The Design of a Mobile Intelligent Tutoring System

Quincy Brown1,**, Frank J. Lee1, Dario D. Salvucci1, Vincent Aleven2

Drexel University1, Carnegie Mellon University2

Abstract. Mobile intelligent tutoring systems have the potential to deliver low-cost, one-to-one
assistance to learners outside of the traditional classroom and computer lab settings. The focus of this
paper is to outline the process involved in researching the role that mobile devices can play in
disseminating and supporting the knowledge gained by intelligent tutors. We review three aspects of
mobile intelligent tutors — namely the interface, learning and teaching strategies, and the software and
hardware architecture — and discuss concepts and preliminary results for consideration of all three
aspects in building the next generation of mobile intelligent tutors.

1. Introduction

The ubiquitous use of cell phones in society has led researchers to investigate methods to employ mobile
devices in education [1-4]. To date, mobile devices have been successfully used in a variety of ways and in
a number of contexts: to create interactive user experiences in museum settings [5, 6], to facilitate scientific
field exploration [7, 8], and of course to extend, both formally and informally, science, math, and language
learning K-12, higher education, and adult education [9-11].

At the same time, some of the most complex and comprehensive computer-based instructional methods
still reside primarily on the desktop. In particular, intelligent tutoring systems (ITSs) have been shown to
be highly successful in improving student learning in the classroom [12]. For example, ITSs currently assist
students in Mathematics, Science, and Language Learning domains for learners in high school and higher
education courses [13, 14]. When integrated into school curricula, students use the tutors during school
hours in computer labs and classrooms. Tutors have also been used to aid students with homework, test
taking, and assessment [15]. Nevertheless, the extension of these desktop ITSs to the world of mobile
learning could provide great benefit for students and teachers alike. A mobile tutor has the potential to
deliver the significant advantages of intelligent tutoring systems to a wide audience of learners and extend
tutor use to outside of computer labs and traditional classrooms thereby providing robust learning
opportunities to students ”anywhere anyplace” [16].

In this paper we present the design of an intelligent tutoring system implemented on a mobile device.
This research explores three areas within the mobile intelligent tutor design, Table 1. First, we will
investigate how to create a user interface for the much smaller screen offered by mobile devices. Second,
we will explore practical prescriptions for mobile tutor usage. Third, we will detail the hardware and
software architecture required to support the integration of mobile tutors into an existing tutor delivery
architecture. The results we present here are preliminary and will serve as the foundation for future
research.

Desktop Tutor Mobile Tutor

Interface 17” and larger displays 2-5” displays
Multi-window design Single-window/panel design
Usage Minutes-to-hours of use Seconds-to-minutes of use
Classroom & computer Labs Anywhere & anytime
Architecture Large memory requirements Small memory requirements
Land or wireless Internet Wireless or desktop synchronization

Table 1 Contrasting mobile and desktop tutor interface, usage, and architecture.
 For purposes of this research, we define mobile devices as cell phones, personal digital assistants
(PDAs) or smart phones; we do not include laptop or micro-sized computers in our definition. The benefit
of using cell phones is that, while students may not have access to computers or the Internet outside of
school, they may be more likely to have access to a cell phone to use the tutors. PDAs are also beneficial
in that they do not require additional costs for voice and data services; therefore, once the initial investment
in the device is made, a standalone or online application can be accessed without users, or schools,
incurring additional recurring costs.

1.1 Background

1.1.1 Mobile Learning

Mobile learning is an emerging discipline in the area of education and educational technology. Just as
personal computer integration predicated computer-enhanced learning, a primary factor in the growth of
mobile learning is the increasingly ubiquitous integration of cell phones into society. In the second quarter
of 2005 there were more than 190.5 million mobile devices and smart phones sold worldwide [17]. It is
expected that in 2015 more than five billion people will utilize services via mobile devices [18]. In
addition to cellular phones recent years have seen an increase in non-phone mobile devices such as MP3
players, such as iPods™, and Palm Pilots. While these devices do not perform voice services they do have
many of the same services as today’s smart-phones such as data and basic word processing. This increase
in wireless connectivity has lead educators and researchers to investigate methods to integrate mobile
devices in education.

Mobile learning has been used to support communication between instructors, students, and peers [9, 19,
20]. It has been used to deliver general as well individualized course content material to K-12, higher
education, and adult learners. It has been employed in the science, museum, math, language acquisition,
and job skills training domains. For instance, in 2005, 19% of schools provided handheld computers to
students while 10% provided laptop computers for home use [21]. Although these numbers do not
represent the majority of schools, they represent a 9% increase in handheld use from 2003 [21]. These
numbers are indicative of a positive trend towards school making use of handheld computing and allowing
students to coordinate technology use between home and school.

For a mobile device to incorporate a full-blown intelligent tutoring system, field researchers must
understand how learning may differ when mobile devices enter the environment. As school systems begin
to integrate handheld devices into their existing technologies, educators must understand how to use the
technology either collaboratively with or instead of desktop computers. Conducting research on mobile
intelligent tutors has the ability to provide insight and begin to answer these currently open issues.

1.1.2 Mobile Intelligent Tutoring Systems

The delivery of tutoring systems on mobile devices has the potential to deliver the significant advantages
of intelligent tutoring systems to a wider audience of learners. Despite the fact that nearly all schools
provide Internet and computer access to students, a deeper examination reveals that the presence of
technology does not equate to effective use of the technology. Perhaps surprisingly, in a recent survey,
only one-third of survey teachers indicate feeling prepared to use computers and the Internet for instruction
[21]. Another factor hindering use is the student-to-computer ratio in schools: in 2005, no school reported
having one computer for each child with the lowest computer to student ratio being approximately 3-to-1.
Unfortunately, schools with greater numbers of minority students enrolled reported a higher number of
students per computer [21]. The low cost of mobile and handheld devices has the potential to deliver a
one-to-one computing solution to the education community [22].
 School systems without the financial resources to invest in and maintain large computer labs will, by
using mobile devices, have the ability to provide learners with proven educational technology. As schools
consider investing in mobile technologies for student use, they will consider lower cost and light weight of
mobile devices. These affordances make mobile or handheld device sometimes preferable to laptop or
desktop computers. Students can more easily transport the tutors between home and school as well as share
the mobile tutors between classes in the same school. The portability affordance of mobile tutors can
extend tutor use to outside of computer labs and traditional classrooms thereby providing robust learning
opportunities to students at home, after school, and in other locations. Considering that less than 40% of
families earning less than $24,999 report have Internet access at home [6], it is plausible to project that in
the near future mobile devices will have the capability to serve as a platform uniting learning between
home and school as well as providing a means for students to employ educational software while outside of
the traditional school environment.

1.2 Hypothesis and Research Questions

It is our hypothesis that mobile intelligent tutors can augment desktop tutors, working hand-in-hand and
enabling more time-on-task, than is currently possible because of classroom and lab limitations. Our
prevailing theory is that mobile tutors will prove to provide learning gains comparable to desktop tutors
provided that the overall time on task remains the same. To understand the learning a mobile intelligent
tutor can provide we seek to answer the following research questions:

• Does shorter, yet more frequent, mobile tutor use provide equivalent learning gains as
compared to longer, yet less frequent, desktop tutor use?
• What types of problems are best suited for short and frequent tutor use?
• Are the learning gains of the mobile tutor equal to those of the desktop tutor?

1.3 Approach to Addressing the Research Questions

To address the research questions we will conduct controlled experiments utilizing a desktop tutor and
two mobile tutors. For the purposes of these experiments we will employ existing tutors from the
Pittsburgh Science of Learning Center (PSLC) as the basis for the desktop and mobile tutors. The desktop
tutor will serve as the control tutor and will be the baseline to which we will compare the mobile tutors.
The first mobile tutor will be a close replica of the desktop tutor. It will be utilized in experiments designed
to understand if a mobile tutor can effectively replace a desktop tutor and determine if the learning gains of
the two tutors are comparable. A second mobile tutor will allow us to experiment with the frequency,
duration, and type of problems presented to learners. We will explore the effects of continuous and
interrupted tutor use and examine how the nature of the tutor questions have to change to fit interrupted
mobile tutor use. Experiments comparing the learning gains of both mobile tutors will be carefully
developed to minimize the Novelty Effect [10, 23] due to the introduction of new technology into the
education process.

Learners will participate in pre- and post- testing allowing us to analyze test scores for learning gains.
We will work closely with classroom teachers to ensure the problems, expert model, student model, hints,
and feedback are consistent with the existing pedagogy, learning and teaching strategies. We will collect
solution time data to gain an understanding of time on task with tutor use. Statistical analysis will be
performed to discover the relationship between variables such as time on task, achievement gains, and
problem structure; and to test the hypothesis that students using the mobile tutor can have learning gains
equivalent to students using desktop tutors.

In the remainder of the paper we will describe how the proposed experiments will provide insight in
answering the research questions. We will describe a completed preliminary interface study provide details
regarding the effect of teaching and learning strategies as well as detail anticipated architectural issues.
2. A Mobile Intelligent Tutor

2.1 The Interface

Intelligent tutors, such as Andes, Algebra I, and AutoTutor, provide users with interfaces that are
divided into regions. Andes, an intelligent tutor for Newtonian Physics, has an interface divided into three
regions/panes [11]. Similarly, Algebra I and AutoTutor developers employed a multi-region interface
design [24, 25]. Users are given the specific problem or scenario they are currently working on, a
workspace, as well as hint buttons, and other areas such as a glossary, or reference material necessary. As
students use the tutors on large desktop displays, they are easily able to see all of these regions easily
without having to manually navigate to each individually. The problem with implementing the tutor on
the smaller device is that the displays are much smaller.

Contrasted with a 17 inch or larger display of the desktop computer, the display of a handheld device
can range between 1.5 to 4 square inches. Although newer smart phones are available offering larger screen
sizes and greater resolutions the complexity of the tutor interfaces would still result in users performing
more scrolling or similarly undesirable navigation to maneuver within the interface. This reduction in real
estate requires the interfaces to be redesigned.

In the context of usability of a mobile tutor interface, we desire to provide mobile users with access to
the information they need without having to perform a large sequence of keystrokes or experience font
sizes too small comfortably read. The simpler interface will ensure that when a user is interacting with the
mobile tutor that their time is spent on problem solving and not on navigation. This will allow the time on
task to be used actually solving problems and not searching for information within the interface. We
recognize that in addition to small displays there are other interface challenges, such as limited user input
mechanisms. However, we are focusing the mobile interface issue to to display for the purposes of this
research.

2.1.1 Preliminary Interface Study Results

Tracking users’ eye movements is a technique used in human-computer interaction to explore the
usability of an interface. In e-learning environments the use of an eye tracker can provide insight into how
learners interact with systems and record their visual attention flow. The trackers employ complex
hardware and software using infrared technology to track the change in users attention. Their benefits
notwithstanding, eye trackers are complex and require many hours of preparation to setup. They also suffer
from limitations in calibrating the software and hardware with multiple users and can be affected by users’
glasses or contacts and even facial gestures [26]. A less complicated alternative “Poor Man’s Eye Trackers”
(PMETs) have been used to evaluate user interfaces without the great expense and time for using an eye
tracker [27]. PMETs cover areas of information with opaque blocks and require the user to uncover
relevant regions, thus enabling data collection on a user’s shifts of attention. This method does provide less
detailed information than a traditional eye tracker; however, it is able to provide data regarding users areas’
interest as they progress through a tutor [28].
Figure 1 Geometry Tutor based on Carnegie Learning framework

Figure 2 Poor Man’s Eye Tracker version of Geometry Tutor

For this study a geometry tutor, Figure 1, based on the Carnegie Learning framework [29] was converted
into a Poor Man’s Eye Tracker (PMET) tutor, as shown in Figure 2. The PMET interface was designed to
provide insight into how users interact with an intelligent tutor that has a complex, multi-region, interface.
The tutor interface was divided into diagram, problem, hint, and glossary regions. Each region was covered
using an opaque mask that is removed when the mouse is clicked on the region label. Using the PMET
tutor, users were forced to view one region at a time while solving four geometry problems.

The study was conducted with eight graduate and undergraduate university students using the tutor to
solve four geometry problems. Study participants had previously completed courses in geometry and were
familiar with basic terminology and concepts comparable to the familiarity expected from mid to high
achieving high school geometry students. Prior to solving tutor problems participants reviewed geometry
terminology and concepts to refresh their memory. Although target students for the original tutor are in
high school we believe the results from the university students can be generalized to reflect behaviors of
high school tutor users because the study was of the interface use and not geometry problem solving
knowledge.

The tutor was designed to record the region and time for each of the users’ mouse clicks. For each user
the software recorded the complete listing of regions selected and the amount of time spent in each region.
To analyze the data we examined the transitions flows between regions and time spent in each region.
Viewing the transition patterns (interaction flows) of users provided insight into the relationship between
the various interface regions. We were able to see the strength of certain transitions, flows as compared to
others. As expected the most frequent transitions occurred between the problem and diagram regions, and
these accounted for approximately 70% of the total transitions. The unexpected result was that transitions
between the hints, glossary, and diagram regions comprised less than 10% of all transitions. Although
these transitions were not as frequent, the timing of them are important in the problem solving process. We
realized that when users needed hints and the glossary, those transitions were mostly to or from either the
problem or diagram. Considering that the study participants were similar to higher achieving students we
would expect lower achieving students to utilize the hints and glossary to an even greater extent. Initially
we intended to place the hints and glossary on a separate tab; however these results caused us to re-examine
our approach. Based upon this preliminary result we believe that placing them on tabs would cause users to
navigate too much at times when they most need clarity. This understanding will inform our decision for
designing the mobile intelligent tutor interface.

2.2 The Effect of Teaching and Learning Strategies

To understand the role that a mobile tutor can play in learning we will examine how mobile tutors can be
integrated into classes and the education. The tutors we are using as a platform for our research, Carnegie
Learning [29], are currently used in conjunction with a curriculum designed to accompany tutors. The
students use the tutors for 40 minutes several times a week.

The structure of the tutor questions have been shown to affect the rate of learning for users [15]. For
example, researchers have experimented with scaffolding within questions, use of feedback, spacing of
worked examples, and the level of mathematical abstraction within problems to improve student learning
[15, 30, 31]. To design the mobile tutor problems we intend to explore problem structures with respect to
varied levels of scaffolding, number of optimal solution steps, and the relationship between individual
questions. During home use, and use in other non-school locations, students may not have the same amount
of time to continuously utilize the tutor. In this case, the teaching strategy may change to support student
learning in various environments.

We anticipate that the primary difference between mobile tutor and desktop tutor use is with respect to
the frequency and duration of a learner’s use. The question we pose is whether it is feasible, considering
the small screen size and limited interaction modes, to expect students to use the mobile tutors with the
same frequency and duration as the desktop counterparts. However, considering that the targeted students
are from the 5th and 6th grades, they are comfortable playing video games on small handheld consoles and
we expect that they would be equally comfortable using a mobile tutor.
2.3 The Architecture

To support a mobile tutor, the chosen platform architecture, PSLC Carnegie Learning Tutor, will be
evaluated to determine the necessary changes required to support mobile tutor delivery. An optimal mobile
tutor delivery system will require minimal data storage of either application or user logs on the device, and
therefore we will examine how memory size differences between mobile devices and desktop computers
may affect this architecture. We will seek to maintain a minimal amount of data on the device.

Proposed solutions include periodic data synchronization, reduction in data logged, and minimization of
application file size. We will evaluate the effect of the small device on the tutor delivery mechanisms by
using measures such as the changes to the client-server protocol and data logging parameters required
during development. For web delivered tutors, we will measure connection times as well as time required
to communicate information to and from the server in an effort to provide a system that is responsive to
user interaction without long delays. Standalone tutors will be monitored to optimize the frequency of
synchronization required to avoid running out of storage space in the devices memory.

Although the previously proposed experiments are designed to answer the research questions, the issues
surrounding the architecture and delivery of a mobile tutor do play a role in mobile tutor efficacy and
feasibility. The three areas of development, interface, teaching and learning strategy, and architecture are
not mutually exclusive and therefore we recognize that decisions made in one area can have an effect in
another and thereby effect the entire system. For example, if the mobile tutors require students or teachers
to synchronize too often with a desktop computer, they may create an additional task in school days already
full of activities. If the devices use web delivery systems the additional cost of wireless data services must
be factored into the total cost of ownership and be considered as a possible limitation when wireless
services are unavailable. While we recognize that these issues are not directly related to our research
questions and goals, we acknowledge that these topics do have an effect on the feasibility of the long-term
adoption of mobile learning applications.

Throughout the development process we will explore the relationship between mobile ITS, teacher, and
student to identify features of a mobile ITS, such as text messaging for just-in-time teacher help, that are
unique to mobile devices.

3. Conclusion

It is our expectation that, with respect to mobile intelligent tutors, this research will bridge the gap
between the theoretical side of the education and learning science disciplines and the practical side of what
is feasible to implement in actual classrooms. This paper outlines a novel approach to the development of a
mobile intelligent tutor. We propose research in three areas of development, interface, teaching and
learning strategies, and architecture and describe the results from a preliminary user interface study.

We expect to contribute knowledge that will assist in the creation of mobile intelligent tutoring systems
that support learning in various locations. As researchers continue to discover how students can best learn
using desktop computers, this research will begin to provide quantitative and qualitative data leading to the
understanding of the role mobile devices can play in education.

As this project involves collaboration with the Pittsburgh Science of Learning Center (PSLC), we expect
to provide a mobile tutor platform that can be used by researchers utilizing their LearnLab facilities. An
artifact of this research will be an authoring tool that supports the rapid development of mobile intelligent
tutors as well as the architecture enabling the PSLC to provide data logging and delivery services to its
customers.
4. Acknowledgements

**The author is supported by National Science Foundation grant #DGE-0538476, and the third author is
supported by National Science Foundation grant #IIS-0426674.

5. References

[1] A. P. Massey, V. Ramesh, and V. Khatri, "Design, Development, and Assessment of Mobile
Applications: The Case for Problem-Based Learning," IEEE Transactions of Education, vol. 49,
2006.
[2] D. P. Nguyen, M. Guggisberg, and H. Burkhart, "CoMobile: Collaborative Learning with Mobile
Devices," in International Conference on Advanced Learning Technologies, 2006.
[3] M. Sharples, "The Design of Personal Mobile Technologies for Lifelong Learning," Computers &
Education, vol. 34, pp. 177-193, 2000.
[4] R. Y.-L. Ting, "Mobile Learning: Current Trend and Future Challenges," in IEEE International
Conference on Advanced Learning Technologies, 2005.
[5] P. Lonsdale, W. Byrne, R. Beale, M. Sharples, and C. Baber, "Spatial and context awareness for
mobile learning in a museum," in CSCL Workshop on Spatial Awareness and Collaboration,
2004.
[6] P. Pierroux, "Mobility In Learning: Meaning Making Across Classroom and Museum Settings," in
International Association For Development Of The Information Society Mobile Learning, Lisbon,
Portugal, 2007.
[7] U. Farooq, W. Schafer, M. B. Rosson, and J. M. Carroll, "M-Education: Bridging the Gap of
Mobile and Desktop Computing," in International Workshop on Wireless and Mobile
Technologies in Education, 2002.
[8] P. Silander, E. Sutinen, and J. Tarhio, "Mobile Collaborative Concept Mapping- Combining
Classroom Activity with Simultaneous Field Explorations," in International Workshop on
Wireless and Mobile Technologies in Education, 2004.
[9] J. Attewell and M. Gustafsson, "Mobile Communication Technologies For Young Adult Learning
and Skills Development (m-Learning)," in International Workshop on Wireless and Mobile
Technologies in Education, 2002, pp. 158-160.
[10] K. Facer, F. Faux, and A. McFarlane, "Challenges and Opportunities: Making Mobile Learning a
Reality in Schools," in Mlearn 2005: 4th World Conference on mLearning, 2005.
[11] A. S. Gertner and K. VanLehn, "Andes: A Coached Problem Solving Environment For Physics,"
in 5th International Conference on Intelligent Tutoring Systems, 2000, pp. 133-142.
[12] H. Cen, K. Koedinger, and B. Junker, "Is Over Practice Nessary? - Improving Learning Efficiency
with the Cognitive Tutor through Educational Data Mining," in International Conference on
Artificial Intelligent in Education (AIED 2007), 2007.
[13] "Andes Physics Tutor," 2008. Retrieved March 10, 2008 from http://www.andes.pitt.edu/.
[14] K. R. Koedinger, J. R. Anderson, W. H. Hadley, and M. A. Mark, "Intelligent Tutoring Goes To
School in the Big City," International Journal of Artificial Intelligent in Education vol. 8, pp. 30-
43, 1997.
[15] L. Razzaq, M. Feng, G. Nuzzo-Jones, N. T. Heffernan, K. R. Koedinger, B. Junker, S. Ritter, A.
Knight, E. Mercado, T. E. Turner, R. Upalekar, J. A. Walonoski, M. A. Macasek, C. Aniszczyk, S.
Choksey, T. L. Ak, and K. Rasmussen, "The Assistment Project: Blending Assessment and
Assisting," in 12th Artificial Intelligence in Education, Amsterdam, 2005, pp. 555-562.
[16] L. F. Motivalla, "Mobile Learning: A framework and evaluation," Computers & Education, vol.
49, pp. 581-596, 2007.
[17] T. Raiciu, "190.5 Million Cell Phones Sold - In the second quarter of 2005," 2005. Retrieved
December, 2008 from http://news.softpedia.com/news/190-5-Million-Cell-Phones-Sold-
6879.shtml
[18] G. Association, "Universal Access How Mobile Can Bring Communication to All -Full Report,"
GSM Association.
[19] J. S. Cabrera, H. M. Frutos, A. G. Stocia, N. Avouris, Y. Dimitriadis, G. Fiotakis, and K. D.
Liveri, "Mystery in the Museum: Collaborative Learning Activities using Handheld Devices," in
Mobile HCI, 2005.
[20] J. Traxler and J. Leach, "Innovative and Sustainable Mobile Learning in Africa," in Fourth IEEE
International Workshop on Wireless, Mobile and Ubiquitous Technology in Education, 2006.
WMUTE '06, Athens, 2006, pp. 98-102.
[21] J. Wells and L. Lewis, "Internet Access in U.S. Public Schools and Classrooms: 1994-2005,"
National Center for Education Statistics, Washington, D. C. 2006.
[22] C. Norris and E. Soloway, "Envisioning The Handheld-Centri Classroom," Journal of Educational
Computing Research, vol. 30, pp. 281-294, 2004.
[23] C. Noessel, "Mobile Learning As A Service Offering With Near-Term Technologies," in MLearn
2003, 2003, pp. 117-126.
[24] A. C. Graesser, G. T. Jackson, and B. McDaniel, "AutoTutor holds conversations with learners
that are responsive to their cognitive and emotional states," Educational Technology, vol. 47, pp.
19-22, 2007.
[25] K. R. Koedinger and V. Aleven, "Exploring the Assistance Dilemma in Experiments with
Cognitive Tutors," Educational Psychological Review, vol. 19, pp. 239-264, 2007.
[26] W. Schroeder, "What is Eye-Tracking Good For," 1998. Retrieved March 20, 2008 from
http://www.uie.com/articles/eye_tracking_benefits/
[27] C. Ullrich, D. Wallach, and E. Melis, "What is Poor Man's Eye Tracking Good For?," in 17th
Annual Human Computer Interaction Conference 2003, 2003.
[28] C. Ullrich and E. Melis, "The Poor Man's Eyetracker Tool of ActiveMath," in World Conference
on E-Learning in Corporate Government Healthcare and Higher Education (eLearn-2002), 2002,
pp. 2313-2316.
[29] C. Learning, "Carnegie Learning, Inc. Math curricula with Proven Success," 2008. Retrieved
March 30, 2008 from http://www.carnegielearning.com/
[30] E. Melis and G. Goguadze, "Towards Adaptive Generation of Faded Examples," in International
Conference on Intelligent Tutoring Systems, ITS 2004, Maceio, Brazil, 2004, pp. 761-71.
[31] J. Tan and G. Biswas, "The Role of Feedback in Preparation for Future Learning: A Case Study in
Learning by Teaching Environments," in 8th International Conference Intelligent Tutoring
Systems, ITS 2006, Jhongli, Taiwan, 2006, pp. 370-381.