Problem-Based Learning Meets Case-Based Reasoning
Janet L. Kolodner, Cindy E. Hmelo, and N. Hari Narayanan
The EduTech Institute
College of Computing
Georgia Institute of Technology
Atlanta, GA 30332-0280
{jlk, ceh, narayan}@cc.gatech.edu
Abstract: The modern education community agrees that deep and effective learning is
best promoted by situating learning in authentic activity. Many in the education
community have put in place constructivist classroom practices that put students into
situations where they must make hypotheses, collect data, and determine which data to use
in the process of solving a problem or participating in some kind of realistic analysis or
investigation. Research in case-based reasoning (CBR), which provides a plausible model
of learning from problem solving situations, makes suggestions about education that are
consistent with these educational theories and methodologies and which can provide added
concreteness and detail. In this paper, we show how CBR's suggestions can enhance
problem-based learning (PBL), which is already a well-worked-out and successful approach
to education. The computational accounts CBR provides of reasoning activities,
especially of knowledge access, access to old experiences (cases), and use of old
experiences in reasoning, suggest guidelines about materials that should be made
available as resources, the kinds of reflection that will promote transfer, qualities of good
problems, qualities of the environment in which problems are solved (e.g., affordances for
feedback), and sequencing a curriculum. The two approaches complement each other well,
and together, we believe they provide a powerful foundation for educational practice in the
constructivist tradition, one that at once combines lessons learned from classroom
practice with sound cognitive theory.
1. Introduction
The modern education community agrees that deep and effective learning is best promoted by situating learning
in authentic activity. Anchored instruction, project-based learning, problem-based learning, and other
constructivist approaches to classroom practice all focus on putting students into situations where they must
make hypotheses, collect data, and determine which data to use in the process of solving a problem or
participating in some kind of realistic analysis or investigation (Barrows, 1985; Blumenfeld, et al, 1991;
CTGV, 1993; Williams, 1993). Research shows that students participating in these kinds of learning activities
are more motivated to learn, that what they learn is more usable than the knowledge learned by students carrying
out rote activities, and that they tend to better learn higher order thinking skills than do students in other
learning situations (Blumenfeld, et al., 1991; CTGV, 1993; Hmelo, 1995).
Designing such activities well requires an understanding of what needs to be learned and the kinds of experiences
that will lead to such learning. But making learning activities work well requires that their use in the classroom
be informed by lessons of practice. The classroom is complicated. Social issues, personality issues, capabilities
of teachers, the comfort of the classroom, and other issues all play a role in student learning, muddling the best
educational designs or allowing poor ones to work despite their weaknesses. Designing effective learning
activities thus requires (1) cognitive (and social) theory to provide guidelines about learning, (2) classroom
methodology, or lessons of practice, to provide guidelines about operationalization, and (3) trial, analysis, and
refinement (Linn & Songer, 1988) over time aimed toward operationalizing the activities well. Logic tells us
that the more operational cognitive theories are, the easier it will be to put their guidelines into practice; and the
closer the match between the guidelines for learning and the classroom methodology, the more chance that each
could contribute toward better understanding of the other.
This logic, plus a desire to put the educational philosophy of case-based reasoning into practice and a need to
advise faculty at Georgia Tech about how to help their students learn from problem solving activity, has led us
to begin to put into place a marriage between problem-based learning and case-based reasoning. Case-based
reasoning (Schank, 1982; Kolodner, 1993), which originated as a methodology for implementing computer
programs that could solve problems based on their past experiences, provides a cognitive theory that situates
learning in reasoning about real-world situations. The computational accounts it provides of reasoning
activities, especially of knowledge access, access to old experiences (cases), and use of old experiences in
reasoning, provide suggestions about what makes a good problem, the range of problems students should solve,
the kinds of materials and resources students need to use, ways to manage the complexity of hard problems and
the kinds of reflection that should be encouraged. Some of CBR's principles have been used to inform the
design of stand-alone learning environments (Schank, et al., 1993), but when we wanted to put its principles
into practice in classrooms, we found that CBR was lacking in telling us about the teacher's role and other
issues of classroom practice. We thus sought an educational methodology that would provide principles of
practice to go along with CBR's educational principles.
We chose problem-based learning because of its parallels to case-based reasoning. Problem-based learning
(Barrows, 1985) provides a classroom methodology that situates learning in problem-solving activity. It has
been in use in medical schools for twenty years and specifies activities of students and teachers that promote
learning from problem-solving experiences. It originated from the intuitions of excellent teachers and has
evolved into a well-developed system of practice through systematic refinement. Problem-based learning gives
reflection on problem-solving activity a central role, and specifies roles for students as researchers who discover
knowledge and teachers as facilitators of this constructivist process.
The two approaches complement each other well, and together, we believe they provide a powerful foundation
for educational practice in the constructivist tradition, one that at once combines lessons learned from classroom
practice with sound cognitive theory.
In this paper, we explore how these two views complement each other,
presenting each in turn, and then synthesizing them and describing the implications of their synthesis. Each
adds to the other; CBR adds reason and specificity to some of PBL's practices and suggests some new ways of
accomplishing PBL's goals, while PBL provides for CBR a vehicle for putting its philosophy into practice.
2. Problem-Based Learning: A Model of the Practice of Learning from Problem-
Solving Activity
In problem-based learning, students learn by solving problems and reflecting on their experiences. PBL is used
substantially at medical and business schools (Barrows, 1985; Williams, 1993). Students learn by solving
authentic real-world problems; in medicine, this means diagnosing and managing patient cases. Because the
problems are complex, students work in groups, where they pool their expertise and experience and together
grapple with the complexities of the issues that must be considered. Coaches guide student reflection on their
problem-solving experiences, asking students to articulate both the concepts and skills they are learning, and
helping them identify the cognitive skills needed for problem solving, the full range of skills needed for
collaboration and articulation, and the principles behind those skills. But coaches do not tell students what to do
or show them how to solve problems; rather students decide how to go about solving problems and what they
need to learn, while coaches question students to force them to justify their approach and explain their
conclusions. In this way, skills needed for life-long learning are also acquired. Research shows that students in
problem-based curricula are indeed learning facts and concepts and the skills needed for critical problem solving
and self-learning (Hmelo, 1995; Norman & Schmidt, 1992).
In its classical form (Barrows, 1985), students take no classes, but rather work on a series of 100 problems over
their first two years of medical school. Small groups of 5-7 students and a coach meet to discuss a patient case.
Cases are presented to students in an authentic way; the presenting symptoms (those the patient reports) are first
presented to students, and as the students feel they need more information about a patient, they ask for it. A
patient record holds all such information and is indexed by questions that traditionally come up in a medical
consultation. Problem solving begins with an attempt to interpret what is wrong with the patient; it continues
with an attempt to manage the patient's care. Students keep track, on a set of white boards, of facts they know,
hypotheses they have about what might be wrong, ideas about treatment, and issues they do not yet understand
and need to learn more about (learning issues). They follow a methodology of first, writing down what they
know, then generating hypotheses, then identifying the things they still need to find out about the patient and
the issues they need to learn more about. After considering the case with their naive knowledge, students divide
up the learning issues they have generated among the group and research them. When they get back together,
they return to their problem-solving activity, this time using what they have learned through research to move
farther forward in their solutions. They reconsider their hypotheses and/or generate new hypotheses in light of
their new learning. This cycle continues until students are satisfied that they have solved the problem. They
are then presented with the expert opinion on the problem and discuss its pros and cons if different than their
conclusion.
3. Case-Based Reasoning: A Cognitive and Computational Model of Learning from
Problem-Solving Activity
The term case-based reasoning refers to reasoning based on previous experience. It might mean solving a new
problem by adapting an old solution or merging pieces of several old solutions, interpreting a new situation in
light of similar situations, or projecting the effects of a new situation by examining the effects of a similar old
situation. In short, case-based reasoning means using the lessons learned in old situations to understand or
navigate new ones. The basic premise underlying CBR is the preference to reason using the most specific and
most cohesive applicable knowledge available. Inferences made using specific knowledge are relatively simple
to make. Inferences made using cohesive knowledge structures, i.e., those that tie together several aspects of a
situation, are relatively efficient. Cases, which describe situations, have both of these properties. In addition,
they record what is possible, providing a reasoner with more probability of moving forward in a workable way
than is provided by using general knowledge that is merely plausible. Furthermore, reasoning based on previous
experience seems natural in people; an understanding of how it is done well and effectively can provide
guidelines for helping people to effectively use this natural reasoning process.
A good example of case-based reasoning comes from architecture. An architect is designing an office building
with a long naturally lit atrium in the middle and a circular row of offices surrounding it. She wants the office
to get as much light as possible so daytime energy consumption can be minimized. She remembers the design
of a library that has no atrium but where the designer solved the problem of bringing in sunlight by
constructing exterior walls of glass. She realizes that this solution can be used in the current building -- the
office space can be separated from the atrium by a circular glass wall. But upon further thought, she remembers
the problems that a courthouse had, in which a glass wall was used in a row of offices with heavy public traffic.
While the offices were well lit, the constant presence of the public interfered with the privacy and work of the
office workers. The library did not have this problem because the glass wall faced a wooded area. While the
first case provides a means of dealing with her new design, the second case, and its difference from the first,
alerts her to a potential problem with that solution. Comparing these two cases with the current one, she
realizes that the potential for the problem exists, but to a lesser degree. While the atrium is not deserted like the
woods, it is not a heavily-trafficked area either. She decides to use the first solution but modifies it slightly by
using translucent glass bricks instead of clear plate glass for the wall.
While originally derived to explain reasoning and problem solving, the cognitive model underlying case-based
reasoning also provides an explanation of the role memory plays in reasoning -- how memory is accessed during
reasoning and how reasoning contributes to changes in the content and organization of memory. Thus, CBR
addresses the ways experiences are recalled to be used in reasoning, the use of old experiences in reasoning, and
the ways in which new experiences are analyzed, indexed, and stored in memory.
Learning, in the CBR paradigm, means extending one's knowledge by incorporating new experiences into
memory, by re-indexing old experiences to make them more accessible, and by abstracting out generalizations
from experiences. Thus, a major issue case-based reasoning addresses is the indexing problem: identifying old
situations that are relevant to a new one. Two sets of procedures allow such recognition to happen: (1) those
that operate when cases or experiences are encoded and inserted into long-term memory and (2) those that operate
at retrieval time. At insertion time, a reasoner interprets a situation and identifies at least some of the lessons
it can teach and when those lessons might most productively be applied. The case is labeled according to its
applicability conditions, i.e., the circumstances in which it ought to be retrieved. The most discriminating
labels on a case will be derived by a reasoner that has taken the time and effort, and that has the background
knowledge, to carefully analyze a case's potential applicability. At retrieval time, a reasoner uses his/her current
goals and understanding of the new situation as a probe into memory, looking for cases that are usefully similar
to the new one. The extent to which a reasoner is willing or able to interpret the new situation determines the
quality of the probe into memory. An uninterpreted situation is likely to yield poorer access to the contents of
memory than is one that is more embellished. The more creative a reasoner is at interpreting a situation, the
more likely he is to find relevant knowledge and experience to use in reasoning about it.
CBR gives failure a central role in promoting learning. During reasoning, when the reasoner's expectations fail,
it is alerted that its knowledge or reasoning is deficient, and a need to learn arises. Similarly, an unsuccessful
outcome or solution warns the reasoner of a deficiency and therefore a need to learn. Failure at applying an old
case in a new situation triggers explanation that might result in reinterpreting (reindexing) old situations and/or
discovering new kinds of interpretations (indexes). Crucial to interpreting failure is useful feedback from the
world. A reasoner that is connected to the world will be able to evaluate its solutions with respect to what
results from them, allowing indexing that discriminates usability of old cases and allowing good judgments
later about reuse.
4. CBR and PBL: A Synthesis
PBL and CBR have much in common. Both point toward a constructivist mode of education, in which one
learns by extracting wisdom from one's experiences. Each focuses on different but complementary aspects of the
experience. The methodology of PBL asks students to solve problems and then reflect on what they have
learned from the experience. The problems students solve serve two purposes: they become cases for use in
later reasoning and they are vehicles for learning. Case-based reasoning suggests that learning well from
experience requires assessing what lessons an experience teaches and predicting the circumstances when those
lessons might be appropriately applied, adding to PBL's call for reflection specificity about what particular
things to reflect on. CBR also makes suggestions about the kinds of problem solving experiences a learner
should have. CBR suggests furthermore the importance of acquiring feedback on decisions made, in order to be
able to identify the holes in one's knowledge.
Overall, there are many areas where CBR and PBL together make more concrete suggestions about
educational practice than does either one alone: identifying the qualities of good problems, the kinds of
materials and resources that ought to be made available to students as they are generating learning issues,
qualities of the environment in which problems are solved, managing the complexity of solving hard problems,
encouraging the kinds of reflection that promote transfer, facilitating cognitive flexibility, and sequencing the
curriculum.
4.1 What makes for a good problem?
Experiences with medical PBL have led to a number of guidelines about what makes for good problems. In
order to learn real-world reasoning skills, problems are purposely complex, ill-structured, and open-ended,
lending themselves to several interpretations and/or solutions, and painting a cohesive, holistic view of an issue
or situation. This is for several pragmatic reasons, most important in order for students to learn skills and facts
in situations of realistic complexity. It is important, also, that problems be realistic, that they resonate with
the experiences and knowledge of students, and that they are problems the students want to solve. This is to
insure that students will become engaged in the activity and will be motivated to learn, and will be capable of
getting started based on what they already know. Also important is that problems are complex enough to have
several interrelated parts, all important to a good solution. This is necessary so that the learning issues that
students generate are related enough so that they have background, having researched one issue, to understand the
explanations of fellow students when they report about other learning issues, and so that, having researched a
single learning issue, they have an increased understanding of the problem as a whole. And problems should
promote conjecture, argumentation, and peer criticism, by lending themselves to multiple interpretations or
solutions, sometimes depending on which of a variety of conflicting perspectives students take on. The
intuition here is that collaboration and collaborative learning will work best if students have something to work
out together and if they have an authentic need to ask each other to justify their points of view.
Student roles and goals should be made clear in a problem as well. They might be asked to act as physicians
and come up with a solution to a problem, as medical students are asked to do, or they may be asked to act as
scientists, engineers, or consultants of some kind to create a product or performance (e.g., Barrows & Kelson,
1995; Boud & Felletti, 1991; Ram & Hmelo, 1995), e.g., design an experiment to evaluate some evidence or
produce a marketing plan.
An added benefit, if problems fit the criteria listed, is that, in general, they will require students to integrate
knowledge from across multiple disciplines. In medical problems, students may deal with anatomy,
physiology, and pharmacology in a single case; a chemistry problem might elicit consideration of biology at the
same time students are learning chemistry; a design problem might integrate mathematics, life science, and earth
science.
CBR adds several things to this already well-articulated list of descriptors, based on its finding that failure is a
powerful motivator of learning and that connections with the world that afford feedback are critical to learning.
These two findings suggest several things about good problems. (1) The most effective problems for learning
will be those where students can acquire feedback along the way that allows them to recognize the holes and
misconceptions in their knowledge, refine their knowledge and reasoning strategies, and evaluate the goodness of
their knowledge and reasoning strategies. (2) Problems should present difficulties for students, to give them an
opportunity to see where the real complexity in situations or domains lies. (3) There is much to learn from
situations where problems are not solved correctly or well. Students do not always have to be successful at
solving problems, as long as they are learning from the experience. These statements also make a suggestion
about the products of problem solving experiences. It is important for the product or performance asked for in a
problem to be something that provides the kinds of feedback that allow students to identify issues that need to
be addressed or identify holes in their knowledge or reasoning. Of course, if students fail in negative ways, they
will give up and lose confidence. It is important, therefore, that difficulties and failures be orchestrated in gentle
ways and that students understand their role in learning.
4.2 Materials and resources
PBL tells us that it is important that resources used during early inquiry into a problem supply authentic
information about the problem situation and make that information available in an authentic way. The patient
case is used for medicine, where patient signs and symptoms and the results of tests are indexed by typical
questions asked during a medical examination. For an engineering or design problem, the kinds of materials that
would typically be available to a practicing engineer or designer working on the problem need to be made
available to learners.
But PBL does not provide all the guidance that might be needed to identify important issues or relevant
subgoals. Nor does it identify materials and resources that will help students to solve a problem. Case-based
reasoning provides guidelines on these things. Cases recalled during everyday reasoning can serve several
purposes. They can point out issues to focus on, suggest solutions to problems, warn of potential pitfalls,
support projection of the effects of a chosen solution, and so on. While a person working alone has only his or
her own experiences to use as cases to guide reasoning, when working with others, the range of experiences of
those in the group can all contribute to reasoning about a situation. The case-based reasoning community has
turned this observation into a guideline for using case-based reasoning to help people solve problems: case-
based design aids (Domeshek, Kolodner & Zimring, 1994) and other kinds of case libraries (Bell et al., 1993,
Kass et al, 1993) store the cases or experiences of others (often experts) for problem solvers to peruse while
reasoning. Human reasoners can use such systems to augment their own memories. Had the architect in the
example above not known of examples of buildings with good daylight lighting, she would probably have
looked in architectural magazines or files to find such examples and reasoned with them as if they were her own
experiences. This is typical of the way expert designers get started solving problems; case libraries make such
examples readily available.
Case-based reasoning suggests making case libraries available to learners as they are solving hard problems.
PBL suggests that the cases in the case library be indexed in ways that will promote identifying issues that need
to be addressed to come up with a good problem solution (learning issues) and that will suggest potential
solutions or parts of solutions or ways of addressing issues. Case-based reasoning suggests that, in addition, it
is important to include cases that can help with projecting effects of potential solutions and to index cases to
facilitate such reasoning.
4.3 Managing complexity and promoting successful problem solving
The PBL approach asks students to solve very hard problems. But the methodology also provides ways of
managing the complexity and helping students to be successful. Complexity is managed in a variety of ways.
(1) Students work together in collaborative groups, pooling their expertise, experience, ideas, and time.
Students can build on each others' strengths. Working in groups also promotes learning how to articulate and