20.3 Contextualizing instruction: cognitive apprenticeships

Collins, Brown, and colleagues (e.g., Collins, Brown & Newman, 1989; Collins, 1991) developed an instructional model derived from the metaphor of the apprentice working under the master craftsperson in traditional societies, and from the way people seem to learn in everyday informal environments (Lave, 1988). The cognitive apprenticeship model rests on a somewhat romantic conception of the "ideal" apprenticeship as a method of becoming a master in a complex domain (Brown, Collins & Duguid, 1989). In contrast to the classroom context, which tends to remove knowledge from its sphere of use, Collins and Brown recommend establishing settings where worthwhile problems can be worked with and solved. The need for a problem-solving orientation to education is apparent from the difficulty schools are having in achieving substantial learning outcomes (Resnick, 1989).

Emulating the best features of apprenticeships is needed because, as Gott (1988a) noted, lengthy periods of apprenticeship are becoming a rarity in industrial and military settings. She termed this phenomenon the lost apprenticeship and noted the effects of the increased complexity and automation of production systems. First, the need is growing for high levels of expertise in supervising and using automated work systems; correspondingly, the need for entry levels of expertise is declining. Workers on the job are more and more expected to be flexible problem solvers; human intervention is often most needed at points of breakdown or malfunction. At these points, the expert is called in. Experts, however narrow the domain, do more than apply canned job aids or troubleshooting algorithms; rather, they draw on the considerable knowledge they have internalized and use it to solve problems flexibly in real time (Gott, 1988b).

Gott's second observation relates to training opportunities. Now, at a time when more problem-solving expertise is needed due to the complexity of systems, fewer on-the-job training opportunities exist for entry-level workers. There is often little or no chance for beginning workers to acclimatize themselves to the job, and workers very quickly are expected to perform like seasoned professionals. True apprenticeship experiences are becoming relatively rare. Gott calls this dilemma-more complex job requirements with less time on the job to learn-the "lost" apprenticeship and argues for the critical need for cognitive apprenticeships and simulation-type training to help workers develop greater problem-solving expertise.

20.3. 1 Features of Cognitive Apprenticeships

The Collins-Brown model of cognitive apprenticeship (see 7.4.4) incorporates the following instructional strategies or components.

1. Content: Teach tacit, heuristic knowledge as well as textbook knowledge. Collins et al. (1989) refer to four kinds of knowledge:

• Domain knowledge is the conceptual, factual, and procedural knowledge typically found in textbooks and other instructional materials. This knowledge is important, but often is insufficient to enable students to approach and solve problems independently.
• Heuristic strategies are "tricks of the trade" or "rules of thumb" that often help narrow solution paths. Experts usually pick up heuristic knowledge indirectly through repeated problem-solving practice; slower learners usually fail to acquire this subtle knowledge and never develop competence. There is evidence to believe, however, that at least some heuristic knowledge can be made explicit and represented in a teachable form (Chi, Glaser & Farr, 1988).
• Control strategies are required for students to monitor and regulate their problem-solving activity. Control strategies have monitoring, diagnostic, and remedial components; this kind of knowledge is often termed metacognition (Paris & Winograd, 1990).
• Learning strategies are strategies for learning; they may be domain, heuristic, or control strategies, aimed at learning. Inquiry teaching to some extent directly models expert learning strategies (Collins & Stevens, 1983).

2. Situated learning: Teach knowledge and skills in contexts that reflect the way the knowledge will be useful in real life. Brown, Collins, and Duguid (1989) argue for placing all instruction within "authentic" contexts that mirror real-life problem-solving situations. Collins (1991) is less forceful, moving away from real-life requirements and toward problem-solving situations: For teaching math skills, situated learning could encompass settings "ranging from running a bank or shopping in a grocery store to inventing new theorems or finding new proofs. That is, situated learning can incorporate situations from everyday life to the most theoretical endeavors" (Collins, 1991, p. 122).

Collins cites several benefits for placing instruction within problem-solving contexts:

• Learners learn to apply their knowledge under appropriate conditions.
• Problem-solving situations foster invention and creativity.
• Learners come to see the implications of new knowledge. A common problem inherent in classroom learning is the question of relevance: How does this relate to my life and goals? When knowledge is acquired in the context of solving a meaningful problem, the question of relevance is at least partly answered.
• Knowledge is stored in ways that make it accessible when solving problems. People tend to retrieve knowledge more easily when they return to the setting of its acquisition. Knowledge learned while solving problems gets encoded in a way that can be accessed again in similar problem-solving situations.

3. Modeling and explaining: Show how a process unfolds and tell reasons why it happens that way. Collins (1991) cites two kinds of modeling: modeling of processes observed in the world and modeling of expert performance, including covert cognitive processes. Computers can be used to aid in the modeling of these processes. Collins stresses the importance of integrating both the demonstration and the explanation during instruction. Learners need access to explanations as they observe details of the modeled performance. Computers are particularly good at modeling covert processes that otherwise would be difficult to observe. Collins suggests that truly modeling competent performance, including the false starts, dead ends, and backup strategies, can help learners more quickly adopt the tacit forms of knowledge alluded to above in the section on content. Teachers in this way are seen as "intelligent novices" (Bransford et al., 1988). By seeing both process modeling and accompanying explanations, students can develop "conditionalized" knowledge, that is, knowledge about when and where knowledge should be used to solve a variety of problems.

4. Coaching: Observe students as they try to complete tasks and provide hints and helps when needed (see 7.4.5). Intelligent tutoring systems sometimes embody sophisticated coaching systems that model the learner's progress and provide hints and support as practice activities increase in difficulty. The same principles of coaching can be implemented in a variety of settings. Bransford and Vye (1989) identify several characteristics of effective coaches:

• Coaches need to monitor learners' performance to prevent their getting too far off base, but leaving enough room to allow for a real sense of exploration and problem solving.
• Coaches help learners reflect on their performance and compare it to the performance of others.
• Coaches use problem-solving exercises to assess learners' knowledge states. Misconceptions and buggy strategies can be identified in the context of solving problems.
• Coaches use problem-solving exercises to create the "teachable moment."

5. Articulation: Have students think about their actions and give reasons for their decisions and strategies, thus making their tacit knowledge more explicit. Think-aloud protocols are one example of articulation. Collins (1991) cites the benefits of added insight and the ability to compare knowledge across contexts. As learners' tacit knowledge is brought to light, that knowledge can be recruited to solve other problems.

6. Reflection: Have students look back over their efforts to complete a task and analyze their own performance. Reflection is like articulation, except it is pointed backwards to past tasks. Analyzing past performance efforts can also influence strategic goal-setting and intentional learning (Bereiter & Scardamalia, 1989). Collins and Brown (1988) suggest four kinds or levels of reflection:

• Imitation occurs when a batting coach demonstrates a proper swing, contrasting it with your swing.
• Replay occurs when the coach videotapes your swing and plays it back, critiquing and comparing it to the swing of an expert.
• Abstracted replay might occur by tracing an expert's movements of key body parts such as elbows, wrists, hips, and knees, and comparing those movement to your movements.
• Spatial reification would take the tracings of body parts and plot them moving through space.

The latter forms of reflection seem to rely on technologies-video or computer-for feasible implementation.

7. Exploration: Encourage students to try out different strategies and hypotheses and observe their effects. Collins (199 1) claims that through exploration, students learn how to set achievable goals and to manage the pursuit of those goals. They learn to set and try out hypotheses, and to seek knowledge independently. Real-world exploration is always an attractive option; however, constraints of cost, time, and safety sometimes prohibit instruction in realistic settings. Simulations are one way to allow exploration; hypermedia structures are another.

8. Sequence: Present instruction in an ordering from simple to complex, with increasing diversity, and global before local skills.

• Increasing complexity. Collins et al. (1989) point to two methods for helping learners deal with increasing complexity. First, instruction should take steps to control the complexity of assigned tasks. They cite Lave's study of tailoring apprenticeships: Apprentices first learn to sew drawers, which have straight lines, few pieces of material, and no special features like zippers or pockets. They progress to more complex garments over a period of time. The second method for controlling complexity is through scaffolding. In this case the cases or content remains complex, but the instructor provides the needed scaffolding for initial performances and gradually fades that support.
• Increasing diversity refers to the variety in examples and practice contexts.
• Global before local skills refers to helping learners acquire a mental model of the problem space at very early stages of learning. Even though learners are not engaged in full problem solving, through modeling and helping on parts of the task (scaffolding), they can understand the goals of the activity and the way various strategies relate to the problem's solution. Once they have a clear "conceptual map" of the activity, they can proceed to developing specific skills.

The three teaching models presented below illustrate various features of the cognitive apprenticeship model. The first two are computer-based environments: Sherlock and goal-based scenarios. The third model is the problem-based learning environment developed by medical educators at the University of Illinois. All three models build instruction around problems or cases that are authentic to real-life situations, within which learners learn the details of a subject matter.

20.3.2 Sherlock

What happens when you combine extensive cognitive-task analysis of a well-defined, technical domain with a situated-learning philosophy and a teaching model based on intelligent tutoring systems (ITS)? Sherlock is an example of such a teaching tool. Sherlock is a computer-coached practice environment developed by Alan Lesgold and colleagues (e.g., Lajoie & Lesgold, 1992; Lesgold et al., 1988; Lesgold, Lajoie, Bunzo & Eggan, 1992) to develop the troubleshooting skills of Air Force electronics technicians. Sherlock was specifically designed to teach the most difficult parts of the troubleshooter's job. Learners are presented a number of troubleshooting problems requiring two kinds of activities:

• The student solves the problem, requesting advice from the intelligent tutor/coach as necessary.
• The student reviews a record of his/her problem-solving activity, receiving constructive critique from the coach (Gott, Lesgold & Kane, 1996).

Sherlock serves not only as an instructional environment and an assessment device but also as a laboratory for instructional research (Lajoie & Lesgold, 1992). Assessment is interwoven with instruction so that coaching is highly individualized. This is achieved using expert systems technology to create two types of student modeling using expert systems technology: a competency model that is updated throughout the program, and a performance model. Together, they provide the basis for diagnosing learner problems and selecting appropriate instructional mediation relating to goals, operators, methods, and strategies.

A central feature of Sherlock is its intelligent hyperdisplay:

When Sherlock constructs a schematic diagram to help illustrate the advice it is providing, that diagram is organized to show expert understanding about the system with which the trainee is working... . What is displayed is approximately what a trainee would want to know at that time, but every display component is "hot" and can be used as a portal to more detail or explanation (Gott et al., 1995).

Sherlock's diagrams are dynamic; that is, they are assembled at any point in the program to be sensitive to immediate conditions. The diagrams are adjusted so that conceptually central components are afforded the most space; diagram boxes and circuit paths are color coded to reflect the learner's prior knowledge about them.

Gott, Hall, Pokorny, Dibble. and Glaser (1992) studied Sherlock learning environments to find out how students made flexible use of their knowledge in novel situations:

Time and again we observed [successful] learners access their existing mental models of equipment structure ... and their schema of the troubleshooting task... . They then used these models as flexible blueprints to guide their performance as they crafted solutions to new problems. Their prior models became interpretive structures, and when these models were inadequate, better learners flexibly used them as the basis for transposed and elaborated structures that could accommodate the novel situations. They were ready and willing to construct new knowledge that was grounded in their existing representational and functional competence (Gott et al., 1996, pp. 37-38).

The current incarnation of the program, Sherlock 2, includes a number of refinements aimed at facilitating students' development of device models and transfer of knowledge. Consistent with the cognitive apprenticeship model, students have a variety of supports and reflective tools available to them:

To complement coached learning by doing, we have developed a collection of tools for post-performance reflection. One provides an intelligent replay of the trainee's actions. A trainee can "walk through" the actions he just performed while solving the problem. In addition, he can access information about what can in principle be known about the system given the actions replayed so far... Also, he can ask what an expert might have done in place of any of his actions, get a critique of his action, and have his action evaluated by the system.... Further, there is an option for side-by-side listing of an expert solution and the trainee's most recent effort (Gott et al., 1996, p. 35).

One key to Sherlock's success is the extensive and sophisticated cognitive task analysis that provided "critical information necessary for developing the appropriate student models of proficiency" (Lajoie & Lesgold, 1992, p. 381). The program addresses eight dimensions of proficiency proposed by Glaser, Lesgold, and Lajoie (1987) and based on research on expert-novice differences (Lajoie & Lesgold, 1992):

1. Knowledge organization & structures
2. Depth of underlying principles
3. Quality of mental models
4. Efficiency of procedures
5. Automaticity to reduce attentional demands
6. Procedural knowledge
7. Procedures for theory change
8. Metacognitive skills

According to Gott et al. (1996, p. 36), the Sherlock team's approach to task analysis is similar but distinguishable from traditional instructional-design approaches. "What is different is that the structure of learning tasks is more authentic, rooted in the needs of practice (or simulated practice) rather than being derived directly from task analysis structure... ." Research has found that learners who used Sherlock improved dramatically in their troubleshooting skills, during training as well as on a posttest (Lesgold et al., 1988; Nichols, Pokorny, Jones, Gott & Alley, 1995). Sherlock 2 yielded effect sizes on posttest measures ranging from .87 to 1.27 (Gott et al., 1996).

20.3.3 Goal-Based Scenarios

Roger Schank and colleagues at the Institute for the Learning Sciences have developed an architecture for the design of learn-by-doing courses and simulations (Schank, Fano, Bell & Jona, 1993/1994). Goal-based scenarios constitute an approach to the design of intelligent tutoring systems (ITS) that combine elements of simulation, case-based reasoning, and traditional ITS modeling techniques. Riesbeck (1996, p. 52) describes the concept of a goal-based scenario:

In a [goal-based scenario], a student is given a role to play, e.g., owner of a trucking company or chief scientist at a nuclear research installation, and interesting problems to solve or goals to achieve. The role and problems should be of real interest to the student, e.g., feeding the world, getting rich, or flying a rocket to the moon, not artificial word problems....

The student engages in a simulation in order to solve the defined problem or achieve the goal. Typically the student interacts with simulated agents and objects within a simulated environment. A goal-based scenario differs, however, from traditional simulations in a number of respects. For example:

When the student gets stuck or in trouble, a tutor, in video form, appears to offer advice, tell stories, and so on.. The stories come from a multimedia archive of texts and video interviews of experts in that domain, telling personal experiences similar to the student's simulated situation. These stories are also organized for browsing in a structure we call ASK networks (Ferguson et al., 1991, cited in Riesbeck (1996, p. 52).

ASK networks are systems for indexing and archiving stories in a way that makes them useful within the goal-based scenario. Stories are brought into the simulation as the need arises, with the indexing sensitive to the learner's progress and other local conditions.

Three goal-based scenarios that have been developed are briefly reported below.

20.3.3.1. Broadcast News. High school students collaborate to produce their own simulated TV news broadcast. "The student first sees a brief introduction informing him or her that he or she will be working on a newscast for a particular day in history ... [say] May 21, 1991. The student is then given a rough draft of a news story that requires revisions so it can go on the air" (Schank et al., 1993/94, p. 309).

Students often lack the historical or political knowledge to understand the draft script, so they consult tools and resources within the program that provide a context for understanding the script.

The student then needs to revise the script and prepare it for broadcast. Rather than personally rewriting the draft, the student submits specifications for revision back to the writers. The program's experts can provide support and advice through this process. As in real life, two experts' feedback, in fact, may conflict with each other, forcing the student to decide how to interpret suggestions and incorporate them into the revised script.

After the student gives final approval to a story, he or she can then choose to play the role of anchorperson for the newscast as well. The program then acts as a teleprompter and editing booth. The student reads the story as the text rolls by on the screen. A video camera controlled by the computer records the student as he or she plays the role of anchor; the computer also supplies the video accompanying the story. A complete videotape of the student's newscast is ready as soon as the newscast ends ( Schank et al., 1993/94, pp. 309-10).

The student can watch the tape and compare it with a professional network newscast covering the same event. This comparison fosters reflection and discussion about the process and decisions made.

20.3.3.2. Sickle Cell Counselor. This is an interactive hypermedia exhibit designed for the Museum of Science and Industry in Chicago (Bell, Bareiss & Beckwith, 1993/94; Schank et al., 1994, pp. 310-11). The user assumes a role of genetic counselor advising couples of the genetic risks of their upcoming marriage. The student is able to run simulated lab tests, interact via interactive video with the couple, collect data from the couple, and offer advice. Research indicated that museum visitors spent considerably more time with the exhibit than for other exhibits and learned something about genetics (based on both self-report and performance on pre- and posttests).

20.3.3.3. YELLO. This is a program designed to teach telephone operators how to sell Yellow Pages advertising (Kass, Burke, Blevis & Williamson, 1993/94). The program follows a framework designed for the teaching of complex social skills. Of particular interest is the interjection of stories into the practice section. The program tracks student performance and retrieves a "story" that matches the student's performance profile, based on a sophisticated indexing scheme. The story-a real-life recounting by an experienced practitioner-is then provided to the student to strengthen motivation and make the task meaningful, as well as to correct the performance.

Schank et al. (1993/1994) outline the principal components of goal-based scenarios, along with criteria for good design. The following four components form the basis of a goal-based scenario.

20.3.3.4. Mission. The mission is the overall goal of the goal-based scenario. A scenario's mission may relate to process skills or outcome achievement skills. Process skills (e.g., running a trucking company, flying a plane, being a bank teller, etc.) lend themselves to role-play scenarios where the learner assumes the role of a character and learns the knowledge and skills related to that role. Outcome achievement skills (e.g., troubleshooting an engine or building a bridge) lend themselves to a scenario focusing on a specific task or achieving a particular result. By accomplishing the task, the student learns relevant skills along the way.

20.3.3.5. Mission Focus. Schank et al. (1993/94) refer to the mission focus as "the underlying organization" of the activities engaged in by students (p. 327). They identified four mission foci:

• Explanation. Students are asked to account for phenomena, predict outcomes, diagnose systems, etc. Sickle Cell Counselor has an explanation mission focus.
• Control. Students are asked to run an organization or maintain and regulate a functioning system. Examples of a control focus would be managing a software project or running a nuclear power plant (Schank et al., 1993/94, p. 331).
• Discovery. Students enter a microworld and explore the features available. They may be asked to infer the microworld's governing principles or participate in activities available. YELLO offers an example of this type of mission focus.
• Design. Students create or compose some product or create the design specifications for some artifact. An example would be Broadcast News.

20.3.3.6. Cover Story. The cover story provides the specifics of the student's role and the surrounding context Schank et al, (1993/94) suggest designing a cover story around something the student might like to do (e.g., be President of the United States or fly an airplane) or something the student would have some strong feeling for (e.g., investigate the Chernobyl accident site, help a person threatening suicide). The details of the story are worked out in the design of the cover story, including the "setup" (explanations to students about why the scenario is important, specification of tools available in solving problems, etc.) and the "scenes" (the specific physical settings encountered in the story).

20-3.3.7. Scenario Operations. Specification of scenario operations is the final stage of a goal-based scenario design. Scenario operations are the discrete, specific responses required of students engaged in the program. Examples might include "adjusting a parameter with a dial, issuing a directive in a social simulation, answering a question, using a tool to shape part of an artifact, searching for a piece of information, and deciding between two alternatives" (Schank et al., 1993/94, p. 336).

In many ways, Schank's work continues the tradition of developing instructional simulations (see 17.4, 17.5), popular forms since the advent of computer-based instruction in the 1960s, including models of role playing, control of dynamic systems, and task performance. Schank's contribution has been in the development of a sound theoretical model based on cognitive memory research, and in the creation of a design laboratory that follows A well-defined development model in creating working products. The costs of developing full-blown goal-based scenarios undoubtedly remain high, but they signal important progress in the design of instructional systems.

20.3.4 Problem-Based Learning

As noted above, intelligent tutoring systems rely on extensive and sophisticated cognitive task analysis to develop expert and student models; moreover, they are usually designed for individual learners. Like Sherlock, they tend to address well-structured problems in well-structured domains and build on a broad base of content knowledge. Problem-based learning (PBL) addresses ill-structured problems and/or ill-structured domains. Koschmann, Myers, Feltovich, and Barrows (1994) stressed the distinction between an ill-structured domain (see 7.5, 23.5.1.3), as characterized by Spiro, Coulson, Feltovich, and Anderson (1988), "in which no single concept, or even a small number of conceptual elements, is sufficient for capturing the workings of a typical instance of knowledge application" (Koschmann et al., 1994, p. 23 1), and an ill-structured problem:

...defining the problem requires more information than is initially available-the nature of the problem unfolds over time; there is no single, right way to get that information; as formation is obtained, the problem changes; decisions must be made in the absence of definitive knowledge; and there may never be certainty about having made the right decision (Barrows & Feltovich, 1987, as summarized by Koschmann et al., 1994, p. 23 1).

Problem-based learning integrates the learning of content and skills, utilizes a collaborative environment, and emphasizes "learning to learn" by placing most of the responsibility for learning on the learner rather than providing a sophisticated pre-designed instructional system (see 7.5, 23.5.1.3).

The PBL model has been implemented in several areas of higher education, including medicine, business, education, architecture, law, engineering, and social work, as well as in high school (Savery & Duffy, 1996). However, the. best known applications of PBL are in medical schools, where it was developed in the 1950s, and whose graduates face particularly ill-structured problems in an ill-structured and ever-expanding domain that requires lifelong learning skills. (See, e.g., Williams, 1992; Savery & Duffy, 1996, and Koschmann et al., 1994, for critical overviews of the use of PBL in medical schools.) More than 100 medical programs include a PBL option (Duffy, 1994). PBL combines the teacher-directed case method that is used extensively in law and business schools with the discovery-learning philosophy of Jerome Bruner (Lipkin, 1989; Schmidt, 1989).

During the first 2 years of medical school, students in a PBL curriculum work in small self-directed groups to learn simultaneously content (basic science knowledge) and skills (the procedural skills of examining and diagnosing patients, and metacognitive skills such as monitoring their comprehension, identifying their learning needs, identifying and using resources efficiently, and reflecting on procedural and metacognitive skills). In lieu of traditional lectures and laboratory exercises, the problem-based curriculum presents a series of authentic patient problems; groups of five to seven students work intensely for about a week on each problem, diagnosing and learning to understand its causes. Authenticity is critical in motivating students and in avoiding the "construction of fictional problems ... [with] symptoms that cannot coexist" (Williams, 1992, p. 404).

The facts of the problem are presented, just as they were initially to a doctor, as an incomplete set of symptoms that must be evaluated and explained. For practical reasons, the presentation is usually simulated on paper or by an actor trained as a patient. The facts of the complete case are contained in a problem-based learning module (Barrows, 1985; Distlehorst & Barrows, 1982), which "is designed to allow for free inquiry, providing responses for any question, examination, or laboratory test an examiner might request for the actual patient" (Koschmann et al., 1994, p . 241) without cueing any factors that are critical to the case (Savery & Duffy, 1995). A tutor facilitates students' negotiation of five recursive stages of the problem-based methodology (Koschmann et al., 1994):

1. Problem formulation. Students isolate important facts froth their ' rich context, identify the problem, and generate hypotheses.
2. Self-directed learning. Group members identify and address information needed to evaluate hypotheses. This list of needed information sets the learning agenda. For example, they might research basic biological mechanisms that might underlie a patient's problems, question or "examine" the patient, review results of tests they "order," and consult with the medical faculty.
3. Problem reexamination. Group members bring to bear their findings from their self-directed learning activities-adding, deleting, or revising hypotheses as warranted.
4. Abstraction. This is an articulation process (cf. Collins, Brown & Newman, 1989) during which members compare and contrast cases, forming cognitive connections to increase the utility of the knowledge gained in specific contexts.
5. Reflection. At this point the group debriefs the experience and identifies areas for improvement in their learning processes.

The role of the tutor, according to Barrows (1992), is critical to "the success of any educational method aimed at (1) developing students' thinking or reasoning skills (problem solving, metacognition, critical thinking) as they learn, and (2) helping them to become independent, self-directed learners ..." (p. 12). The tutor must not provide mini-lectures or suggest solutions but help each member of the group internalize effective metacognitive strategies by monitoring, modeling, coaching, and fading. This role includes

not only moving students through the various stages ... but also monitoring group process and the participation of individuals within it, guiding the development of the clinical reasoning process by strategically questioning the rationale underlying the inquiry strategy of the group or individuals, externalizing self-questioning and self-reflection by directing appropriate questions to individuals or the group as a whole, and evaluating each student's development (Koschmann et al., 1994, p. 243).

The tutor externalizes higher-order thinking that students are expected to internalize, by asking students to justify not only their inferences (e.g., "How do you know that's true?") but also any question they ask the patient (Savery & Duffy, 1995; Williams, 1992). Thus the students learn to identify and challenge superficial thinking and vague notions. In PBL, the tutor carefully monitors each student's development and forces him or her to remain an active learner.

Medical students appear to be very positive about the approach, particularly in their first year. However, many issues are still to be addressed. Because so much of the work is group oriented, PBL cannot guarantee that the student will do his or her own thinking. As Koschmann et al. (1994) noted:

Teaching methods that depend on group interaction often experience what is termed the polling problem; the opinions of individuals vary as a function of the order in which their views are gathered. Contributions of less dominant members may be suppressed or contaminated by the more dominant members; convictions of any single individual in the group may be inappropriately influenced by other members; individuals can find means to hide or ride on the coattails of other group members. The polling problem, therefore, can result in the suppression of ideas, reducing the multiplicity of viewpoints expressed (p. 243).

Compared to the computer-based programs reviewed above, PBL is relatively ill-defined; that is, students' specific interactions cannot be pre-specified. Because of this, good design and successful implementation become intertwined: Design must happen constantly throughout the course of students' activities. The instructor must remain vigilant to ensure that less-dominant members still have a voice within the group. Reluctant learners need to be monitored and encouraged to participate. Thus, more than some forms of controlled instruction, PBL depends on high-quality implementation by skilled instructors and participants.

PBL is also time consuming. Problem-based activities can become tedious and boring, especially for students who have already internalized the clinical reasoning process. These complaints lead one to ask: Is it essential that students discover or seek out primary sources in the library for every case? Or might learning be just as effective but more efficient if hypermedia programs consolidated answers to relevant questions to reduce the temporal and cognitive loads, at least for some of the cases? A careful analysis to identify the specific activities most valuable in generating new knowledge seems in order (cf. Collins's epistemic games, described below).

Indeed, faculty have begun to modify PBL programs. For example, faculty at the University of New Mexico School of Medicine designed focused cases "to solve these problems by encouraging learning on a single topic rather than leading to the generation of multiple, diverse hypotheses" (Williams, 1992, pp. 403--04). Koschmann et al. (1994) are developing computer-assisted programs to augment PBL, several of which directly or indirectly reduce cognitive load. These programs will facilitate:

• Students' expression of their honest viewpoints
• "A retrievable record of the group's deliberations for previously studied cases" (p. 247)
• The tutor's monitoring of the development of each person's progress
• The collection of authentic cases
• The selection of cases appropriate to the needs of the students
• Students' communicating outside group meetings
• Students' access to learning resources
• Students' access to information in their notes

Koschmann et al. (1994) are designing programs that utilize groupware (Stefik & Brown, 1989), hypertext/hypemedia (Conklin, 1987; see Chapter 21), database technologies (see 24.9), and electronic mail.