images. Images can be shocking and unusual, beautiful or ugly, funny or rude. Good stories create emotions that are likely to create useful indices that will facilitate remembering. « The reason that we remember the stories teachers tell is that human memory is set up to retrieve and tell stories, as well as to capture the stories that others tell. The story is a unit of memory. Furthermore, good stories contain good images, novel ideas, or particularly poignant passages that enable our memories to create indices that make retrieval of these stories easier. Storytelling depends on being reminded of a good story to tell. And being reminded depends on having labeled the stories we have heard or have created well enough so that when those labels appear naturally in the course of a day, we can use them to find relevant stories. » (Schank & Jona, 1991.) There are talks that we attend that do not make sense because they are not illustrated enough with stories. They are so abstract that we cannot get to the point. In addition, the speaker cannot motivate us enough to create our own stories about what they are saying. The point here is that abstract information « remembered » by an organizational memory system is likely to create the same effect if it is not associated with good stories. Stories play the role of loci in the Art of Memory. A story can be summarized into a picture or even an appropriate icon that may be linked to a hidden abstract mechanism designed to provide requested information. Information access involves complex mechanisms that the user needs to learn and become expert with. Information access can Page 9


be facilitated by facing the user with interface objects that motivate the search and afford natural interaction with this information in his or her terms.

2. A cognitive function analysis of socio-technical systems CFA is adapted here as a methodology for human-centered design of electronic memory systems that involves the definition of a multi-agent cognitive engineering framework (Boy, 1998). This framework will be presented. In addition, a specific emphasis will be made on the representation of context for socio-technical systems. Contextual indexing will be taken as an example of cooperation among human and software cognitive functions. Active documents are introduced as a technological support of an organizational memory system.

The multi-agent cognitive engineering framework The agent-orientation of human-machine interaction is not new. Autopilots have been commonly used since the 1930's. Such agents perform tasks that human pilots usually perform, such as following a flight track, maintaining an altitude, etc. Transferring such tasks to the machine modifies the original task of the human operator. Thus, the job of the human operator evolves from a manipulation task (usually involving sensorymotoric skills) to a supervisory task (involving cognitive processing and situation awareness skills) (Sheridan, 1992). Software agent technology enables users to center their interactions at the content level (semantics) partially removing syntactic difficulties. It also enables users to index (contextualize) content to specific situations that they understand better (pragmatics). Hierarchy versus jazz orchestra Various organizations can be proposed for a society of agents. The hierarchical organization often comes to mind because it reflects the way conventional organizations work in our occidental social world (Figure 1). Each agent has a specific task to perform. Usually this task is assigned by an agent in the hierarchy. At the bottom of the hierarchical structure, basic agents perform elementary tasks. The more an agent nears the top of the organization, the more its job is cognitive. We say that a job is more cognitive than another when it involves more cognitive tasks and activities, i.e., more information processing. Our industrial cultural heritage is strongly based on taylorism that is a very good example of hierarchical organizational philosophy. This philosophy is based on a top-down transfer of information and control. Most actors are not globally responsible because they are not involved in the overall decision process. They are responsible for their job within the limited scope of their assignments.

Page 10


Figure 1. Hierarchical organization of a society of agents. An alternative to hierarchy is the jazz orchestra organization. It is represented in Figure 2. In a jazz orchestra, each agent performs an equally complex task. However, in contrast to the hierarchical organization where agents either control other agents and/or report to another agent, each orchestra agent acts according to a common set of rules (that can be transposed). These are coordination rules. These rules are defined by a syntax and a semantics. For instance, scores and music theory provide a set of such rules for musicians in an orchestra. In addition, there are transversal communication channels, and the overall coordination is managed by a single conductor. Actors are learning to interact and communicate with each other. Shared knowledge is often very useful to individual actors for the benefit of the overall organization. All actors are globally responsible because they are involved in the overall « symphony ». They are responsible for their job as well as for the result of the group. In the air traffic management domain for instance, the more pilots/aircraft become autonomous, the more they should become knowledgeable and informed agents. By autonomous, we mean that they know where they are located and what their environment (situation assessment) is, and where they need to go according to the current and future situation (goal-orientation). In the past, the hierarchical organization of the airspace was necessary and sufficient because aircraft were not autonomous from a navigational point of view. Today (and tomorrow), this state is changing. Most aircraft are now equipped with global positioning systems (GPSs) that provide more autonomy (Stix, 1994). The GPS provides pilots with a more precise location of the aircraft, thus pilots’ situation awareness is improved in theory, and they are in a better position to decide by themselves (autonomy). This statement is valid if they trust the GPS. The hierarchical concept is no longer compatible with this evolution. However, the orchestra concept is more appropriate. Each aircraft is an agent and control centers are other agents that work as conductors. The difficulty with several possible conductors is guidance coordination and consistency. Coordination, safety and consistency require the definition of cooperation rules that need to be applied by all agents in the society. This is the case with cars. There is a road rule system that guides the behavior of car drivers. There are also policemen who make sure that these rules are well applied to insure the safety of the road system. By analogy with a jazz orchestra, the conductor makes sure that each musician plays the right music at the right time with the right mood to contribute to the harmony of the symphony. Note

Page 11


that these rules can be explicit or implicit. In this view, data link needs to be thought of in a very different way since the current air-ground communication system is purely hierarchical. My claim is that new technology brings more autonomy to aircraft if it provides appropriate peripheral information to the pilots. This means that new cooperation and coordination rules need to be defined and incrementally refined.

Figure 2. Orchestra organization of a society of agents. Various types of agent communication When designing integrated systems the designer needs to consider the nature of communication among both human and machine agents. The type of interaction depends, in part, on the knowledge that each agent has of the others. An agent interacting with another agent, called a partner, can belong to two classes: (class 1) the agent does not know its partner; (class 2) the agent knows its partner. The second class can be decomposed into two sub-classes: (subclass 2a) the agent knows its partner indirectly (using shared data for instance), (subclass 2b) the agent knows its partner explicitly (using communication primitives clearly understood by the partner). This classification leads to three relations between two agents communicating: • (A) non-cooperation (class 1); • (B) cooperation by sharing common data (subclass 2a); • (C) cooperation by direct communication (subclass 2b). In the non-cooperation case, the agent is partially or totally ignorant of the behavior and reactions of the other agents. This can lead to conflict for available resources. Thus, it is necessary to define a set of synchronization rules for avoiding problems of resource allocation between agents. Typically, these synchronization rules have to be handled by a supervisor (Figure 3). The supervisor can be one of the partners or an external agent. Obviously, if available resources exceed the requirements of all the agents, conflict is automatically avoided.

Page 12


Figure 3. Non-cooperation: agents need to have a supervisor to manage their activities. In the case of cooperation by sharing common data, the agent knows that its partner exists because he/she is aware of the results of (at least) some of the partner’s actions. Both of them use a shared data base (Figure 4).

Figure 4. Cooperation by sharing common data: agents manage to communicate through a common database that is an interface between the agents. Such a shared data base can be an agent itself if it actively informs the various agents involved in the environment, or requests new information (self updating) from these agents. Agents use and update the state of this data base. An example would be both agents noting all their actions on a blackboard to which the other agents refer before acting. Agents have to cooperate to manage the shared data base. This is no longer a problem of resource allocation, but a problem of sharing data which each agent can use as it is entitled to. This paradigm is called a data-oriented system. Such a system has to control the consistency of the shared data. Cooperative relations between agents do not exclude competitive relations, i.e., shared data are generally supported by resources for which the corresponding agents may be competing. In this case, synchronization rules have to deal with resource allocation conflicts and corresponding data consistency checking. In the previous cases, the interaction is always indirect. In the case of cooperating by direct communication, agents interact directly (Figure 5). They share a common goal and a Page 13


common language expressed by messages, e.g., experts in the same domain cooperating to solve a problem. Agents communicate by incrementally constructing and sharing a common context. Each agent always attempts to construct a meaningful representation of the other agents in order to anticipate their behavior and reactions. Cooperation by direct communication involves learning about the other agents. Difficulty in or absence of learning may cause switching to either non-cooperation (and need a supervisor or a referee) or cooperation by sharing common data (and need a manageable interface understandable by each agent).

Agents Figure 5. Cooperation by direct communication: agents interact directly with each other. Flexibility and guidance in human-software-agent interaction The three types of agent communication and the distinction between hierarchical organization and orchestra can be very useful to analyze, design and evaluate humansoftware-agent interaction. In the non-cooperation case, a human agent knows very little of the other human and software agents constituting the organizational memory system. This human agent needs to be mentored by an OMS-knowledgeable agent. This type of agent can be an operation manual including organizational procedures implemented in the form of an intelligent assistant system. In the case of cooperation by sharing common data, a human agent understands the software interface of the organizational memory system. He or she typically interacts with (manipulates) clearly perceived interface objects that are familiar to him or her, e.g., menus, icons, electronic documents, and simulations. Interaction is more guided by the interface, and thus less flexible. The fact that the human agent has less options to consider results in cognitive load decrease, and does not need to be mentored by an OMS-knowledgeable agent. In the case of cooperating by direct communication, a human agent interacts directly with the other human and software agents constituting the organizational memory system. They must share goals, representation of the situation and contexts. Active documents enable this type of interaction. The problem is no longer an issue of a static number of options available on the user interface. Software agents are designed to dynamically provide the right option at the right time using the right format clearly perceivable and understandable by users. Guidance becomes an intrinsic asset of software agents. The

Page 14


flexibility needs to be analyzed at the level of the organization of human and machine agents.

Context representation The three levels of investigation Recently, Mantovani proposed a model of social contexts describing the agentenvironment interaction that is represented by the three following levels (Mantovani, 1996): • Level 1: construction of the context, i.e., social context; • Level 2: interpretation of the situation, i.e., everyday situations; • Level 3: local interaction with the environment, i.e., artifact use. Level 1 is more general than Level 2 which is more general than Level 3. At level 1, the social context is determined by values that are very general goals determined by the culture (Thomas & Alaphilippe, 1993). At Level 2, the psychological level, motivations are more precise but less persistent than attitudes, and are influenced by values and needs. At Level 3, the biological level, needs are biological strengths such as eating, drinking or sleeping. Mantovani describes Level 3 using the user-task-artifact triangle (Boy, 1998). He proposes equivalent triangles at Level 2, i.e., interests-goals-opportunities, and Level 1, i.e., action-history-structure. Mantovani’s three level model can be interpreted as follows (Figure 6): • a user may have situated interests for action in a given context of action; • the task that a user performs is based on situated goals coming from a social history; and • the tool is built from opportunities that arise from a cultural structure. Example of contexts in a design memory: legal versus design expertise reuse In current human-centered design, we tend to store design history to provide explanations to other people on how an artifact has been designed and why design choices were made at the time they were made. Traceability of design rationale has become a crucial issue in modern industrial organizations for two different types of issues: legal and design expertise. Legal issues are related to the process of certification of an engineered system as well as to the process of incident and accident investigation. In a certification process, investigators have to show that the engineered system is safe, comfortable and efficient to use, for instance. Accident investigators need to find the reasons why a catastrophe happened. Designers are required to provide explanations that guaranty them legally from being prosecuted. The level of explanations is usually safety-related. Design expertise issues are very different. They involve finer grain of knowledge that people tend to hide because it may not be safety-related but it is related to specific know-how that is often very difficult to formalize. Domain experts tend to be kept in their organization because this type of knowledge is unique and difficult to transfer. For instance, the number of remaining engineers who designed the Concorde supersonic airplane has decreased to the point that the corresponding design expertise has almost vanished. Since this type of knowledge is very precious, retired engineers are often solicited by the manufacturer to help new designers. Corresponding knowledge transfer is currently investigated (Israel, 1995, 1998).

Page 15


History

Structure

1 Goal Action Opportunities

2 Task Interest Artifact

3 User

Figure 6. Three level of investigation. Spatial versus temporal perspectives Organizational memory systems can be seen from two perspectives: spatial and temporal. Knowledge is usually spatially distributed within an organization. Knowledge distribution within an organization is handled by cooperation/coordination or hierarchical mechanisms for instance. Either people discover such mechanisms by themselves or they use an already existing cooperation mechanism provided by the organization. In hierarchical organizations for instance, people tended to create parallel cooperation mechanisms that are not provided by the organization. Information technology provide means to enhance cooperation using email for instance. Knowledge is produced to be reused. But knowledge producers do not know who will need this knowledge and for what type of reuse. This makes it difficult for the knowledge producer to package his or her production. This is clearly a usability issue where the context of knowledge production should be represented as clearly as possible. For example, someone who needs to reuse design knowledge produced by Concorde’s designers more than 35 years ago may face a difficult problem of interpretation. He or she may call the person directly to obtain the context in which this knowledge was produced. This shows that context is an important issue in organizational memory systems. In particular, context should be included in the indexing mechanism. Contextualization of information to memorize People usually find it difficult to elicit and formulate what they know. In addition, they only express what they can according to their role within the organization, i.e., trust, power and hierarchical relationships strongly influence the elicitation process. Finally, it is not because we can formalize knowledge that it becomes available to the community. Additional work is always necessary to make the produced elicited information understandable and transferable. In particular, facts to be stored need to be contextualized (Israel, 1998). This chapter advocates the use of human factors dimensions for the contextualization of organizational information to be preserved. These dimensions take into account training, operations and maintenance requirements. Function allocation within an

Page 16


organization should be based on these criteria as well as previous knowledge and skills. These dimensions can be included in one or several of the following classes of factors: • Socio-cognitive factors such as novelty, delegation, cooperation, coordination, training, personality, qualification, new job design, roles and work organization, efficiency, motivation, skills, control, ease of learning, information retention, complexity, reliability, errors and satisfaction. • Physiological factors such as stress, headaches, muscular and biomechanical problems. • Ergonomic factors linked to the interface such as input devices, displays, use of color, direct manipulation, graphics, natural language, 3D representation, operational documentation and multimedia. • Environmental factors such as noise, air conditioning, lighting or ventilation. • Productivity factors such as operator assistance, quality assurance, cost decrease, human error decrease, workload decrease, production time decrease and innovation increase. • Constraints such as costs, time scales, budgets, personnel, equipment and building structure. • Product functionality such as hardware, software and applications. Contextual indexing As already described, organizations are vivid repositories of information and knowledge. Appropriate information often needs to be retrieved at the right time to solve a specific problem. Use of documentation is very context-sensitive. Information retrieval cognitive functions often fail because they do not use the appropriate context. These cognitive functions may belong to people or machines. Cognitive functions providers cannot anticipate what users will need and use in the set of cognitive functions that they are developing. They do not know how users will enter into the documentation. For instance, let us assume that you need some very specific information on the air conditioning in your house. The first thing you may try is to use the keyword "air conditioning" to find the right cognitive function that would help to solve your problem. If you can specify the context of your problem, e.g., "you are a designer and are concerned by the connection of the air conditioning system, and have very little information about the electrical circuitry in your house", then you will probably find a list of vendors of pieces of equipment that may solve your problem. The solution will not be the same if you mention that "you are in your house and are freezing". You may find a way to repair the air conditioning system. In other words, stating a problem requires good contextual conditions, if one wants to solve this problem easily and appropriately. It is very difficult to elicit such contextual conditions since information holders do not know what they know, and in the best case often consider that such information is common sense. This is due to the fact that such knowledge is very compiled. However, if we consider the reasonable assumption that contextual knowledge is acquired incrementally, then incremental knowledge acquisition techniques are useful for on-line elicitation of context (Boy, 1991). Indeed, it is difficult for expert users to attach the right situation to any information retrieval strategy, simply because they do not remember well what they would do in a given situation. It is however, very easy to ask them to describe the relevance of retrieved information just after the fact (i.e., on-line elicitation). Obviously, the question is how to ask for such additional information from users Page 17


without overloading or "annoying" them. One partial answer is certainly to reduce the amount of interaction users will have to perform to accomplish this additional task, or to provide specific high-quality interaction. An organizational memory system is composed of agents that now can be humans or machines. Therefore, the definition of shared context between these agents is crucial to insure an appropriate communication among them. In the next section, the concept of active document will be defined in order to enable the user of OMS to easily store and retrieve any kind of information that is necessary at any time.

Active documents One way to avoid the need for extra training or workload is to produce appropriate software agents that can be naturally used by people. The CID project is an example of integration of software agents into active documents (Boy, 1991b). Direct manipulation improves the design and use of active documents. A user-centered answer to facilitate OMS integration is to satisfy conditions such as consistency of knowledge, internal consistency of the system that insures human reliability, context-sensitivity to the task, expert advice when it is needed, etc. Current documents are constructed from a variety of knowledge sources. They may have various formats according to the target and the available technology. The form and content of a document are both task-dependent (context of use) and domain-dependent (content). One of the main difficulties in designing active documents is to anticipate a very large number of contexts of use. Context of use is usually related to other entities such as situation, behavior, viewpoint, relationships among agents, discourse, dialogue, etc. Contextualization is extremely difficult using the conventional paper technology. It is made easier using computer technology when appropriate software agents are available or easy to construct. If active documents are understood by the user without external help, then they are selfexplanatory. Complementary documents are commonly used to understand original documents. In active documents, explanations should be formalized and transferred into a software agent that will help the user to better understand. For instance, in physics lab exercises, diagrams are presented to the students with missing parts that the students need to add in order to complete a consistent electrical circuit. On paper, these diagrams are presented to the student with a text explanation to explain what he/she needs to do. Using the computer, the same diagrams are active, so that by clicking on each part of them, hypertextual information (text or graphics) appears and explains what to do.

3. Active design documents In the classical cycle of technical documentation, design teams write requirement documents for manufacturing teams who then write documents for users. Design documents usually describe the way the designed artifact works. Operational documentation is usually developed at the end of the artifact development process. It unfortunately often attempts to compensate for design flaws. This paper proposes a different perspective. It presents the motivations of the approach, the definition of an active design document, related functionality in terms of cognitive functions involved in the interaction and the issue of traceability of design decisions. Page 18


Design teams exchange vivid knowledge of artifacts that they develop. For instance, design team players talk about the artifact, reinforce ideas, disagree with each other, or reach consensus. Descriptions and arguments remain traditionally documented in the form of text and drawings. Hypertext linking between technical documentation and artifacts provides a more active description of the ways in which the artifact works or should be used. Resulting documents enable the description of how the artifact works and how it should be used. In addition, linking interaction descriptions to corresponding artifact functions (Figure 7) is a step towards the formalization of cognitive functions involved in the use of the artifact. Cognitive function elicitation enables the design of interface objects that afford direct manipulation (Boy & L’Ebraly, 1994; Broigniez, 1996; Boy, 1998). As already mentioned, the CFA methodology and more specifically the distribution of these functions between human and artifact are based on the assumption that the artifact behavior induces user attitudes. The user reacts to artifact behavior and constructs his or her own attitudes to avoid cognitive dissonance (Festinger, 1957). Cognitive dissonance results from bad pattern matching between user expectations and real possibilities that the artifact can afford. User attitudes are based on a sum of beliefs and implicit evaluations of use possibilities. These attitudes lead to intent and behavior formation. Designers

Developers Artifact Technical documentation as user interface

Users

Certifiers

Figure 7. Technical documentation as a user interface of the artifact. Our approach supports the thesis that the quality of technical documentation contributes to the quality of design. We usually write for potential readers. In the same way, we design for potential users. We know that papers that we write must be reviewed by several persons before being delivered outside. We also know that artifacts must be tested by several persons before being delivered outside. The reader of a multimedia document has become a user of a software application. From this viewpoint, reading has evolved towards human-computer interaction (HCI). Writing has also evolved towards the design of interactive software. Writing words, phrases, paragraphs and chapters has become designing objects and software agents (Bradshaw, 1997). Static paper documents have become (inter)active documents.

Page 19


The active part of a book (system) is the reader (user). In addition, the organization of the book (system), the way phrases (objects) are written (designed), style and lexicon used suggest reader (user) activity. Sometimes, the reader (user) hardly understands what the author (designer) wanted to express. Instead of mobilizing reader (user) cognition on interaction problems, the most important part of the cognitive activity of the reader (user) should be centered on the understanding and interpretation of (active) document content. Human-centered design methods take into account users’ needs and requirements in the design/evaluation process. Instead of designing an artifact and documenting it later, we design and evaluate documented prototypes, called active design documents (ADD) in this paper, incrementally until they become acceptable prototypes. A main difficulty in technical document design is to anticipate a very large number of contexts of use. Context of use is related to entities such as situations, behavior, viewpoints, and dialogue. Conventional paper technology is not an appropriate support for contextualization. Software technology provides more contextualization capabilities. In addition, contextualization is both an intra-document as well as an inter-documents issue, for support of traceability.

Active design document: Definition An active design document is a hypermedia application usable by a community of persons. It describes various attributes of an artifact (being or actually designed). It is defined by three aspects: • interaction descriptions constitute the task space — they describe how to use the artifact, e.g., a procedure to follow (left part of Figure 8); • interface objets connected to interaction descriptions constitute the activity space — they enable the user to actually use a simulation of the artifact, e.g., a pilot can test a software prototype of a flight management system (FMS) interface (right part of Figure 8); • contextual links between the interaction descriptions and the interface objects constitute the cognitive function space — they enable the evaluator to annotate and comment the active design document during a usability test. Interaction descriptions of an active design document constitute the core of the humanartifact dialogue requirements. Interaction descriptions may be expressed either in natural language, or in a domain-specific language ranging from Simplified English to a knowledge representation such as knowledge blocks. Knowledge block descriptions enable semi-formal analysis of interaction complexity (Boy & Bertuccio, 1997), and elicitation of contexts of use as well as abnormal conditions. The use of knowledge blocks as a support for the generation of interaction descriptions is presented elsewhere (Boy, 1998). In this paper, active design documents are textual descriptions, e.g., operational procedures (see Figure 8). Interface objects of an active design document provide an appropriate, useful and natural illusion of the artifact. Interface objects enable the user to interact. They include important dynamic aspects of artifact properties such as color changes with respect to specific semantics or continuous parameters evolution. A user following interaction descriptions and interacting with interface objects is able to test artifact usability. For example, if the use of an artifact requires too much learning or is of little interest, it Page 20


might not be used. Interface objects may constitute the final product or an intermediary prototype. An active design document is equipped with an evaluation support that includes indexing, annotating (evaluation history) and browsing (hypertextual traceability, i.e., relating interface objects to interaction descriptions as well as active design documents among each other). The notion of context enables the customization and adaptation of multimedia documents to user requirements. In CID for instance (Boy, 1991), contextual links are incrementally generated by interpreting and annotating an integrated documentation. They are processed by a machine learning mechanism. For example, interaction descriptions generated by an author-designer or by a reader-user are often not the same, even if the same person is a designer and a user. The CFA methodology generalizes the CID approach: a contextual link base is incrementally generated by specifying, interpreting and annotating active design documents (Figure 9). Interaction descriptions and interface objects are concrete implementations of descriptors that can be clickable strings or graphical areas (e.g., an instrument or a part of an instrument such as a speed indicator in an aircraft cockpit). When a user selects a descriptor, he or she obtains one or several referents. These referents are consistent windows including interaction descriptions or interface objects. Contextual links elicitation consists of providing viewpoints on descriptor-referents relations. Either the user follows interaction descriptions and produces an activity by using corresponding interface objects, or the user interacts directly with interface objects and verifies the validity of corresponding interaction descriptions. Contextual links are generated and used incrementally to improve active design documents. They describe both design knowledge and artifact usability properties associated to contexts of use.

Page 21


Figure 8. An active design document: textual descriptions of required interaction (left part) and interface objects (right part).

Active DESIGN documents as a mediating environment for cognitive functions elicitation The basic claim of the CFA approach is that elicitation of cognitive functions involved in the interaction is equivalent to incrementally constructing a cognitive model of the use of the artifact, and to evaluating appropriate matching between interaction descriptions and interface objects. Drawings are useful for explaining an idea or a concept. These drawings can be done on any kind of support including a restaurant’s paper tablecloth or a blackboard. People usually choose cheap easy ways to discuss and formalize ideas. active design documents are intermediary supports between this type of communication and cooperation support, generally represented by conventional paper documents, and fully dynamic simulators. Though paper documents are cheap to generate, they are usually difficult to read and understand because they are not vivid enough. Simulations are much more expensive to generate, but can be very useful to test artifact usability. Active design documents constitute a compromise between these two extremes. By using rapid

Page 22


prototyping tools, active design documents are easier to generate. They are interactive and enable global and local usability testing. The prediction power of an active design document depends on the level and nature of its details. The more people are able to relate an active design document to the real world, the more it enables believable evaluation. There is, however, a compromise to be made between the level of necessary details as far as the development effort is concerned. Two types of active design documents can be distinguished: • shallow-global active design documents that provide a global active view of the artifact; they do not have full deep functionality, but they offer the possibility of using supervisory control, management or coordination cognitive functions; • deep-local active design documents that provide a local active view of the artifact being designed; they have full deep functionality of the artifact for very narrow tasks, and offer the possibility of using control and monitoring cognitive functions. Expert description of cognitive functions involved in the interaction is a way to test artifact usability. Cognitive function descriptors (CFDs) must be clearly defined to be compared and widely accepted. A CFD has the following properties: two CFDs provided by two different experts must be comparable (comparison property); and any CFD must be defined with respect to current scientific results in human-computer interaction and cognitive engineering, and domain terminology. The following CFDs were used as usability criteria in human-centered automation: prediction (capacity of anticipation of action consequences in highly automated systems); feedback (quality and speed); autonomy (domain of artifact autonomous performance); elegance (artifact capacity to avoid additional inappropriate cognitive workload, essentially in critical situations); trust; qualification level (ranging from the need for expertise to an interaction based on common sense); and programmability. The main issue is to understand if identified design flaws come from prototype approximations (shallow-global or deep-local) or from the developed concept itself. In the first case, it is a matter of explanation (e.g., wrong granularity, too shallow representation, bad assessment of the context of use). In the second case, flaw identification leads to revising design rationale.

Page 23


Figure 9. An example of comments as contextual links. Beginning in 1994, we have tested the use of active design documents on three types of applications in aeronautics: redesign of the user interface of a flight planning system onboard new-generation airplanes; design and evaluation of the user interface of an airground datalink communication system; and redesign of an Electronic Centralized Aircraft Monitoring System (ECAM) (example presented below). In these three cases, we directly faced industrial safety-critical systems. Results were well received by industrial experts in aeronautics (Boy & L’Ebraly, 1994; Broignez, 1996; Krishnakumar, 1996). In these three applications, hypermedia technology was used to build active design documents. An active design document is constituted of three windows: • the interaction descriptions window; • the interface objects window; • the contextual links window. Figure 10 presents an example of an active design document for the Level Change procedure of a new generation commercial aircraft. This example has been developed using Hypercard software and its scripting language Hypertalk. Figure 10 presents an interaction descriptions window (right side) and an interface objects window (left side). Page 24


The user has already selected seven procedure items (they are highlighted). Each time a procedure item (i.e., an interaction description) is selected a contextual link automatically sends a message to the corresponding interface objects window that produces an appropriate behavior. This behavior is evaluated and the corresponding contextual link is informed by generating relevant CFDs.

Figure 10. Example of an ADD of an Airbus procedure (Ids) and a simulation of an ECAM screen (IOs).

Traceability, design rationale and external memory Active design documents are permanent records supporting communication between the actors involved in the life cycle of an artifact. Active design document’s generation and maintenance enable domain actors to share concepts by writing and reading them (in the multimedia sense), and to be part of the artifact design-use-evaluation spiral. This approach concretizes Muller’s arguments advocating participatory design (Muller, 1991): • to combine diverse sources of expertise; • to formalize the ownership and commitment by all of the people who will eventually work on or with the designed artifact; • to participate in decision-making by the people who will be affected by the design decisions. The main difference between classical human-factors-oriented design and this type of participatory design is that instead of simply analyzing the existing artifact life-cycle, actors train themselves by cooperating throughout active design documents. The first approach is based on observation, the second one is based on cooperation. By providing users with design-aid tools such as active design documents, we enable them to contribute actively to design. Our first validation results (still preliminary) show this trend of social integration of users in the design process. It should be noted that active design documents can be used in conjunction with methods such as the Group Elicitation Method (GEM) (Boy, 1997). Active design documents are continually modified with respect to opinions of various artifact life-cycle actors, evaluation criteria (CFDs for instance) and domain and cultural requirements. When the active design document evolution leads to dead-ends, a Page 25


backtracking is performed to specific decisions that were made earlier, and a design history is kept in an external memory (Carroll, Alpert, Karat, Van Deusen & Rosson, 1994). This approach reveals that indexing (Boy, 1991) is a crucial issue to enable the traceability of design decisions that are included both in active design documents and in the relations between active design documents (Figure 11). The resulting library of active design documents is defined as an external memory. The active use of active design documents, i.e., not only reading but also writing, will contribute to change the organization of the designers’-users’ space and will really define a human-centered design environment. Basically, in the beginning of the design process active design documents include design-centered interaction descriptions that document a preliminary task analysis, roughly sketched interface objects, and contextual links mainly defined by the design rationale based on a first set of overall requirements. Later in the life cycle of the artifact, interface objects become more sophisticated and user-friendly, interaction descriptions should become minimal, and contextual links richer in comments and feedback from tests. The easier the interaction with interface objects is, the shorter and crisper interaction descriptions are. An important issue involves how to handle the growth of contextual links. This is precisely where traceability problems arise. Contextual links should be classified, generalized and incrementally simplified (sometimes forgotten) in order to be used efficiently. A first solution is to group them by viewpoint.

Traceability

Currentdocument

Initial document

Figure 11. Evolution of a design document

4. Active documents in education The evolution of learning technology shows that we are heading towards the construction of pedagogical tools that add pragmatics to current educational materials. Creating software agents involves new cooperation and coordination processes that were not explicitly obvious before. A specific case of computer-supported cooperative learning in physics will be given. I will then focus on the requirements for an educational environment based on the construction and exchange of documents. Examples of software agents for cooperative learning will be provided. Page 26


In the following example, software agents are added to existing documents to enhance their usability. Software agents provide pragmatics to the existing documents where syntax and semantics are already defined and will not be modified. This feature corresponds to the French unified school program. Even if this approach fits well with the French education system, we think that the separation of semantics and pragmatics is a general and useful concept for the design of active documents, i.e., electronic documents that include software agents.

An example in physics Let us take an example of a formal course on electrical tension. In this example, we show how a conventional physics exercise can be transformed into an active document by the addition of appropriate software agents. A conventional page describing the notion of potential difference or tension follows (Figure 12). NOTION OF POTENTIAL DIFFERENCE OR TENSION We observe a river water current. The altitude difference between two points of the river causes the existence of a water current between these two points. In the same way, we observe an electrical current in a closed circuit. The potential difference between two points of the circuit causes the existence of an electrical current between these two points. This analogy is displayed in the following figure:

Figure 12. Basic pedagogical document. Teachers may add appropriate agents such as denotation agents that show relevant parts of graphics explained in the text. These agents associate text descriptions to corresponding graphical objects, and conversely. For instance, by dragging the mouse on the sentence "We observe a river water current", the denotation agent shows the relevant part (Figure 13). In the same way, a definition agent can be programmed to establish the correspondence between a text description and a mathematical formula. When the text "altitude difference between two points" is activated by putting the mouse on top of it, a mathematical formula appears in context. The context is defined by the corresponding picture and the denotation of the two points A and B (Figure 14). NOTION OF POTENTIAL DIFFERENCE OR TENSION Page 27


The altitude difference between two points of the river causes the existence of a water current between these two points. In the same way, we observe an electrical current in a closed circuit. The potential difference between two points of the circuit causes the existence of an electrical current between these two points. This analogy is displayed in the following figure: We observe a river water curren t.

Figure 13. Use of a denotation agent. NOTION OF POTENTIAL DIFFERENCE OR TENSION We observe a river water current. The altitude difference bet ween tw o poin ts of the river causes the existence of a water current between these two points. In the same way, we observe an electrical current in a closed circuit. The potential difference between two points of the circuit causes the existence of an electrical current between these two points. This analogy is displayed in the following figure:

Figure 14. Use of a definition agent. An analogy agent gives the equivalence between various entities such as VA and ZA. By dragging the mouse on top of VA, the altitude ZA is highlighted and shows the analogy (Figure 15). These are very simple software agents that enhance the pragmatics of already designed physics courses. In this particular case, agents are basically hypermedia links between objects. Objects can be overlaid on top of graphical or textual parts of a conventional Page 28


document to create active documents. There is a tool box of agent types that the teacher can choose to program his own agents by analogy. Agent types can be: denotation, definition, analogy, suggestive question, problem solving (decomposition of a problem into sub-problems), video management, evaluation, hypermedia link to another active document, etc. Once the teacher has chosen an agent type in the tool box, a procedure helps him/her to design the corresponding agent by clicking on appropriate objects or locations on the screen. When in use, both students and teachers browse at their own speed active documents related to the lesson of the day. Individual backtracking is possible and encouraged. Eventually new agents can be created to enhance understanding of the concept to be learned. Students practice exercises by solving problems presented in active document exercises. In these documents, problem statements are put in context using agents in the same way as presented above. Suggestive questions guide the students. Hypermedia links to other relevant documents enable the student to remember concepts previously learned. An evaluation agent records students’ paths in the various active documents, as well as the answers to the questions posed. By the end of a session active documents are collected and analyzed by the teacher either on-line with the students, or off-line. NOTION OF POTENTIAL DIFFERENCE OR TENSION We observe a river water current. The altitude difference between two points of the river causes the existence of a water current between these two points. In the same way, we observe an electrical current in a closed circuit. The potential difference between two points of the circuit causes the existence of an electrical current between these two points. This analogy is displayed in the following figure:

Figure 15. Use of an analogical agent.

An educational memory in use Typical active documents such as those described above can be exchanged between teachers, students, parents, schools and homes. An educational memory is not a dead body of information but an actively growing accumulation of beliefs that have been put together (related or not) by people involved in the education process. These beliefs may evolve with time according to tests. An active document cannot become a stable and

Page 29


trustworthy knowledge entity1 until it has been adequately communicated to and approved by other people. This is a reason to enhance the educational memory interactivity both within the education system itself, and with other parties such as industry and the civil organizations. The educational memory can be seen as a large set of interconnected active documents that are logically and historically organized. This logical and historical organization is performed using contextual descriptions of the documents as described previously. It also includes a classification of software agents. This classification is incrementally acquired using a concept clustering process applied to software agents constructed by teachers. The block representation handles the construction and re-construction of such documents' organization (Boy, 1991b). Active documents should have appropriate indexing mechanisms. In the CID system, we have already developed an indexing mechanism that is suitable for incrementally updating descriptors of documents and attaching context to these descriptors. A descriptor is a partial description of the document that defines a particular semantic direction of search. A document is always partially described. This is why information retrieval is an abduction process (Pierce, 1931-1958). Abduction is the selection of a hypothesis from a predifined set. Context is added to the descriptors within a document to reduce the uncertainty characterizing the information retrieval process. Context is usually added either positively or negatively to descriptors after successful or unsuccessful document retrievals. When a document is retrieved, it not only provides content knowledge, but also contextual information such as who designed it, why it was designed the way it is (design rationale), who used it, who did not like it (user feedback), etc. Let us take a scenario of active document search and reuse. First, a physics teacher decides to give a course on the notion of potential difference and tension. She decides to retrieve active documents generated by other people. She makes the assumption that using the educational memory, she will find interesting active documents that she can reuse and adapt to her course. She tries to describe what she needs by specifying a list of descriptors such as "potential difference" or "tension". After a first information retrieval attempt, she gets more than 100 active documents. She does not have time to examine the whole set. She then decides to add some context to the descriptors by specifying "tenth grade" and "physics course". She then gets 7 documents that she can browse. She sees that some of the documents mention that the evaluation feedback provided by other teachers on 4 of them is not acceptable. She decides not to consider these anymore. To decide which one of the 3 remaining documents she will keep, she reads the annotations provided by other teachers and uses the documents themselves. Once she has used the selected active document, she provides feedback information on her own use of it. She may say that some children could not understand some parts of it. Thus, she has made some modifications that are included and contextualized in the current active document. The document is returned to the educational memory. In addition, a physics teacher may design a set of software agents that he/she can send to the educational memory for experimentation. Other teachers may use them and give their feedback. We think that this is a way to converge towards a normalization of 1A trustable knowledge entity is guarantied to work in a given context of validity. This is the case of physics formula

such as Newton's law "f=ma" to measure forces at the surface of the Earth.

Page 30


pragmatics in the teaching of physics. The main problem is for teachers to carefully annotate the active documents that they create, modify or use. In the current project, we try to better understand the human factors involved in the use of such an educational memory, as well as the underlying mechanisms that are required to support it.

5. Discussion and perspectives This chapter presents a view on organizational memory systems based on the analysis of cognitive functions involved in corresponding socio-technical systems and on the use of software agents. It takes into account the emergence of computer networks and their repercussions on people within organization. The design of an OMS may be very different according to the background of the design team and the requirements of the organization itself. We will start a discussion on a few possible views. (1) Research work is carried out in the knowledge management domain that includes requirements engineering, enterprise integration, use of artificial intelligence techniques for the development of OMS tools (e.g., knowledge acquisition, ontologies, data mining), and intelligent interfaces for knowledge retrieval. (2) Process modeling will be illustrated on an example of a large conference paper review management. (3) Traceability of information within an organization will be discussed using the active document approach. (4) Social issues of OMSs will be briefly remained, and finally (5) the conclusion of the chapter will present the perspective of novel research efforts.

Knowledge management and organizational memory systems There are various approaches to knowledge management including the use of knowledge-based systems. In this chapter, the objective is not to list the current systems that are developed or being designed because technology is changing so fast that when you will read this, these systems will probably be obsolete. Instead, the goal is to describe the key requirements, concepts, or challenges in developing an organizational memory system? An organizational memory system is not a static artifact that can be reduced to database management. An organizational memory system evolves at all times. An organizational memory system should facilitate information sharing and knowledge sharing. For these reasons, four concepts were chosen as fundamental to describe it: agents, decisions, knowledge, and context. As already presented in the introduction, the concept of cognitive function is crucial for the analysis, design and evaluation of an a organizational memory system. Human and machine agents include networks of cognitive functions that need to be described in order to represented and assess the complex relationships within an organization. These cognitive functions represent knowledge and know-how that the organization has produced. Decisions are made by all agents and shape the organization, i.e., the organization is as a living entity characterized by its agents and the dynamic relationships among them that are incrementally constructed from the decisions they make. Some agents may make minor decisions, others major ones. Agents exchange knowledge more or less formally according to the culture of the organization. Without context, this knowledge is useless. Since the difficulty is to contextualize knowledge, OMS research should focus on methods and tools that enhance the production of contextualized knowledge, as opposed to context-free knowledge. It is the price to pay to enable agents to trace knowledge within an organization.

Page 31


OMS support tools enable people to manage formalized knowledge as well as semiformal and completely informal representations that may vanish once they are used. Formalization is often obtained from classification and standardization of domain concepts and terms. When a concept is well accepted by a community of people, it can be used as a formalizing representation that will be shared by the members of this community. We cannot formalize everything. Only a few concepts can be elaborated and formalized. Among the rest, some pieces of knowledge can be represented using a semi-formal representation such as mark-up languages (SGML or HTML for instance), or ad-hoc structuring paradigms (hierarchical organization for instance.) Active documents are good support tools for the management of various kind of knowledge including interactive interfaces. OMS support tools enable people to maintain knowledge by assisting them in incrementally constructing contextual links among pieces of knowledge such as in the CID system. Machine learning should play an important role in the future of organizational memory systems. Software agent technology is likely to help in this direction. Classification mechanisms can be extremely useful to categorize information as well as cognitive functions that are needed to process this information. Motivation to store and share information and knowledge within an organization is not guarantied. Individuals may not see the real benefit of an organizational memory system. There should be a willingness to share knowledge. Human resource mechanisms should be discussed to improve cooperation and coordination. For this very specific reason, an ethnographic investigation of the culture of the organization is crucial. There are normative cognitive functions that can be easily identified, but there are also metacognitive functions that people use to work easier or faster, or for any other reasons that are related to the deeper culture of the organization. These meta-cognitive functions are essential to identify in order to better understand the structure and functioning of the organization. They often lead to « invisible » work that would lead to a disfunctioning of the organization if they were not performed.

The example of a large conference paper review management Communities have been formed on a persistent but evolutionary theme called HumanComputer Interaction. The largest community is the Special Interest Group on Computer-Human Interaction (SIGCHI) of the Association for Computing Machinery (ACM) that sponsors the CHI Conference every year. Over the years, CHI Conference Paper Chairs have developed a very good organizational memory system that preserves information on reviewers and paper selection mechanisms. It should be noted that (volunteer) Paper Chairs change every year, and it is not easy to learn and perform the job. Assistance is provided in the form of a overall process including several databases and several subprocesses. Even if most people are new every year in the Program Committee, a tradition persists over the years. The organizational memory system of the CHI Conference paper review management can be modeled as presented in Figure 16. This example shows that an organizational memory system cannot be reduced to a database management system. There are several processes that need to be handled by humans such as checking continuing interest and obsolescence of the produced literature, reviewing, mentoring, as well as deciding themes and writing the call for papers of the conference. Other processes such as paper/reviewer matching can be handled by software agents (computer programs) when they are appropriately fed by reviewer expertise descriptions and selection rules. Even if Figure 16 presents a very Page 32


simple overall process, it illustrates that human and software agents can be associated to keep an organizational memory system alive. This simplified example is intended to present the interest of process modeling. There are processes that are crucial to bring to the front, others can be kept implicit. In this example, we see that processes produce documents that need to be further characterized.

Figure 16. Conference paper review management.

Traceability Who is responsible for this decision? This is the type of question that one asks within a large organization everyday. The decision process within an organization is usually complex and results from the integration of several micro-decisions. Some of these micro-decisions are sometimes important to trace back in order to understand the rationale of a synthetic information. Traceability is not a simple concept to define. It involves at least the following dimensions: • design rationale, i.e., record design objectives and justification of a design decision, make the design process explicit;

Page 33


• legal rationale, i.e., demonstrate compliance to existing rules, or get convinced that things are done correctly; • context description, i.e., reserve time and project startup to use past experience (organizational memory), model the situation (roles, persons, reflexive behavior, viewpoints, constraints), agree about assumptions, or express design process towards a reader in order to become explicit – working in more contexts improves your sensitivity of context; various viewpoints (information and context descriptions that meet the customers needs); • anonymous versus responsible reporting, i.e., responsibility engaged in writing personal information, responsibility for a decision, accountability; • implicit versus explicit knowledge; • alternatives and decision criteria; • information granularity (how heavy the documentation is, bureaucracy versus efficiency); • indexing (adaptive, cost, knowledge should be contextualized); • tutoring other people; • remembering versus reinventing (knowledge reuse versus knowledge design). Traceability involves issues such as: • What do we need to document? Could we provide guidelines for things that are important to document, and other things that are not important? • When do need to document? We can document things as they occur, or document syntheses only. • How do we need to document? Using paper, electronic media, or people. Using narratives or formal representations. Using a distributed network memory system or a single database. • Why do we need to document? For legal reasons or knowledge reuse. Traceability is not only based on an indexing mechanism, but also on important questions such as remembering active knowledge, not in interpreting passive information. What is the role of software agents in supporting traceability? Two illustrative examples were provided in the previous sections of this chapter. Active documents are good candidates for the management of active knowledge.

Social issues The following social issues are crucial: • who defines the structure, function and content of an OMS; • an OMS and the overall organization influence each other; • an OMS should be defined according to the needs and goals of the organization and the personnel in the organization; • objects available in the OMS should be sharable; • knowledge transferred within the OMS should be controlled and evaluated; • an OMS tends to facilitate transversal exchanges within an organization (is the organization culture ready to accept and manage them?); • an OMS should be able to capitalize projects events across several work groups; • individuals may not see the real benefit of an OMS.

Page 34


Research perspectives In this chapter, we assumed that if technology is not a panacea for the organization, it can serve the proximal cause for mobilizing folk to actions (Soloway, 1995). Three main concepts have emerged: active documents, software agents and organization. Active documents are used as repositories of organizational and technical knowledge. People should be able to easily create active documents, as well as modify old ones. To facilitate active document design and publishing, libraries of software agents need to be created and maintained2. Software agents are observers, information processors, and proposers. They can be active entities added to conventional documents transcribed into an electronic form. Some of them observe user's interactions as it is the case in CID. They are able to produce actions from the data they have acquired from the user. The action performed by a software agent ranges from the activation of other agents to the execution of (computational) operations. Software agents are easy to manipulate and relate to each others. They provide vivid behavior for a user interface. They can be visible (audible), or invisible (inaudible). When they are sensorial they have a presentation shape (usually called a metaphor) on the screen, or a sound, and a behavior. Otherwise, the user does not know that they exist. In the field of electronic documentation, agent adaptivity has been shown to be extremely useful in information retrieval (Boy, 1991b). In this case, software agents are knowledge-based mechanisms that enable the management of active documents. By manipulating active documents, it is anticipated that the education organization will evolve. It will produce a shareable memory that can be capitalized by the corpus of the organization. In the aeronautical domain, Airbus Training has implemented a procedure used by instructors that enables them to provide experience feedback, i.e., instructors ask for improvements or corrections of flaws in training tools based on the experience they have on these tools. Experience feedback is based on positive or negative experience that is interpreted by training specialists to generate or modify corporate knowledge. A corporate memory of the description of the various pedagogical tools is maintained using this procedure. The main point of such an organization is the optimization of the end product destined for the students. There are several issues that need to be further investigated such as interoperability of cooperation of human and software agents (Bradshaw, 1997), forgetting and standardization of software agents that enhance organizational memory systems capabilities. Three theoretical issues should be better investigated: • Information Technology (IT) shapes new distribution of labor, i.e., whenever possible, a cognitive function analysis of the organization should be carried out to improve the allocation of human and software functions – social repercussions of enterprise automation are not easily controllable. • Context representation, i.e., people productions should be shared within the organization; the way to reach this goal is to implement a situated cognition approach 2A major issue is the interoperability of software developed in a specific software environment. Software agents

should be platform-independent. Furthermore, the combination of object-oriented techniques (a software agent is a software object) and component-based software has some essential benefits listed by Rappaport (1995): reuse, extendibility, customization, distributability, and standardization. An example of standardization of agent-based software is given in (General Magic, 1994).

Page 35


instead of a pure goal-driven (top-down) approach. Context representation is useful to better handle information structuring, indexing and retrieval. • Memory models, i.e., a classification of use of static information stores should be performed in order to design them for maximum usability. A careful use of human memory metaphors is likely to help in the definition of an OMS. Knowledge management tools and decision aids should be based on memory models that users understand. This is to improve the affordance of an OMS. Three human factors issues need to be considered: • Usability of organizational memory systems, i.e., OMS-specific usability methods and techniques should be developed to test attributes such delegation, control, trust, information management, forgetting, coordination and cooperation. • Cooperation and coordination, i.e., better understand if the processes of cooperation and coordination need to be adapted to the culture of the organization, and find out new resource management paradigms. These tests should be centered on an acceptable level of activity (vigilance/stress) – sometimes cooperation/coordination requires additional activities that may not be compatible with the organization culture. • Traceability of design decisions, i.e., improve the way individual and group decisions are modeled. Experts should be better modeled in order to take them into account when recording information. A distinction should be made between legal knowledge and design expertise. Both should be remembered but for different reasons. Usually, the former is public, the latter is private. Three technological issues should be better emphasized: • Human-centered design, i.e., technology should not be developed without a cognitive function analysis (agent modeling) that shows the pros and cons of various function allocations. Rapid prototyping and large-size testing are recommended. • Integrated technical documentation, i.e., as in all user interface design, the procedure/interface duality should be analyzed (Boy, 1998). The active document approach is recommended both to master the technology being developed and to involve organization members into the design of the OMS. • Software agent technology, i.e., the evolution of information management through computer networks has promoted three important concepts that are intelligent assistance (performance support), adaptivity to users, and interoperability. By definition, software agents incorporate these three concepts.

Acknowledgments I thank my colleagues Mike Atwood, Jeff Bradshaw, Jonathan Grudin, Norbert Streitz, Bea Zimmerman, and the participants to the Industrial Summer School in HumanCentered Design of Organizational Memory Systems, held in August 1997 in the Southwest of France. They provided me with many comments and suggestions on the content of this chapter that was under construction at that time. I also thank my coworkers at EURISCO Alfred Attipoe, Markus Durstewitz, Martin Hellander, Rachel Israel and David Novick, as well as my Aerospatiale colleagues Brigitte Daniel, JeanPierre Heckmann and René Peltier for insightful discussions on the topic. Special thanks

Page 36


to Rachel. Many thanks to Helen Wilson for many useful and vivid exchanges on the topic.

References Ackerman, M.S. & Malone, T.W. (1990). Answer Garden: A tool for growing organizational memory. Proceedings of the Conference on Office Information Systems, ACM, New York, pp. 31-39. Attipoe, A. & Boy, G.A. (1995). Modeling knowledge and access in corporate distributed information systems. Proceedings of the IJCAI'95 Workshop on Artificial Intelligence in Distributed Information Systems. Berlin, L.M., Jeffries, R., O’Day, V.L., Paepcke, A. & Wharton, C. (1993). KAPTUP: Knowledge acquisition for preservation of tradeoffs and uderlying rationale. Proceedings of INTERCHI’93 Conference on Human Factors in Computer Systems, ACM, New York, pp. 23-30. Bisseret, A., Figeac-Letang, C. & Falzon, P. (1988). Modeling opportunistic reasoning: the cognitive activity of traffic signal setting technicians. Research Report no. 898, INRIA, Roquencourt, France. Boehm, B.W. (1988). A spiral model of software development and enhancement. IEEE Computer, Vol. 21, no. 5, May, pp. 61-72. Boy, G.A. (1987). Operator Assistant Systems. International Journal of Man Machine Studies. 27, pp. 541-554. Boy, G.A. (1991a). Intelligent Assistant Systems. Academic Press, London. Boy, G.A. (1991b). Computer Integrated Documentation. NASA Technical Memorandum. Ames Research Center, California. Boy, G.A. (1991c). Indexing Hypertext Documents in Context. In the Proceedings of Hypertext'91. San Antonio, Texas, December. Boy, G.A. (1993). HCI in the cockpit. Keynote Lecture. INTERCHI’93. Amsterdam, The Netherlands. Boy, G.A. (1995). Supportability-Based Design Rationale. In the Proceedings of the 6th IFAC/IFIP/IFORS/IEA Symposium on Analysis, Design and Evaluation of Man-Machine Systems, MIT, Boston, USA - July 27-29, 1995 Boy, G.A. (1996). The Group Elicitation Method: An Introduction. Proceedings of the European Knowledge Acquisition Workshop (EKAW’96), Springer Verlag, Berlin. Boy, G.A. & L’Ebraly, H. (1994). Modèle de tâche et representation de l’expertise pour la conception de dispositifs d’interaction homme-machine. Exemple d’application en aéronautique. Rapport EURISCO no. T-94-009-VI. Boy, G.A. (1997). The Group Elicitation Method for Participatory Design and Usability Testing. Interactions. March-April issue, Vol. IV.2, pp. 27-33. Boy, G.A. & Streitz, N. (1997). Hypertext/hypermedia for organizational memory systems. Panel at the ACM Conference on Hypertext. In the Proceedings of Hypertext’97, ACM Press, New York. Boy, G.A. (1998). Cognitive Function Analysis. Ablex Pubishing Corporation, Greewich, CT. To appear. Boy, G.A. & Bertuccio, E. (1997). Interaction complexity testing using the knowledge block representation. EURISCO publication to appear. Bradshaw, J. (1997). Software agents. AAAI/MIT Press, Cambridge, MA. Broigniez, T. (1996). Application d’une méthode de conception interactive pour l’amélioration d’interfaces. Exemple d’application en aéronautique. Rapport EURISCO no. RS-96-006-TB.

Page 37


Card, S.K., Moran, T.P. & Newell, A. (1983). The Psychology of Human Computer Interaction, Lawrence Erlbaum Ass., Pub., Hillsdale Carroll, J.M., Alpert, S.R., Karat, J., Van Deusen, M.S. & Rosson, M.B. (1994). Raison d’Etre: Capturing design history and rationale in multimedia narratives. Proceedings of the ACM CHI’94 Conference. (Boston, April 24-28). New York: ACM Press, pp. 192-197, 478. Carroll, J.M. & Moran, T. (1991). Special Issue on Design Rationale. Human-Computer Interaction, Lawrence Erlbaum Associate, Inc., Volume 6, pp. 281-318. Carroll, J.M. & Rosson, M.B. (1991). Deliberated Evolution: Stalking the View Matcher in Design Space. Human-Computer Interaction, Lawrence Erlbaum Associate, Inc., Volume 6, pp. 281-318. Conklin, J. & Begeman, M. (1988). gIBIS: A hypertext tool for exploratory policy discussion. ACM Transaction on Office Information Systems, Vol. 6; No. 4, October, pp. 303-331. Conklin, E.J. & Burgess Yakemovic, S.C. (1991). A process oriented approach to design rationale. Human-Computer Interaction, 6, pp. 357-391. Curtis, B., Kellner, M.I. & Over, J. (1992). Process modeling. Communications of the ACM, 35(9), pp. 75-90. Curtis, B., Krasner, H. & Iscoe, N. (1988). A field study of the software design process for large systems. Communications of the ACM, 31, pp. 1268-1287. Davis, P.J. & Hersh, R. (1986). Descartes’ Dream. The World according to Mathematics. Harcourt Brace Jovanovich, Inc. De Brito, G., Pinet, J. & Boy, G. (1997). L’utilisation des checklists dans un cockpit d’avion de nouvelle génération. EURISCO Technical Report no. T-97-042. Durstewitz, M.D. (1994). Reuse of Experience for Industrial Applications - A formal approach, CSI'94, Cognitive Science in Industry, Luxembourg, 28-30 September. Festinger, L. (1957). A thery of cognitive dissonance. Evanston, Row, Peterson. Fischer, G. (1990). Communication requirements for cooperative problem solving systems. International Journal of Information Systems, 15(1), pp. 21-36. Fischer, G., Grudin, J., Lemke, A.C., McCall, R., Ostwald, J. & Shipman, F. (1992). Supporting indirect, collaborative design with integrated knowledge-based environments. Human-Computer Interaction, Lawrence Erlbaum Associate, Inc., Volume 7, pp. 281-314. Fischer, G., Lemke, A.C., McCall, R. & Morch, A.I. (1991). Making argumentation serve design. Human-Computer Interaction, Lawrence Erlbaum Associate, Inc., Volume 6, pp. 393-419. Fisher, D. (1987). Knowledge acquisition via incremental conceptual clustering. Machine Learning, 2, pp. 139-172. Flores, F., Graves, M., Hartfield, B. & Winograd, T. (1988). Computer systems and the design of organizational interaction. ACM Transactions on Office Information Systems, Vol. 6, No. 2, pp. 153-172. Floridi, L. (1995). The Internet and the future of organised knowledge. UNESCO Conference, Paris, France. March 14-17. Gaines, B. (1989). Social and cognitive processes in knowledge acquisition. Knowledge Acquisition, 1(1), pp. 39-58. Grudin, J. (1988). Why CSCW application fail: Problems in the design and evaluatoin of organizational interfaces. Proceedings of CSCW’88 Conference on Computer-Supported Cooperative Work, ACM, New York, pp. 85-93.

Page 38


Guidon, R., Krasner, H. & Curtis, B. (1987). Breakdowns and processes during the early activities of software design by professionals. In G. Olson, S.S. Sheppard & E. Soloway (Eds.), Empirical Studies of Programmers: Second Workshop. Ablex Publishing Corporation, Norwood, NJ. Haake, S.H. & Wilson, B. (1992). Supporting collaborative writing of hyperdocuments in SEPIA. Proceedings of the ACM Conference on Computer-Supported Cooperative Work, CSCW’92. ACM Press, New York. Hayakawa, S.I. (1990). Language in Thought and Action. Fifth Edition. Harcourt, Brace & Jovanovich, New York. Hollnagel, E. (1989). Performance improvement through cognitive task analysis. ESA/ESTEC Workshop: A task oriented approach to human factors engineering. Nordwijk, The Netherlands. November 21-23. Hollnagel, E. & Woods, D.D. (1983). Cognitive systems engineering: New wine in new bottles. International Journal of Man-Machine Studies, 18, pp. 583-600. Hollnagel, E. (1993). Human Reliability Analysis: Context and Control. Academic Press, London. Hutchins, E. (1995). How a Cockpit Remembers its Speeds. Cognitive Science, 19, pp. 265288. Hutchins, E. (1995). Cognition in the wild. MIT Press, Boston, USA. Israel, R. (1995). L’oubli de l’oubli. Mémoire de DEA. Sciences de l’Homme et Technologie, Université Technololgique de Compiègne, France. Israel, R. (1998). Valorisation d’expérience et mémoire d’entreprise: aspects organisationnels. EURISCO Technical Report T-98-048, Janvier. Israel, R. (1998). Une méthodologie pour la mémoire collective: le cas Concorde. EURISCO Technical Report T-98-052, In press. Karsenty, L., Bigot, V. & de Brito, G. (1995). L'utilisation des procédures écrites dans un avion de nouvelle génération. EURISCO Technical Report no. T-95-023. Kolodner, J.L. Jona, M.Y. (1991). Case-based reasoning: An overview. Technical Report 15, Institute for the Learning Sciences. Krasner, H., Terrel, J., Linehan, A., Arnold, P. & Ett, W.H. (1992). Lessons learned from a software process modeling system. Communications of the ACM, 35(9), pp. 91-100. Krishnakumar, S. (1996). Conception d’une interface homme-machine et d’un document actif en aéronautique embarquée. Mémoire de Mastère TAS. ENSAE-EURISCO. Lave, J. (1988). Cognition in practice. Cambridge University Press, Cambridge, England. MacLean, A., Young, R.M., Bellotti, V.M.E. & Moran, T.P. (1991). Questions, options and criteria: Elements of design space analysis. Human-Computer Interaction, Lawrence Erlbaum Associate, Inc., Volume 6, pp. 201-250. Mallery, J.C., Hurwitz, R. & Duffy, G. (1987). Hermeneutics: From Textual Explication to Computer Understanding? Published in The Encyclopedia of Artificial Intelligence, Stuart. C. Shapiro, editor, John Wiley &Sons, New York. Mantovani, G. (1996). Social context in HCI: A new framework for mental models, cooperation, and communication. Cognitive Science, 20, pp. 237-269. McQuail, D. (1987). Mass communication theory. An Introduction. Newbury Park, Sage. Mead, G.H. (1963). L’esprit, le soi et la société. Presses Universitaires de France, Paris. Minsky, M. (1985). The Society of Mind. Touchstone Simon and Schuster. Monod, J. (1970). Le hasard et la nécessité. Le Seuil, Paris. Muller, M. (1991). Participatory design in Britain and North America: Responding to the « Scandinavian Challenge ». In Reading Through Technology, CHI’91 Conference Proceedings. S.P. Robertson, G.M. Ohlson and J.S. Ohlson Eds. ACM, pp. 389-392. Page 39


Newell, A. (1990). Unified Theories of Cognition. Harvard University Press, Cambridge, MA. Nielsen, J. (1993). Usability engineering. Academic Press. London. Norman, D.A. (1986). Cognitive Engineering. In User-Centered System Design., D. Norman & S. Draper (Eds.), Hillsdale, NJ: Lawrence Erlbaum Associate, pp. 31-61. Norman, D. A. (1992). Cognitive artifacts. In J. M. Carroll (Ed.), Designing Interaction: Psychology at the Human-Computer Interface. (pp. 17-38). Cambridge: Cambridge University Press. Norman, D.A. (1992). Turn signals are the facial expressions of automobiles. AddisonWesley, Reading, MA. Norman, D.A. (1993). Things that make us smart. Addison-Wesley Publishing Company, Reading, MA. Oatley, K.G. (1977). Inference, navigation, and cognitive maps. In Thinking - Readings in Cognitive Science, P.N. Johnson-Laird & P.C. Wason, Cambridge University Press, Cambridge, UK, pp. 537-547. Peirce, C.S. (1931-1958). Collected Papers, 8 vols. Edited by C. Hartshorne, P. Weiss, and A. Burks. Cambridge, Massachusetts: Harvard University Press. Rappaport, A.T. (1995). Context, Cognition and Intelligent Digital Information Infrastructure. Personal Communication to be published. Rasmussen, J. (1986). Information Processing and Human-Machine Interaction - An Approach to Cognitive Engineering. North Holland Series in System Science and Engineering, A.P. Sage, Ed. Rogers, W.P. (1986). Presidential Commission on the Space Shuttle Challenger Accident Report. U.S. Government Printing Office: 688-274/58381, June. Rogoff, B. (1984). Introduction: Thinking and learning in social context. In Everyday Cognition: Its Developement in Social Context, Rogoff, B. and Lave J., Harward University Press Pub., Cambridge, MA. Rumbaugh, J., Blaha, M., Permerlani, W., Eddy, F. & Lorensen, W. (1991). Object oriented analysis and design. Prentice Hall International. SAE (1994). Systems Integration Requirements. ARP 4754, System Integration Requirements Task Group, AS-1C, ASD, Draft 43a, March 13, 1994. Schank, R.C. & Jona, M.Y. (1991). Empowering the student: New perspectives on the design of teaching systems. The Journal of the Learning Sciences. Volume 1, Number 1, pp. 7-36. Sheridan. T.B. (1992). Telerobotics, Automation, and Human Supervisory Control. MIT Press, Cambridge. Shipman, F.M. & McCall, R. (1994). Supporting knowledge-base evolution with incremental formalization. Proceedings of CHI’94 Conference on Human Factors in Computer Systems, ACM, New York, pp. 285-291. SIGCUE (1992). Computer supported collaborative learning. Bulletin of the Special Interest Group for Computer Uses in Education (SIGCUE). ACM Press, Spring, Vol. 21#3. Sober, E. (1990). Teleological functionalism: Putting the function back into functionalism. In Mind and Cognition, William G. Lycan (Ed.), Basic Blackwell, Inc., Cambridge, MA. Soloway, E. (1993). Interactive Learning Environments. INTERCHI'93 Tutorial Notes, ACM, Amsterdam. Soloway, E. (1995). Beware, Techies Bearing Gifts. Communications of the ACM, Vol. 38, No. 1, January. Page 40


Suchman, L. (1987). Plans and situated actions: The problem of human-machine communications. Cambridge University Press, New York. Suchman, L. (1993). Situativity and symbols. Response to Vera and Simon. Cognitive Science. Vera, A. & Simon, H. (1993). Situated actions: A symbolic interpretation. Cognitive Science. Visser, W. (1989). More or Less Following a Plan during Design: Opportunistic deviations in Specification. International Journal of Man-Machine Studies, Special Issue on Empirical Studies of Programmers. Also INRIA Research Report, Roquencourt, BP 105, 78153 Le Chesnay Cedex, France. Winograd, T. & Flores, F. (1986). Understanding computers and cognition-A new foundation for design. Addison-Wesley Pub. Comp., Inc. Reading, MA. Woods, D.D. (1993). Towards a theoretical base for representation design in the computer medium: Ecological perception and aiding human cognition. In J. Flach, P. Hancock, J. Craig, & K. Vicente (Eds.), The Ecology of Human-Machine Systems, Hillsdale, NJ: Lawrence Erlbaum. Yates, F.A. (1966). The art of memory. French translation by Daniel Arasse, 1975, Editions Gallimard, Paris, France.

Page 41