including environments for kitchen design [Fischer, McCall, Morch 1989], ... The learner selected a learning example from the catalog and brought it in the work ...
Learning Opportunities Provided By Domain-Oriented Design Environments
Gerhard Fischer Department of Computer Science and Institute of Cognitive Science University of Colorado, Campus Box 430 Boulder, Colorado 80309 USA
Abstract. Information overload, the advent of high-functionality systems, and a climate of rapid technological change have created new problems and challenges for education and training. New instructional approaches are needed to circumvent the difficult problems of coverage (i.e., trying to teach people everything that they may need to know in the future) and obsolescence (i.e., trying to predict what specific knowledge someone will need in the future). Learning on demand is a promising approach to address these problems for the following reasons: (1) it contextualizes learning by allowing it to be integrated into work rather than relegating it to a separate phase; (2) it lets learners see for themselves the usefulness of new knowledge for actual problem situations, thereby increasing the motivation for learning new skills and information; and (3) it makes new information relevant to the task at hand. In educational settings, learning is often restricted to the solution of well-defined problems; our research, in contrast, focuses on environments that support learning in the context of realistic, open-ended, ill-defined problems. In our environments, learners explore information spaces relevant to a self-chosen task at hand. Learning on demand provides learner-centered alternatives to teacher-centered tutoring systems and it augments open-ended, unsupported learning environments by providing advice, assistance, and guidance if needed in breakdown situations. Creating computational environments in support of learning on demand generates a large number of challenging problems. We have developed integrated, domain-oriented design environments supporting learning on demand that satisfy the following requirements: (1) they allow users to create new artifacts and understand existing ones, (2) they tailor instruction to serve the accomplishment of the task at hand, and (3) they do this without disrupting or otherwise interfering with the task. Keywords. learning on demand, design, domain-oriented design environments, coverage, obsolescence, information overload, making information relevant to the task at hand, critiquing.
1. Introduction Innovative uses of computers in education have focused on two major approaches: intelligent tutoring systems [Anderson, Reiser 1985; Psotka, Massey, Mutter 1988; Polson, Richardson 1988; Wenger 1987] and open learning environments [Boecker, Eden, Fischer 1991; diSessa, Abelson 1986; Wenger 1987]. The strength of intelligent tutoring systems lies in their ability to teach basic concepts and skills of a problem domain. However, the problems presented by tutoring systems are prespecified by the authors of the systems rather than being ill-defined and arising out of real-world contingencies. Thus, tutoring systems, similar to school education, leave it to the learner to relate training to realworld problem situations. Open learning environments avoid the problem of presenting instructional material in a system-controlled order without regard to the learner's situation, but they provide limited support in helping learners detect mistakes or overcome breakdowns [Fischer 1995]. Misconceptions may accumulate into chains, in which each later misconception is based on a previous one. Learners get trapped on suboptimal plateaus because they fail to discover the knowledge needed to make better use of their tools and to create better artifacts. For many years we have been trying to find new ways of creating computer-based learning environments, combining the strengths of both approaches while at the same time eliminating some of the weaknesses. Our goal has been to integrate working and learning and to support learning on demand.Domain-oriented design environments (based on a multi-faceted architecture [Fischer 1992; Fischer, McCall, Morch 1989]) have proven to be powerful and versatile environments for learning that (1) address the limitations of intelligent tutoring systems and open learning environments, and (2) provide multiple learning opportunities [Weyer 1987]. Pursuing this line of research, we have emphasized AI research directions and techniques that augment and complement human intelligence with rich computational environments, including critics, agents, assistants, adaptable and adaptive tools, information access and information delivery mechanisms [Fischer, Nakakoji 1992; Stefik 1986]. This paper (1) discusses the problems that are the focus of our research, (2) describes conceptual frameworks that address these problems, (3) illustrates prototype systems that instantiate these frameworks, and (4) concludes by giving an assessment of these efforts and by describing plans for future research.
2. Problems Addressed by Our Work This section briefly describes some of the problems that educational uses of computers need to address. Deschooling Learning Learning should be part of living, a natural consequence of being alive and in touch with the world, and not a process separate from the rest of life [Clancey 1992; Lave, Wenger 1991; Resnick 1989]. What learners need, therefore, is not only instruction but access to the world (in order to connect the knowledge in their head with the knowledge in the world [Norman 1993]) and a chance to play a meaningful part in it. Education should be a distributed
lifelong process by which one learns material as one needs it. School learning and work place learning [Scribner, Sachs 1990] need to be integrated. We refer to work place learning not as it is currently practiced (e.g., companies imitating school learning by sending their employees to decontextualized classrooms), but as it could be or should be. Examples include apprenticeship type relationships such as internships for doctors and PhD studies. In such learning situations, problems are not given, but need to be framed. Collaboration is critical and learning is firmly integrated with working. Figure 1 compares school and work place learning along a number of dimensions. Schools
Workplace
EMPHASIS ON:
“basic” skills
education embedded in ongoing work activities
POTENTIAL DRAWBACKS:
decontextualized, not situated
important concepts are not encountered
PROBLEMS:
given
constructed
NEW TOPICS: DEFINED BY:
curricula
arise accidentally from work situations
STRUCTURE:
pedagogic or “logical”structure
work activity
ROLES:
expert ´ novice model
reciprocal learning
TEACHERS/ TRAINERS:
expound subject matter
engage in work practice
MODE:
instructionism (knowledge absorption)
constructionism (knowledge construction)
Figure 1: Integration of School and Work Place Learning
Information Overload. Information overload, the advent of high-functionality systems, and a climate of rapid technological change have created new problems and challenges for education and training. In our information-rich society, humans must cope with a tremendous amount of information [Norman 1993]. New information technologies must take into account not only the producers of that information but also its consumers. Providing more information will not make computers more helpful; instead, there is a need for systems that help us attend to the most useful, most interesting, or most valuable information [Simon 1981], [Fischer 1991]. New instructional approaches are called for to solve the difficult problems of coverage (trying to teach people everything that they may need to know in the future) and obsolescence (trying to predict what specific knowledge someone will need in the future). Most traditional approaches to learning are based on a model in which one attends school or college to prepare for the situations that will occur in later life. These approaches are doomed to fail because there is too much to learn and because the content that is learned becomes outdated too quickly.
Beyond Passive Tools. Passive tools do not support mixed-initiative dialogs, i.e., dialogs in which either the learner or the system may initiate an interaction or a change of topic [Carbonell 1970]. Passive tools are often inadequate for the demands of situations in which we envision our systems being used. For example, browsing does not scale up to large information spaces, and passive help systems are of little use if learners do not know that relevant information exists. Information access mechanisms must be combined with information delivery mechanisms [Nakakoji 1993]. Passive tools must be complemented by intelligent agents, such as critics [Fischer et al. 1991]), which act on their own initiative. Production Paradox. Educational approaches must avoid the “production paradox” [Carroll, Rosson 1987], in which learning is inhibited by lack of time and working is inhibited by lack of knowledge. Learners must regard the time and effort invested in learning as immediately worthwhile for the task at hand—not as valuable merely for some putative long-term gain. Insufficient Attention to Motivation. Many approaches to learning have paid little attention to motivation. Ignoring motivational issues [Csikszentmihalyi 1990], [Norman 1993] will make even the most technologically sophisticated efforts fail. To motivate learners, we advocate the following: (1) learning must be actively desired and controlled by the learner, (2) learners must be able to see the immediate benefit of learning something new to their current working situation, (3) environments must be intrinsically motivating so users can achieve large effects with reasonably small efforts, and (4) environments must allow learners to develop a functioning artifact from a design idea in a short time, so they can spend more time on things they really want to do, invest less ego in a particular product, and therefore be more willing to criticize it and change their designs.
3. Conceptual Frameworks This section describes conceptual frameworks addressing the problems of major concern to us, as outlined above. Learning on Demand. By putting the choice of tasks and goals under the control of the learner, learning on demand can contribute to the goal that learning should simply be a natural consequence of being alive and in touch with the world rather than a process separate from the rest of life. Learning on demand provides and exploits opportunities and support for helping people take charge of their own learning. As a consequence, ensuring coverage is impossible, and obsolescence cannot be avoided. We strongly believe that learning on demand should not be restricted to quick fixes of some problems. It should provide the starting point of learning episodes that in some cases may allow learners to get a job done, but in other cases may provide the global motivation to learn new concepts and new domains. Learning on demand raises many important questions in reconsidering (1) the role of learning over the course of an individual's life, (2) the organization of learning opportunities within our society in the future, and (3) the technological environments needed to support new forms of education. Design. The task domains that we have chosen for our research efforts belong to the “sciences of the artificial” [Simon 1981]. Design as an intellectual activity creates objects by successive refinement through a process of construction, breakdown, interpretation, and reflection [Schoen 1983], [Rittel 1984]. The design of artifacts is in most cases an “ill-defined” or “wicked” problem [Rittel 1984], creating the following dilemma: (1) one cannot gather information meaningfully
unless the problem is understood, and (2) one cannot understand the problem without having a concept of the solution in mind and without information about that solution. In real life, as opposed to typical classrooms, problems and tasks themselves are moving targets requiring an integration of problem setting and problem solving [Lave 1988]] the work in progress suggests ways to proceed, and the development of a solution causes the understanding of the task to grow and change. Design can best be understood and described as reflection in action [Schoen 1983]. Action is governed by a nonreflective thought process and proceeds until it breaks down. A breakdown occurs when the designer realizes (being assisted by the “back-talk” of the situation [Schoen 1983]) that nonreflective action has resulted in unanticipated consequences. Reflection is used to repair the breakdown, and then situated action continues. Critiquing [Fischer et al. 1991] plays an important role in this process by increasing the back-talk from the situation and by providing access paths to relevant information. Using Examples and Case-Based Reasoning. Examples can play a crucial role in design. Lewis and Olson [Lewis, Olson 1987] observed that the productivity of casual programmers can be increased by their adapting examples through analogy. Examples have been the focus of studies in learning LISP[Pirolli, Anderson 1985] as well as other areas [Chi et al. 1989]. The catalog in our systems [Redmiles 1992] serves as a library of designs, which users can access and extend. The use of examples is similar to work in case-based reasoning, in which previous solutions or cases form the basis of new plans or task solutions [Riesbeck, Schank 1989]. Making Information Relevant to the Task at Hand. High-functionality systems confront us with too much information. The challenge is to make the information relevant to the task at hand—that is, delivering the right knowledge, in the context of a problem or a task, at the right moment for a human professional to consider [Fischer, Nakakoji 1991]. Making information relevant to the task at hand poses many challenges for the design of interactive computer systems, and it also sets computer systems well apart from other technologies such as printed material. Integrating Knowledge in the Head and Knowledge in the World. Learning, understanding, and working in the real world rely heavily on integrating knowledge in the head and knowledge in the world [Lave 1988; Norman 1993]. Rather than emphasizing closed-book exams and toolfree performance, cognitive artifacts are used to expand people's mental power. There is less insistence on symbolic performances. Instead actions are intimately connected with objects and events in the world, and people use the objects themselves (rather than abstractions) in their reasoning.
4. Examples of Learning Opportunities Supported by Domain-Oriented Design Environments To address the problems and to support the conceptual framework outlined in the two previous sections, we have developed a domain-independent multifacted architecture for design environments (see Figure 2). By populating all the components with domain-specific knowledge, we have created a variety of different domain-oriented design environments over the last years, including environments for kitchen design [Fischer, McCall, Morch 1989], computer network design [Fischer et al. 1992], graphic arts design [Eisenberg, Fischer 1994], and lunar habitat design [Stahl 1993].
Construction Construction Analyzer
Simulation
Argumentation
Catalog Explorer
Specification
Argumentation Illustrator
Catalog Catalog Explorer
Figure 2: The Multifaceted Architecture HYDRA H YDRA (as a domain-independent architecture) supports five essential components of domainoriented design environments. The links between the components are crucial for the synergy of the environment. Through domain-oriented instantiation, H YDRA provides the foundation for design creation tools and information repositories reflecting the real-world contexts of design processes.
Scenario. Here we use the JANUS kitchen-design system [Fischer, McCall, Morch 1989], [Fischer, Nakakoji 1992] as an “object to think with” to illustrate the multiple learning opportunities supported by domain-oriented design environments (see Figure 7 for a summary). In this scenario, rather than building a kitchen floor plan by composing parts out of the palette, the learner tries to understand some of the principles behind good kitchen design by looking at one of the learning examples contained in the catalog (see Figure 3). Other learners might have started with a partial specification (e.g., that they are interested in a safe eat-in kitchen because they have small children) and the system, using CATALOG-E XPLORER, would have located for them the examples corresponding most closely to their partial specification (see Figure 2 and [Fischer, Nakakoji 1992]). Still other learners might explore different classes of kitchens (e.g., L-shaped, U-shaped, island, one-wall, etc.). The catalog component may also be simply browsed (to explore “classic” examples and components), and arguments motivating particular design choices associated with catalog entries may be examined independently. Having brought the learning example from the catalog into the working area, the critiquing component can be used to analyze it. The shortcomings of the design are displayed in the message pane (see Figure 3). In our scenario, the learner is most interested in exploring the argumentation behind the message that “the stove should be away from the window.” By clicking on this
Figure 3: Exploring Learning Examples from the Catalog The learner selected a learning example from the catalog and brought it in the work area. The CONSTRUCTION-ANALYZER identifies several problematic design choices. The learner decides to investigate the issue “stove not away from window.”
message, the system brings up the argumentation (see Figure 4) surrounding this particular design issue, showing arguments in favor and against it. By selecting the “Show Example” command, the learner uses the ARGUMENTATION-ILLUSTRATOR (see Figure 2) to contextualize the arguments with a specific example that is displayed in the upper left corner of the screen image. Critiquing. The critiquing approach [Fischer et al. 1991]is an effective way to use computational knowledge bases to aid users in their work and to support learning on demand. In many design situations, artifacts do not speak for themselves. To address this problem, we augmented our environments with computational critics identifying breakdowns [Fischer 1995; Nakakoji, Fischer 1995], which might have remained unnoticed without them. These breakdowns provide learning opportunities for designers by supporting them to reframe problems, to attempt alternative design solutions, and to explore relevant background knowledge.
Figure 4: Exploring the Rationale Behind the Example The hypermedia-based argumentation component displays the rationale behind the critiquing messages. By selecting the “Show Example” command, the ARGUMENTATION-ILLUSTRATOR contextualizes the argumentation in the context of a specific example retrieved from the catalog.
The early work of our research group focused on building and evaluating stand-alone critiquing mechanisms [Fischer 1987]. Critical analyses of these and other systems [Fischer et al. 1991], combined with empirical evaluations, led us to realize that the challenge in building critiquing systems is not simply to provide design feedback: the challenge is to say the right thing at the right time [Burton, Brown 1982]. What is needed is a strategy that: (1) alerts learners to problematic solutions, (2) avoids unnecessary disruptions, and (3) allows learners to control the critic's intervention strategy. The shortcomings in early critiquing systems that hindered their ability to say the right thing at the right time were: (1) lack of domain orientation, (2) insufficient facility for justifying critic suggestions, (3) lack of an explicit representation of the user's goals, (4) no support for different individual perspectives, and (5) timing problems with critic intervention strategies [Fischer et al. 1993].
All of our critic mechanisms use a production system style of knowledge representation, consisting of condition and action clauses plus links into the argumentation system. The condition clause checks whether a certain situation exists in the current design construction. The action clause notifies the designer that a particular situation has been detected. Each critic rule is linked to a particular issue in the argumentation base. The designer can view the critic's associated argumentation by selecting the initial notification message, which creates an entry-point into the hypermedia system (see Figure 3). Such argumentative explanations help designers determine why the design situation identified by the critic message may be significant or problematic. Designers can optionally explore the issue-base or select an issue and an associated answer in the argumentation and request to see a positive example or a counter-example from the catalog of designs (see Figure 4). By embedding critics into design environments, three different mechanisms (generic, specific, and interpretive) can be supported. These mechanisms differ from one another in how they determine which set of critic rules should be enabled. Generic critics evaluate the construction situation based on an assumption that a designer wants to design a “good” kitchen as the default design requirement. Specific critics evaluate the construction situation for compliance with the partial specification. Specific critics are constructed dynamically from the partial specification to reflect current design goals and are specialized by adding requirements to reflect the current specification (see Figure 5 and [Nakakoji 1993]). Critiquing criteria are tied to a representation of the partially articulated goals of a specific design project. Interpretive critics [Stahl 1993] provide support for design as a hermeneutic process [Winograd, Flores 1986]. They allow designers to interpret the design situation according to their own interests. By being associated with design perspectives rather than with partial specifications, perspectives provide a mechanism for creating, managing, and selectively activating different sets of critics and design knowledge, such as spatial relations, domain distinctions, palette items, and argumentation. Figure 6 describes these different critics for a specific design issue (further details of this example are described in [Fischer et al. 1993]). Learning from Examples. As illustrated with the scenario, the catalog, by being integrated into the design environment, provides a number of different learning opportunities: (1) it supports design by modification (rather than by composition using components from the palette; see Figure 3); (2) it provides learning examples for exploring important domain issues and design rules in context; (3) it shows particularly successful designs used as standard examples by the design community in a certain discipline; and (4) it supports contextualized tutoring. Contextualized tutoring allows learners to explore relevant domain concepts in depth. Contrary to the situation in tutoring systems, the initiative for in-depth exploration is not preprogrammed into the system but is derived from the interaction of the learner with the design environment. Contextualized tutoring requires an indexing mechanism that links examples with the specific domain concepts they illustrate (concepts such as safety, flammability, ease of cleaning, etc.; see Figure 4). Making Information Relevant to the Task at Hand. Simon [Simon 1981] argues convincingly that a design representation suitable for a world in which the scarce factor is information may be exactly the wrong one for a world in which the scarce factor is attention. When presenting people with information, the primary concern is to present items that are relevant to the task at hand. Design environments are complex systems in which finding relevant information may be difficult.
Figure 5: The Specification Component Designers can select answers presented in the Questions window. The summary of currently selected answers appears in the Current Specification window. Each answer is accompanied by a slider that allows designers to assign a weight representing the relative importance of the answer.
Information needs and resulting queries or browsing behavior always arise from the larger context of a problem that needs to be solved. Stand-alone access mechanisms, such as queries and browsing, often represent a decontextualized information need. Although context can in principle be explicitly stated in the query, this greatly complicates the query formation process. Design environments attack these problems by supporting both information delivery and access [Nakakoji 1993]. Information delivery is a mechanism by which a system detects the potential for an information need and presents relevant information for the user's perusal. Information access is the
Generic Critic: • specification: none available • rule: the dishwasher should be to the left of the sink • rationale: provides good work flow for right-handed people & a large majority of the population is right-handed • perspective: none available Specific Critic: • specification: the cook is left-handed • rule: the dishwasher should be to the right of the sink • rationale: provides good work flow for left-handed people • perspective: none available Interpretive Critic: • specification: the cook is left-handed • rule: the dishwasher should be to the left of the sink • rationale: interpreted with a resale perspective, stating that the design should not violate basic default assumptions • perspective: resale perspective Figure 6: A Specific Example Comparing Generic, Specific and Interpretive Critics The specific design issue illustrated by the three critiquing rules shown is how dishwasher and sink should be positioned with respect to each other.
user-initiated elicitation of information given a perceived information need that must be refined in order to be satisfied. Taking advantage of the presentation of an evolving artifact in the design environment, our systems partially identify the designer's task at hand from a partial requirement specification and design construction situation. They identify the designer's information needs, deliver the information relevant to the task at hand, and provide background context for the subsequent information access. A problem for effective delivery mechanisms is that, although they may be contextualized, they must work with incomplete knowledge of users' intentions. Delivery techniques may misinterpret users' goals because they lack some important contextual information. Although the mechanisms for delivery can be designed and tailored for minimum disruption, a conflict will always arise between the need to inform users and the desire not to inundate them with unnecessary messages. To address these problems, domain-oriented design environments assist learners in the identification and resolution of information needs by supporting a representation of the task at hand, expressed as a partial requirement specification (see Figure 5) and a construction situation (see Work Area window in Figure 3). This representation is used to reduce the information space to the possibilities within the current design task. Delivery mechanisms such as critics initiate an information seeking process by placing users in an information space relevant to their task at hand (the argumentation displayed in Figure 4 is relevant to the placement of stove and window).
Access mechanisms [Fischer, Nakakoji 1992] are then used to help users explore relevant information. Summary. Figure 7 summarizes the multiple learning opportunities provided by domain-oriented design environments. Learners have considerable control using our environments (e.g., they can disable the critiquing mechanisms, they can ignore the argumentation, and they can engage in tutorial episodes). Design environments, rather than being only passive tools or repositories of information, have evolved into agent-like systems cooperating with learners in searching, interpreting and creating knowledge [Eisenberg, Fischer 1994]. The choice of tasks and goals must be under the control of the learner, not the system. The multiple learning opportunities that are provided must pay attention to the following requirements: The environment must be able to situate learning, allow situations to ``talk back,'' support reflection-in-action, identify the instructional information relevant for tasks at hand, turn breakdowns from disasters into opportunities for learning, and enable learning processes not to disrupt or interfere with solving a problem.
LEARNING OPPORTUNITIES SUPPORTED BY discover and explore relevant domain distinctions at a problemoriented level
domain-orientation of the design environment
notice impasses and experience breakdowns
C ONSTRUCTION-ANALYZER
investigate information relevant to C ATALOG-EXPLORER the task at hand contextualize information
ARGUMENTATION -ILLUSTRATOR
use on demand (the environment makes the information available)
use the world to complement knowledge in the head
learn on demand
increase knowledge in the head and remember access paths to relevant information
learn fundamental domain concepts
contextualized tutoring
deepen the understanding by integrating structural and functional properties
simulation
Figure 7: Summary of Learning Opportunities Supported by Design Environments
5. Assessment This section briefly assesses our efforts (1) by illustrating its relationship to new cognitive science approaches toward learning, (2) by mentioning the contributions of AI techniques used in our
approach as well as the challenges created for AI, and (3) by describing how evaluation efforts have driven our system-building efforts. An Assessment of Our Efforts Based on New Cognitive Science Approaches Toward Learning Current cognitive theories emphasize that learning is a process of knowledge construction, not one of knowledge recording or absorption [Resnick 1989], [Papert 1993]. In our environments, this aspect is supported by the active engagement of the learner. Successful learners elaborate and develop self-explanations [Chi et al. 1989] that extend the information in texts or other instructional materials. Learning is knowledge dependent; people use current knowledge to construct new knowledge and to restructure existing knowledge [Kintsch 1988]. This aspect is supported by the argumentation and catalog and the access mechanisms provided by the C ONSTRUCTION-ANALYZER and CATALOG-EXPLORER. Learning is also highly tuned to the situation in which it takes place [Lave 1988],[Suchman 1987]. No amount of knowledge of principles suffices to account for or guarantee the success of action in real-world problem situations. This challenges many of the basic assumptions about learning general problem-solving skills in a decontextualized way. Contextual elaboration is needed to devise specific courses of action in order to go beyond general procedural prescriptions. This aspect is supported by the contextualization achieved by the ARGUMENTATION-ILLUSTRATOR (the Catalog Example window in Figure 4 contextualizes the rule that stoves should be away from windows with a concrete example) and by the mechanisms making information relevant to the task at hand. Challenges in Creating Instrumental Computational Environments in Support of Learning on Demand. System-building efforts supporting learning on demand face the challenge of creating a system that (1) can relinquish control of task selection yet maintain knowledge of users' goals, plans, and background knowledge; and (2) can function effectively in large solution spaces. Our work addresses these problems by representing specific problem domains with design environments. A partial understanding of the task at hand (as captured by the domain-orientation, by a specification and a construction) allows the system (1) to obtain a partial shared understanding of the users' goals and intention, and (2) to prioritize information spaces in support of learning on demand. Domain-oriented design environments raise a large number of interesting questions about (1) knowledge acquisition (our current approach is based on creating a seed, which evolves and grows through use and a subsequent reseeding process [Fischer et al. 1994]); (2) knowledge representation (our current systems integrate frame-based, rule-based, case-based, and hypermedia-based representations [Fischer, Nakakoji 1992]); and (3) knowledge utilization (we support incremental query formulation, allowing users to initiate their own searches as well as delivering information to them [Nakakoji 1993]). The details of these techniques are described in the mentioned publications. Evaluation. Our work has proceeded as cycles of “design—assessment/evaluation—redesign.” As mentioned before, the evaluation of our early critiquing systems [Fischer et al. 1991] led to the need for embedding them in design environments. The recognition that design is an argumentative process [Rittel 1984] demonstrated the need to back up the critiquing messages with argumentation. Argumentation itself needed to be contextualized through examples. The growing amount of information in our design environments made browsing infeasible and required
components for information delivery and access. In the past, our evaluations were done primarily in controlled experiments rather than in more naturalistic settings. We are currently developing an undergraduate course that will allow us to test some of our current working hypotheses. We expect to find that learning within our design environments will cause major improvements with respect to motivation, fragmentation of knowledge, retention of knowledge, and making learners active participants in the learning process. Observations of learners in our environments in the past indicated that they took advantage of the multiple learning opportunities. However, we intend to carefully investigate possible limitations of our approach, such as: (1) the acquisition of certain essential skills should not be deferred until they are needed, because the time to learn them may be not available or the environment may be too dangerous for safe learning processes; (2) learning on demand is task driven and therefore may be limited to exposing users to isolated pieces of knowledge while providing only limited support for learning essential principles; (3) users may encounter difficulties in decontextualizing knowledge so that it can be used in new settings; and (4) whereas learning on demand may be well suited for evolutionary extensions of a knowledge base, it may not support substantial restructuring, because the additional features learned occur only in the neighborhood of what learners already know.
6. Conclusions Reconceptualizing learning as a lifelong process must be a guiding principle for rethinking our approaches to education in school and work place environments. The College of Engineering at the University of Colorado is in the process of creating a new "Integrated Teaching Laboratory,"which will provide an ideal environment to assess learning on demand in a broad context. We will continue to collaborate with industrial partners to rethink and reconceptualize the relationship between school and work place learning. We will develop our conceptual frameworks by rooting them in new cognitive science approaches to learning, and we will create technological environments that will provide the necessary support to make learning on demand a reality.
Acknowledgments The author would like to thank the members of the Human-Computer Communication group at the University of Colorado who contributed substantially to the conceptual framework and the systems discussed in this paper. The research was supported by (1) the National Science Foundation under grants No. IRI-9015441 and MDR-9253245, (2) NYNEX Science and Technology Center (White Plains, N.Y.), and (3) Software Research Associates, Inc. (Tokyo, Japan).
References [Anderson, Reiser 1985] J.R. Anderson and B.J. Reiser, The LISP Tutor, in (eds.), BYTE, 1985, [Boecker, Eden, Fischer 1991] H. Boecker, H. Eden and G. Fischer, Interactive Problem Solving Using Logo, Lawrence Erlbaum Associates, Hillsdale, NJ, 1991.
[Burton, Brown 1982] R.R. Burton and J.S. Brown, An Investigation of Computer Coaching for Informal Learning Activities, in J.S.B. D.H. Sleeman (eds.), Intelligent Tutoring Systems, Academic Press, London New York, 1982, [Carbonell 1970] J.R. Carbonell, Mixed-Initiative Man-Computer Instructional Dialogues, Report 1971, May, 1970 1970. [Carroll, Rosson 1987] J.M. Carroll and M.B. Rosson, in J.M. Carroll (eds.), Paradox of the Active User Interfacing Thought: Cognitive Aspects of Human-Computer Interaction, The MIT Press, Cambridge, MA, 1987, [Chi et al. 1989] M. Chi, M. Bassok, M. Lewis, P. Reimann and R. Glaser, Self-Explanations: How Students Study and Use Examples in Learning to Solve Problems, in (eds.), Cognitive Science, 1989, [Clancey 1992] W.J. Clancey, Representations of Knowing: In Defense of Cognitive Apprenticeship, in (eds.), The Journal of Artificial Intelligence in Education, 1992, [Csikszentmihalyi 1990] M. Csikszentmihalyi, Flow: The Psychology of Optimal Experience, HarperCollins Publishers, 1990. [diSessa, Abelson 1986] A. diSessa and H. Abelson, Boxer: A Reconstructible Computational Medium, in (eds.), Communications of the ACM, 1986, [Eisenberg, Fischer 1994] M. Eisenberg and G. Fischer, Programmable Design Environments: Integrating End-User Programming with Domain-Oriented Assistance, in (eds.), Human Factors in Computing Systems, CHI’94 Conference Proceedings (Boston, MA), 1994, [Fischer 1987] G. Fischer, A Critic for LISP, in J. McDermott (eds.), Proceedings of the 10th International Joint Conference on Artificial Intelligence (Milan, Italy), Morgan Kaufmann Publishers, Los Altos, CA, 1987, [Fischer 1991] G. Fischer, Supporting Learning on Demand with Design Environments, in L. Birnbaum (eds.), Proceedings of the International Conference on the Learning Sciences 1991 (Evanston, IL), Association for the Advancement of Computing in Education, Charlottesville, VA, 1991, [Fischer 1992]
G. Fischer, Domain-Oriented Design Environments, in (eds.), Proceedings of the 7th Annual Knowledge-Based Software Engineering (KBSE-92) Conference (McLean, VA), IEEE Computer Society Press, Los Alamitos, CA, 1992, [Fischer 1995] G. Fischer, Turning Breakdowns into Opportunities for Creativity, in (eds.), Knowledge-Based Systems, Special Issue on Creativity and Cognition, 1995, [Fischer et al. 1992] G. Fischer, J. Grudin, A.C. Lemke, R. McCall, J. Ostwald, B.N. Reeves and F. Shipman, Supporting Indirect, Collaborative Design with Integrated Knowledge-Based Design Environments, in (eds.), Human Computer Interaction, Special Issue on Computer Supported Cooperative Work, 1992, [Fischer et al. 1991] G. Fischer, A.C. Lemke, T. Mastaglio and A. Morch, The Role of Critiquing in Cooperative Problem Solving, in (eds.), ACM Transactions on Information Systems, 1991, [Fischer, McCall, Morch 1989] G. Fischer, R. McCall and A. Morch, JANUS: Integrating Hypertext with a Knowledge-Based Design Environment, in (eds.), Proceedings of Hypertext’89 (Pittsburgh, PA), New York, 1989, [Fischer et al. 1994] G. Fischer, R. McCall, J. Ostwald, B. Reeves and F. Shipman, Seeding, Evolutionary Growth and Reseeding: Supporting Incremental Development of Design Environments, in (eds.), Human Factors in Computing Systems, CHI’94 Conference Proceedings (Boston, MA), 1994, [Fischer, Nakakoji 1991] G. Fischer and K. Nakakoji, Making Design Objects Relevant to the Task at Hand, in (eds.), Proceedings of AAAI-91, Ninth National Conference on Artificial Intelligence, AAAI Press/The MIT Press, Cambridge, MA, 1991, [Fischer, Nakakoji 1992] G. Fischer and K. Nakakoji, Beyond the Macho Approach of Artificial Intelligence: Empower Human Designers - Do Not Replace Them, in (eds.), Knowledge-Based Systems Journal, 1992, [Fischer et al. 1993] G. Fischer, K. Nakakoji, J. Ostwald, G. Stahl and T. Sumner, Embedding Computer-Based Critics in the Contexts of Design, in (eds.), Human Factors in Computing Systems, INTERCHI’93 Conference Proceedings, 1993, [Psotka, Massey, Mutter 1988] L.D.M. J. Psotka S.A. Mutter, Intelligent Tutoring Systems: Lessons Learned, Lawrence Erlbaum Associates, Hillsdale, NJ, 1988. [Kintsch 1988] W. Kintsch, The Role of Knowledge in Discourse Comprehension: A Construction-Integration Model, in (eds.), Psychological Review, 1988,
[Lave 1988] J. Lave, Cognition in Practice, Cambridge University Press, Cambridge, UK, 1988. [Lave, Wenger 1991] J. Lave and E. Wenger, Situated Learning, Cambridge University Press, Cambridge, UK, 1991. [Lewis, Olson 1987] C.H. Lewis and G.M. Olson, in E.S. G.M. Olson S. Sheppard (eds.), Can the Principles of Cognition Lower the Barriers of Programming? Empirical Studies of Programmers (Vol. 2), Ablex Publishing Corporation, Lawrence Erlbaum Associates, Norwood, NJ, 1987, [Polson, Richardson 1988] J.J.R. M.C. Polson, Foundations of Intelligent Tutoring Systems, Lawrence Erlbaum Associates, Hillsdale, NJ, 1988. [Nakakoji 1993] K. Nakakoji, Increasing Shared Understanding of a Design Task Between Designers and Design Environments: The Role of a Specification Component, Thesis1993. [Nakakoji, Fischer 1995] K. Nakakoji and G. Fischer, Intertwining Knowledge Delivery and Elicitation: A Process Model for Human-Computer Collaboration in Design, in (eds.), Knowledge-Based Systems Journal, 1995, [Norman 1993] D.A. Norman, Things That Make Us Smart, Addison-Wesley Publishing Company, Reading, MA, 1993. [Papert 1993] S. Papert, The Children’s Machine, Basic Books, New York, 1993. [Pirolli, Anderson 1985] P.L. Pirolli and J.R. Anderson, The Role of Learning from Examples in the Acquisition of Recursive Programming Skills, in (eds.), Canadian Journal of Psychology, 1985, [Redmiles 1992] D.F. Redmiles, From Programming Tasks to Solutions -- Bridging the Gap Through the Explanation of Examples, Ph.D. Dissertation 1992. [Resnick 1989] L.B. Resnick, Knowing, Learning, and Instruction: Essays in Honor of Robert Glaser, in (eds.), Lawrence Erlbaum Associates, Hillsdale, NJ, 1989, [Riesbeck, Schank 1989] C. Riesbeck and R.C. Schank, Inside Case-Based Reasoning, Lawrence Erlbaum Associates, Hillsdale, NJ, 1989. [Rittel 1984]
H. Rittel, in N. Cross (eds.), Second-Generation Design Methods Developments in Design Methodology, John Wiley & Sons, New York, 1984, [Schoen 1983] D.A. Schoen, The Reflective Practitioner: How Professionals Think in Action, Basic Books, New York, 1983. [Scribner, Sachs 1990] S. Scribner and P. Sachs, On The Job Training: A Case Study, in (eds.), National Center on Education and Employment, 1990, [Simon 1981] H.A. Simon, The Sciences of the Artificial, The MIT Press, Cambridge, MA, 1981. [Stahl 1993] G. Stahl, Supporting Interpretation in Design, in (eds.), Architecture and Planning Research, Special Issue on Computational Representations of Knowledge, 1993, [Stefik 1986] M.J. Stefik, The Next Knowledge Medium, in (eds.), AI Magazine, 1986, [Suchman 1987] L.A. Suchman, Plans and Situated Actions, Cambridge University Press, Cambridge, UK, 1987. [Wenger 1987] E. Wenger, Artificial Intelligence and Tutoring Systems, Morgan Kaufmann Publishers, Los Altos, CA, 1987. [Weyer 1987] S.A. Weyer, As We May Learn, in (eds.), Multimedia in Education: Interfaces to Knowledge, Education Advisory Council Conference Proceedings, Apple Computer, 1987, [Winograd, Flores 1986] T. Winograd and F. Flores, Understanding Computers and Cognition: A New Foundation for Design, Ablex Publishing Corporation, Norwood, NJ, 1986.
