Intelligent Support for Exploratory Environments: Where are We and Where Do We Want to Go?

Abstract

In the last years, there is a rising interest in exploratory learning environments, due to their positive effects on learning. However, their lack of structure makes the provision of intelligent support and feedback a very challenging problem. This workshop tries to shed some light on several of the aspects of this complicated issue. This preface makes a summary of all topics to be discussed, as well as the innovative methodology that will be used.

Full text

Intelligent Support for
Exploratory Environments:
Where are We and Where Do We Want to Go?
Sergio Gutierrez-Santos1and Manolis Mavrikis2
1Birkbeck College
2Institute of Education
1,2London Knowledge Lab
23-29 Emerald Street, London, WC1N 3QS, UK
{sergut,m.mavrikis}@lkl.ac.uk
Abstract. In the last years, there is a rising interest in exploratory
learning environments, due to their positive effects on learning. However,
their lack of structure makes the provision of intelligent support and
feedback a very challenging problem. This workshop tries to shed some
light on several of the aspects of this complicated issue. This preface
makes a summary of all topics to be discussed, as well as the innovative
methodology that will be used.
1 Introduction
Exploratory learning environments (ELE) support a constructionist approach for
learning, encouraging learners to create their own solutions to problems. This
has shown to be particularly beneficial in terms of providing opportunities for
acquiring deep conceptual and structural knowledge. However, the key learning
over many years is that the level of support of the learning process is crucial;
this support can be provided by teachers, peers, technologies and the structure
of the activity sequences. This is particularly true in the case of mathematics
where, unlike physics or other science domains, knowledge is rarely a directly
observable outcome of a simulation under exploration and therefore other more
expressive tools are required to permit students to externalise their ideas.
Despite important attempts in the last few years to improve the effectiveness
of exploratory learning environments (ELEs) employing AI techniques, there is
a lack of a place where issues that pertain to intelligent support for ELEs can be
discussed and shared among researchers. This has been one of the driving forces
behind starting the series of this International Workshop on Intelligent Support
for Exploratory Environments (ISEE).
ISEE’08 brings together researchers from different fields of expertise to ad-
dress the challenging problems posed by the application of intelligent support
to exploratory learning environments. The accepted seven full papers will form
the basis of a discussion during the workshop held in conjunction with the third
European Conference on Technology-Enhanced Learning (EC-TEL’08) in Maas-
tricht (The Netherlands) between the 17th and the 19th of September, 2008.

2 Sergio Gutierrez-Santos and Manolis Mavrikis
The innovative format of the workshop, using an adaptation of the Learning
Cafe methodology, will ensure productive inter-disciplinary discussions about
both technical and pedagogical issues.
2 Where are we now (topics covered)?
The research issues addressed from the submitted papers cut across several sub-
jects. This section provides a summary of the main topics that are relevant to
the workshop.
2.1 Design of exploratory learning environments
The designers of exploratory learning environments face a challenging task of
striking several balances. The environment must be rich enough to provide good
opportunities for learning while, at the same time, being relevant to the domain.
It must be general so that different tasks can be developed on top of it, but
its interface must not be too cumbersome or the learner will quickly disengage.
From the point of view of this workshop, the most important balance is be-
tween giving freedom of action to the learner, which recognises the importance
of students’ autonomy and responsibility over their learning; and providing a
structured environment, that eases the intelligent analysis of learners’ actions.
These considerations are apparent behind the design of the systems discussed
in the papers contributed to the workshop. A particular type of an ELE under
development, a mathematical microworld, is presented in [1]. The paper describes
its features, as well as the potential for intelligent support. It also highlights the
fact that most related work focuses on developing intelligence for exploratory
learning environments that are open yet well-defined, meaning that they usually
allow exploration of pre-determined models or simulations. It seems however
that there are quite a few lessons to be learned from the way ELE are designed
for simulations and the requirements for enabling intelligent support. Example
of such systems are presented in [2] and [3], which are targeting the domain of
physics and chemistry by providing laboratory-like exploratory activities.
2.2 Balance between freedom and guidance and other pedagogic
strategies
One of the most important pedagogical considerations in exploratory learning
environments is the need for the teacher to maintain a balance between (i) the
freedom they allow for learners when interacting with ELEs and (ii) the guidance
they provide in order to ensure students’ interaction is effective and meaningful.
This is discussed in detail in [4]. An important part of the paper is devoted
to present pedagogic strategies, outlining the role of the teacher as that of a
‘facilitator’ who maintains this balance. This seems an important requirement
of intelligent support, and a starting point for any exploratory learning situation.

Intelligent Support for Exploratory Environments: Where. . . ? 3
Another paper that deals explicitly with this issue is [5], although from a
different point of view. The focus here is in one particular ELE, and situations
that can happen when learners interact with it. In order to provide support for
their actions, there are times in which the need for support is unclear from the
actions of the student. The paper suggest different strategies for this problem,
including getting help from the human teacher.
Although the issue is not discussed explicitly in other papers, it is implicit be-
hind the design of the systems presented, all of which are driven by constructivist
principles. For example, [2] discusses briefly on the number of states in which one
task can be expected (in their system) to offer feedback to the learner. Another
paper [3] shows how a prompting strategy has evolved from one with a very clear
structure to a more loose one, as the authors’ experience with a past version of
their system showed that too much structure was overwhelming the students.
2.3 Teachers need support too
When ELE are integrated in a classroom, support for teachers’ work becomes
an important issue. The goal is then twofold: to support the students in order to
ease the burden of the teacher and to support the teacher in her specific tasks. As
discussed in [4], due to the need to attend to all students individually, teachers
find it difficult to accomplish their role as facilitators in a classroom. Although
this paper only suggests ways that an intelligent ‘computer-based facilitator’ can
serve the role of a teacher assistant, the strategies presented can form the basis
of the development of ISEE. This last aspect is specifically covered in [5].
Some brief suggestions about the potential of intelligent support in their ex-
ploratory learning environment are also provided in [1], while [2] goes a step
further and provides tools for teachers to (i) author exercises and (ii) enable
them to investigate patterns of students’ behaviour, offering visual information
about how students interact with the ELE. Additionally, some of the techniques
presented in [6], although designed to help the students reflect on their interac-
tion with the system, could be used to provide visual information to the teacher.
2.4 Authoring of open-ended tasks with intelligent support
The development of activities in exploratory environments, similar to any other
intelligent environment, is an expensive and complex process. Although no anal-
ysis has been made in the literature (to the extent of our knowledge) about the
cost of developing this kind of systems, they are more complex than the typical
ITS. The cost of creating an ITS (i.e. software, rules, content, etc) was estimated
by Murray [7] in about 100 hours of work per hour of instruction. The cost of
designing a meaningful and general set of activities in an intelligent exploratory
environment can be expected to be even higher.
One of the papers [2] discusses ways to simplify the authoring of activities
by separation of concerns: some stakeholders design virtual tools that can be
used in the environment (with an API), some others design questions relating to
the tools, etc. Questions are viewed as related to the status of the whole system,

4 Sergio Gutierrez-Santos and Manolis Mavrikis
according to some rules. Another paper [8] focuses on developing a framework for
the authoring of rules for non-technical users, by using a rule-based expert system
combined with a particular framework for the design of exploratory activities.
2.5 Research Methodology
Although the fields of AIEd and ITS are well researched, the particular issues
that pertain to ISEE were discussed mostly in the early 90s (e.g. [9]). They have
re-appeared the last few years in an attempt to develop support that extends the
effectiveness of exploratory learning environments. Therefore, work on research
methodologies that can facilitate the development of intelligent support in ELE
is required.
The research presented in [3] emphasises the need to develop intelligent sup-
port based on models of the cognitive processes applicable to the particular
learning scenarios. The authors present an experimental study which uses a
Wizard of Oz approach where participants interact through an interface with a
human ‘wizard’. This is commonly used to investigate human-computer interac-
tion in systems under development in order to inform the design of intelligent
support. A complementary approach is followed in [4] . This relies on eliciting
knowledge from observing realistic teacher-student interactions, as well as iden-
tifying strategies based on relevant theories of learning. Another approach is the
use of data-mining techniques a posteriori on data collected from the learning
environments, as hinted in [2] and [5], although neither paper focuses on this.
2.6 Intelligent inference and analysis
In order to understand the actions of the learner, artificial intelligence, machine
learning and data mining techniques can be used to analyse the data provided
by the system. This information can be used in order to make inferences and
take appropriate actions: e.g. present relevant information to the teacher, suggest
collaboration between peers, etc.
A possible approach is the use of expert systems based on rules [8]. This
paper focuses on the issues that appear when a rule-based system is integrated
with an already existing exploratory environment, providing some interesting
discussion about the best architectural approach and the implications of using
fuzzy logic. Rules are also used in [2] to determine when feedback is needed.
Another approach is the use of case-based reasoning for detecting similarities
in the actions of the learners [5]. The paper illustrates how such a technique
opens the door to providing hints that do not distinguish correct from incorrect
actions, but rather show what other students have done. This could be a way to
address the difficulty in ELEs where the number of possible courses of actions
is higher than in other constrained environments; it must be noted that in some
activities for ELEs the distinction between correct and incorrect solutions does
not always make sense.

Intelligent Support for Exploratory Environments: Where. . . ? 5
Most of the other papers provide suggestions for future work and pointers
to relevant literature. Discussions during the workshop will potentially lead to
interesting interactions between all the participants.
2.7 Modelling in exploratory environments
In order to understand the actions of the learner, a modelling strategy is needed.
Many intelligent systems use a strategy for modelling both the learner and the
domain. The user model usually tries to represent the knowledge level or the
learners, but they can take also care of other aspects (e.g. their preferences,
learning styles [10], etc).
In the last years, many researchers have focused on the use of open learner
models, that is, models that are shown to the student to help them reflect on
their own actions. This is the main topic of [6], and it depicts the different levels
of granularity in which the modelling of the learner takes place, the different
indicators that they show to the learners, etc. The effectiveness of the approach
is being evaluated.
Modelling the actions of the student in an exploratory environment is the
main objective of [5]. The modelling of the actions of the learners is used as an
indirect approach to model the users. Possibilities of feedback or collaboration
are evaluated according to similarities between the learner actions and some
paradigmatic actions stored in a knowledge base. The paper describes some of
the details of the modelling strategies used (based on case-based reasoning), as
well as the metrics that are used to compare different actions and strategies.
2.8 Collaboration scenarios and tools for collaboration
Collaborative learning has been shown to provide a deeper understanding of con-
cepts, and longer-lasting retention [11]. In ELE, if collaboration is introduced
properly, it can provide a means of managing the complex balance between free-
dom and guidance. From the point of view of this workshop, there were two
strands in which we were specially interested: collaboration scenarios and tools
for collaboration. The first strand is of a more pedagogical nature, and deals with
the different possibilities in which collaboration can take place in an exploratory
environment. These environments open themselves to many interesting possi-
bilities: co-construction with peers, challenges against other peers in the frame
of the system, comparison of equivalent or complementary solutions, etc. The
second strand is more technical, studying which tools can be used for collabo-
ration, and how they help overcome some of the problems that take place in a
traditional collaborative learning scenario (e.g. domination issues).
This second strand has shown to be interesting to the community, as seen
from the papers received. Selecting the best tools for collaboration and designing
the right affordances in them is important, because the learning process is influ-
enced by the tools available. Such tools are presented in [3] as extensions to a
simulation environment in order to promote collaboration between students. [1]
discusses the potential of students co-constructing of models in their microworld

6 Sergio Gutierrez-Santos and Manolis Mavrikis
and [4, 5] provide suggestions on how to support students’ collaboration. [6] al-
ludes to the use of open-learner modelling tools for encouraging and facilitating
collaboration and constructive competition between students by inspection of
their own learner model and comparison with that of others.
3 Future work
When we where preparing the workshop, there were several topics that appeared
on the table as interesting and related to the general theme. Given that the
workshop is quite specialised, we tried to connect it to several research strands
that are relevant to researchers related to learning technologies. In order to do
this, we designed a series of questions that could be answered from different
points of view. Those that were answered have been discussed before. In this
section we present some of those that were not covered. They will have to be
addressed in future editions of the workshop.
3.1 Affect in ELEs
The number of papers that deals with affective aspects of the learning process,
from the point of view of learning technologies, increases every year. From our
point of view, the main concern related to intelligent support in exploratory
environments is that of motivation and encouragement.
Learners in an ELE need a certain level of motivation to remain focused.
Without it, they are unlikely to feel they are learning anything and may lose in-
terest in using the system. Detecting motivation automatically is a hard problem
in itself. The open nature of exploratory learning activities makes the problem
even more difficult to address (e.g. the same behaviour can sometimes be in-
terpreted as meaningful exploration or as disengaged playing with the system).
[4] provides pointers to AIEd literature that demonstrate progress in detecting
certain student affective factors. It also suggests that, in the cases where the in-
formation is very uncertain, the system can provide this information to teachers
and let them intervene instead of interacting with the student directly. Given
the interest in these issues for the last few years, we expect that future research
in this field will be applicable to ISEE.
3.2 What counts as correct?
Traditional Intelligent Tutoring Systems (for a classical survey, see [7]) deal with
domains clearly defined, usually of a scientific or technological kind. In these
domains, users have to introduce answers into the system, and the answers can
be corrected directly because they are either right or wrong.
However, in exploratory environments there are no clear right and wrong
answers. Sometimes, there is not even an answer, because the learner is only
expected to explore the domain. However, this does not mean that the system
should not be able to provide support for this exploration. As a matter of fact, an

Intelligent Support for Exploratory Environments: Where. . . ? 7
exploratory environment with no guidance has been reported as being harmful
for learning [12].
Knowing when to provide support is far more complicated than just matching
answers with a table or a set of calculations: it requires that the system has some
knowledge of the task at hand, some understanding of the actions of the learner
and some ideas about the support that can be provided. All three of these are
open and challenging research questions by themselves.
3.3 Activity model
An intelligent system that aims at supporting the learning process usually follows
some strategy for modelling the user (i.e. learner) and the domain in which the
learning takes place. The learner model usually tries to capture the knowledge of
the student, but could also express other characteristics (e.g. emotional states,
preferences, etc). The domain model describes the domain, and is used to assess
the knowledge of the student.
Modelling the learner in an exploratory environment poses a difficult chal-
lenge, as there are no clearly correct or wrong answers, and there are no clearly
correct or wrong behaviours. Exploratory environments are usually used on do-
mains that are not clearly defined, and are difficult to model (a similar problem
is discussed in [13]). Furthermore, exploratory environments are sometimes used
for different tasks; the actions of the learner are very different depending on the
task, making the modelling more difficult.
Therefore, it may be necessary to have an activity model separated from the
domain model. Activity or task models have been used in the past in the field
of Intelligent User Interfaces [15] and Adaptive Hypermedia Systems [14].
3.4 Learning standards
The number of learning support systems is huge. Usually each of them uses their
learning strategy, their own content, etc. This makes it impossible to collaborate
between different systems, share resources, etc.
In order to solve this, several initiatives exist to create standards that foster
intercommunication between different systems. Arguably the most important are
the SCORM initiative3, the IMS Consortium4and the SISO standards5. How
these initiatives are relevant to exploratory environments and intelligent support
remains an unanswered question.
4 Conclusions
The provision of intelligent support in exploratory environments poses a lot of
interesting challenges. Some of them are going to be covered in the workshop. We
3http://www.adlnet.gov/scorm/
4http://www.imsglobal.org
5http://www.sisostds.org

8 Sergio Gutierrez-Santos and Manolis Mavrikis
hope that the discussions that will take place will produce interesting synergies
and help to find some answers, leading to a productive session for all participants.
We anticipate that the outcomes of the discussion will provide new insights
into these challenges, and serve as a catalyst for other lines of research. The
main findings of the workshop will be published in the future for the benefit of
the community.
5 Workshop Methodology
One of the goals of the workshop was to be used as a platform to share ideas
and developments, proving worthwhile to all participants. Therefore, we tried to
detach ourselves from the mini-conference style in which many workshops are
run. In this fashion, most of the time is consumed in presentations of authors’
work, which leaves a little time window for discussion among the workshop at-
tendants. We thought that the time spent in discussions should make the best
part of the workshop.
That is why we tended towards an adaptation of the ‘learning discussion
forum’ (aka Learning Cafe Methodology) which focuses on discussions ensur-
ing that all participants can have a direct impact in addressing the workshop
questions. This methodology (with some variations) had been successfully im-
plemented in at least two previous conferences [16, 17].
The working methodology for this workshop is made up of the following steps:
1. A short discussion was included in the call for papers that raised some open
problems in the field and posed challenging questions that the participants
of the workshop had to answer in advance. Those interested in the workshop
submitted their papers, covering some of the challenging questions. Each
paper was reviewed by 3 members of the Program Committee to provide
feedback from different angles.
2. After the final papers have been collected into the Proceedings, a collabo-
rative environment is set up to facilitate open discussion among workshop
participants previous to the workshop day. Open discussion among partici-
pants can take place in advance of the workshop day.
3. On the workshop day, participants bring a sheet with their relevant conclu-
sions for each topic they have worked in their paper.
4. Two round tables are settled, each managed by a Moderator (more details
bellow). The tables have A-1 sheets and markers. A brief overview of each of
the topics is done, raising the challenging questions. Each table brainstorms
led by the Moderator. The ideas are written down in the sheet. When the
sheet is full, they are stuck on the wall. The brainstorm and the discussion
continues until the break.
5. After the break, people switch to the other table. The Moderators stay in
their table and summarise to the new people what was discussed with the
previous group. The participants continue the brainstorm and discussion
until lunch, taking special care of discussing topics that were left behind by
the former group.

Intelligent Support for Exploratory Environments: Where. . . ? 9
6. After lunch, for each table, Moderators present the conclusions to the audi-
ence. Open discussions with all participants are expected to be risen. The
Moderators summarise the conclusions with the collaborative help of the
group. Final conclusions are structured and uploaded to the website to be
shared with the EC-TEL community and participants that have not been
present in the workshop.
The two tables cover different topics. Discussion on the first table will be
related to the design of exploratory environments and activities: microworlds,
methodology, authoring, the role of the teacher, reflection on students, etc. The
second table will be devoted to the technical aspects of intelligent support: user
modelling, data mining, visualization techniques, rule-based and case-based rea-
soning, etc.
Acknowledgements
Many people have worked hard in order to make this workshop a reality. We
would like to sincerely thank them for their help and (intelligent) support.
First of all, many thanks to all the other members of the Organising Commit-
tee: Richard Noss and Celia Hoyles (Institute of Education, UK), and Alexandra
Poulovassilis and George D. Magoulas (Birkbeck College, UK). Then, we would
like to thank all the members of our international Program Committee: Ryan
Baker (Carnegie Mellon, USA), Jesus G. Boticario and Olga C. Santos (Uni-
versidad Nacional de Educacion a Distancia, Spain), Paul Brna (University of
Glasgow, UK), Andrea Bunt (University of Waterloo, Canada), Cedric d’Ham
(University of Grenoble, France), Vania Dimitrova (University of Leeds, UK),
Ken Kahn (University of Oxford, UK), Piet Kommers (University of Twente,
The Netherlands), Chronis Kynigos (University of Athens, Greece), Muriel Ney
and Sophie Soury-Lavergne (CNRS, France), Abelardo Pardo (University Car-
los III of Madrid, Spain), Cristobal Romero (University of Cordoba, Spain) and
Niall Winters (Institute of Education, UK).
Additionally, we would like to acknowledge the financial support of TLRP
(e-Learning Phase-II, RES-139-25-0381), and the important (and very intelli-
gent) support we have got from other members of the MiGen team during the
preparation of this workshop, specially Eirini Geraniou and Darren Pearce.
References
1. Darren Pearce, Eirini Geraniou, Manolis Mavrikis, Sergio Gutierrez-Santos and Ken
Kahn. Using Pattern Construction and Analysis in an Exploratory Learning Envi-
ronment for Understanding Mathematical Generalisation: The Potential for Intelli-
gent Support. In: [18]. (2008)
2. Dror Ben-Naim, Nadine Marcus, Mike Bain. Visualization and Analysis of Student
Interactions in an Adaptive Exploratory Learning Environment. In: [18]. (2008)

10 Sergio Gutierrez-Santos and Manolis Mavrikis
3. Bruce M. McLaren, Nikol Rummel, Niels Pinkwart, Dimitra Tsovaltzi, Andreas
Harrer and Oliver Scheuer. Learning Chemistry through Collaboration: A Wizard-
of-Oz Study of Adaptive Collaboration Support. In: [18]. (2008).
4. Manolis Mavrikis, Eirini Geraniou, Richard Noss and Celia Hoyles. Revisiting ped-
agogic strategies for supporting students’ learning in Mathematical Microworlds.
In: [18]. (2008).
5. Mihaela Cocea, Sergio Gutierrez-Santos and George D. Magoulas. Challenges for
Intelligent Support in Exploratory Learning: the case of ShapeBuilder. In: [18].
(2008).
6. Kyparisia A. Papanikolaou and Maria Grigoriadou. Sharing knowledge and promot-
ing reflection through the learner model. In: [18]. (2008).
7. Tom Murray. Authoring Intelligent Tutoring Systems: An analysis fo the state of
the art, in Int. Journal of Artificial Intelligence in Education, 10, pp.98-129 (1999)
8. Charles Hunn. Employing a Java Expert System Shell for Intelligent Support in
Exploratory Activities. In: [18]. (2008).
9. Mark Elsom-Cook. Guided Discovery Tutoring: A Framework for ICAI Research.
Paul Chapman Publishing (1990).
10. Enrique Alfonseca, Rosa M. Carro, Estefania Martin, Alvaro Ortigosa, Pedro Pare-
des. The impact of learning styles on student grouping for collaborative learning: a
case study, in User Modeling and User-Adapted Interaction, 16 (3-4) (2006)
11. Anastasio Ovejero Bernal, Maria de la Villa Moral Jimenez and Juan Pastor Mar-
tin. Aprendizaje cooperativo: un eficaz instrumento de trabajo en las escuelas mul-
ticulturales y multietnicas del siglo XXI, in Revista Electronica Iberoamericana de
Psicologia Social, 1 (2) (2002).
12. P. Kirschner, J. Sweller and R. Clark. Why minimal guidance during instruction
does not work: An analysis of the failure of constructivist, discovery, problem-based
experiential and inquiry-based teaching, in Educational Psychologist, 41(2), 75-86
(2006)
13. Vincent Aleven, Kevin Ashley, Collin Lynch and Niels Pinkwart. Workshop on
Intelligent Tutoring Systems for Ill-Defined Domains, in Int. Conf. on Intelligent
Tutoring Systems (ITS’08) (2008)
14. Garlatti, S., Iksal, S., and Kervella, P.: Adaptive on-line information system by
means of a task model and spatial views. Computer Science Report, Eindhoven
University of Technology, Eindhoven, pp. 59-66 (1999)
15. Ulrich Hoppe. Intelligent user support based on task models, in Schneider-
Hufschmidt, M., Khme, T. and Malinowski, U. (eds.): Adaptive user interfaces:
Principles and practice, pp. 167-181 (1993)
16. Jesus G. Boticario and Olga C. Santos (eds.). Workshop ‘To-
wards User Modelling and Adaptive Systems for All’, in Int.
Conf. on User Modelling (UM’07). Proceedings available online at
http://adenu.ia.uned.es/workshops/um07/tumasa07/proceedings.htm, last ac-
cessed Aug 2008 (2007)
17. Vana Kamtsiou, Tapio Koskinen and Paul Lefrere (eds.). Workshop Making of the
Future of Technology Enhanced Professional Learning, in Eur. Conf. on Technology
Enhanced Learning (EC-TEL’06) (2006)
18. Proceedings of the 1st Int. Workshop in Intelligent Support for Exploratory Envi-
ronments on European Conference on Technology Enhanced Learning (EC-TEL’08).
(2008).